9512.net
甜梦文库
当前位置:首页 >> >>

Finance


International Federation of Automatic Control

Symposium on System Identi?cation

SYSID 2000
21st – 23rd June, 2000 Santa Barbara, California, U.S.A.

SYSID 2000 is sponsored by: International Federation of Automatic Control (IFAC), Technical Committee on Modeling, Identi?cation and Signal Processing; American Automatic Control Council (AACC); Department of Electrical & Computer Engineering, University of California, Santa Barbara.

2

National Organizing Committee
Co-chair Dale Seborg Dept. of Chemical Engineering, University of California, Santa Barbara, CA 93106, U.S.A. +1 (805) 893-3352, +1 (805) 893-4731 (fax) seborg@engineering.ucsb.edu Finance Duncan Mellichamp Dept. of Chemical Engineering, University of California, Santa Barbara, CA 93106, U.S.A. +1 (805) 893-2821, +1 (805) 893-4731 (fax) dmell@engineering.ucsb.edu Registration Dan Rivera Dept. of Chemical, Bio. & Materials Engineering, Mail stop 876006 Arizona State University Tempe, AZ 85287-6006, U.S.A. +1 (480) 965-9476, +1 (480) 965-2910 (fax) daniel.rivera@asu.edu Co-chair Roy Smith Dept. of Electrical & Computer Engineering, University of California, Santa Barbara, CA 93106, U.S.A. +1 (805) 893-2967, +1 (805) 893-3262 (fax) roy@ece.ucsb.edu Facilities Kameshwar Poolla Dept. of Mechanical Engineering, University of California, Berkeley, CA 94720, U.S.A. +1 (510) 642-4642, +1 (510) 643-5599 (fax) poolla@jagger.me.berkeley.edu Publications Mario Sznaier Dept. of Electrical Engineering 227E Electrical Engineering West The Pennsylvania State University University Park, PA 16802, U.S.A. +1 (814) 865-0196, +1 (814) 865-7065 (fax) msznaier@frodo.ee.psu.edu

International Program Committee
Mich`le Basseville, e David Bayard, Robert Bitmead, Mogens Blanke, Jie Chen, Han-Fu Chen, Manfred Deistler, Bart De Moor, Guy Dumont, Peter Gawthrop, Laszlo Gerencs`r, e Michel Gevers, Laura Giarr?, e Mike Grimble, Lei Guo, Fred Hadaegh, France U.S.A. Australia Denmark U.S.A. China Austria Belgium Canada U.K. Hungary Belgium Italy U.K. China U.S.A. Shinji Hara, Japan H? akan Hjalmarsson, Sweden Rolf Isermann, Germany Hidenori Kimura, Japan Vikram Krishnamurthy, Australia Kousuke Kumamaru, Japan Ioan Landau, France Wally Larimore, U.S.A. Lennart Ljung, Sweden Pertti M¨kil¨, a a Finland Iven Mareels, Australia Tomas McKelvey, Sweden Mario Milanese, Italy Brett Ninness, Australia Kameshwar Poolla, U.S.A. Boris Polyak, Russia

3 Phillip Regalia, Dan Rivera, Johan Schoukens, Roy Smith (Chair), Torsten S¨derstr¨m, o o Roberto Tempo, France U.S.A. Belgium U.S.A. Sweden Italy Heinz Unbehauen, Paul Van den Hof, Bo Wahlberg, Eric Walter, Germany The Netherlands Sweden France

Welcome
It is a pleasure to join the Organizing and Program Committees in welcoming you to Santa Barbara for the SYSID 2000 in June. The stimulating technical program contrasts well with the relaxing location. We hope that you enjoy both aspects of SYSID 2000. The Symposium on System Identi?cation is IFAC’s longest running, and largest, symposium. It is without doubt the premier forum for the presentation of research and application results in our ?eld. SYSID is held every three years and the previous U.S. location was Baltimore in 1982. Santa Barbara is the ideal location for SYSID’s return to the U.S., being centrally located in California with close connections to a large number of active research centers; some local, some further a?eld. A total of 291 manuscripts were submitted to SYSID 2000. Each was sent to at least two members of the International Program Committee who were responsible for obtaining reviews. The ?nal program contains 226 papers. Of these 183 will be presented in a lecture format, 28 will be given in poster sessions and 15 are software demonstrations. We have adopted a novel scheduling arrangement for the poster sessions. One time slot—Thursday midday—will be dedicated solely to poster presentations. The three plenary talks cover an interesting range of topics, giving us a di?erent view of our own ?eld, as well as a look at other areas of potential application. The abstracts of the talks, and the biographies of the speakers are given in this program. David Luenberger, System Identi?cation in Modern Finance, Wednesday, 8.30 am. Jorma Rissanen Complexity and Information in Data, Thursday, 8.30 am. Tomas McKelvey, Frequency Domain Identi?cation, Friday, 8.30 am. The symposium venue is one of Santa Barbara’s premier resorts, located by the beach and close to the downtown area. Within a mile of the hotel you will also ?nd: the Santa Barbara Zoo; Stearns Wharf; the harbor; and the many restaurants and shops located in and around State Street. The region is known for its distinctive Mediterranean style architecture; beautiful beaches with a mountain backdrop; ocean and outdoor activities; numerous diverse restaurants; and excellent local vineyards. I hope that you enjoy your time in Santa Barbara. Roy Smith

4

Venue
The Symposium will be held at Fess Parker’s Doubletree Resort in the city of Santa Barbara, along the ocean beachfront and near the downtown area. Fess Parker’s Doubletree Resort 633 East Cabrillo Boulevard Santa Barbara, CA 93103 +1 (877) 893–0892

Transportation
The most convenient way of getting to Santa Barbara is by ?ying into Santa Barbara airport. A reasonable alternative is to ?y to Los Angeles and catch the Santa Barbara Airbus.

By Airline:
The Santa Barbara Municipal airport (SBA) has direct connections to Los Angeles, San Francisco, Denver, and Phoenix. The airport is approximately 8 miles (13 km) from the Doubletree Resort and the taxi fare to the Resort is approximately $25. Shuttle service is also available for $15. The Doubletree Resort also o?ers a free airport shuttle; however this must be reserved in advance. Refer to the Resort web site for details. The Los Angeles International Airport (LAX) is located approximately 100 miles (160 km) south of Santa Barbara. Rental cars are available at both the Santa Barbara and Los Angeles airports, but reservations should be made prior to arrival. Additional airport information is available on the websites for the Santa Barbara (http://www.?ysba.com) and Los Angeles (http://www.quickaid.com/airports/lax/) airports. The Santa Barbara Airbus is a bus service operating between LAX and several locations in Santa Barbara. There are 7 scheduled trips in each direction and the journey takes approximately 2 and a half hours. The stop in downtown Santa Barbara is at 1111 E. Cabrillo Blvd; several blocks from the symposium hotel. Call +1 (805) 964–7759 for schedule and price details.

By Train:
There is limited train service—one or two trains per day—to Santa Barbara from Los Angeles and San Francisco via the national railroad, Amtrak (http://www.amtrak.com). However, there is no easy connection between the airport and the train in either San Francisco or Los Angeles.

By Car:
Outdoor parking is available at the hotel. The driving instructions to the Doubletree Resort are shown below. Traveling North (from Los Angeles): 1. Take Highway 101 north to Santa Barbara. 2. In Santa Barbara, take the Cabrillo Blvd/Beach Area exit (It is a left exit.). 3. At the stop sign at the bottom of the o?-ramp, turn left onto Cabrillo Blvd.

5 4. Continue about 1.5 mile (2.4 km) to a tra?c signal at the intersection of Cabrillo and Milpas; turn right onto Milpas Street. 5. Within 0.5 mile (0.8 km), turn left onto Calle Puerto Vallarta. 6. The hotel entrance is the ?rst driveway on the right. Traveling South (from San Francisco or Santa Barbara airport): 1. Take Highway 101 south to Santa Barbara. 2. Exit at the Garden Street exit. Turn right onto Garden Street and follow it all the way to the end, less than a kilometer. 3. Turn left onto Cabrillo Boulevard. At the third light, turn left onto Calle Puerto Vallarta. 4. The hotel entrance is the ?rst driveway on the left.

Symposium Location Map
Downtown Santa Barbara
Ga
co cis ran nF Sa

Ca z ave Ch sar Ce lle
N

Mi s lpa

n rde

101 Freeway Bird Refuge Cabrillo to Los Angeles

Cabrillo

SYSID 2000 Calle Puerto Vallarta

to

East Beach

Pacific Ocean

Symposium Registration
The symposium registration costs are: Symposium attendee Student attendee Banquet tickets $US $US $US 450 75 50

In keeping with recent SYSID practice, the symposium banquet is not included in the registration fee. Banquet tickets cost $US 50 each and can be ordered on the registration form. The registration cost includes a preprint of the proceedings on CD, which will be given to attendees at the symposium. The preprints will not be mailed or made available for sale. The o?cial proceedings will be produced by Elsevier Science Ltd., and available for purchase after the symposium. Refund policy: Registration for the symposium may be refunded in full if a written request is received by the Registration Chair prior to 15th May, 2000.

6

Registration and Information Desk
The Registration and Information Desk is located on the ground ?oor of the hotel, near the main entrance. The Desk will be open during the following times: Tuesday, June 20: Wednesday, June 21: Thursday, June 22: Friday, June 23: 4.00 8.00 8.00 8.00 to 8.00 pm am to 6.00 pm am to 6.00 pm am to 12.00 noon

Registered attendees will receive a symposium package that contains the badge, ?nal program, proceedings preprint CD, tourist information, and one drink ticket for the Welcoming Reception. Tickets for the Symposium Banquet on thursday evening will be enclosed if they were ordered on the registration form.

Symposium Preprints
The symposium preprints will be in electronic (CD) form only. A printed copy will not be produced. Attendees will each receive a copy of the CD in their registration packets. Preprints cannot be mailed, or made available for sale. Please note that the o?cial proceedings of the symposium is produced by Elsevier Science Ltd., after the symposium.

Hotels
A limited number of rooms have been reserved at the symposium hotel, and are available to SYSID 2000 attendees at a reduced rate. Fess Parker’s Doubletree Resort 633 East Cabrillo Boulevard Santa Barbara, CA 93103 +1 (877) 893–0892 http://www.fpdtr.com The SYSID 2000 group discount rates (available only if reserved prior to the 20th May, 2000) are as follows: Singles Doubles Triples Quadruples $US $US $US $US 160 160 175 190

Tourist Information
Santa Barbara is a popular tourist destination due to its picturesque ocean and mountain scenery, excellent weather, and the many things there are to do and see. Tourist information will be enclosed with your registration package and will also be available at the hotel. Additional information is also available via the internet: http://www.santabarbaraca.com. The Tourist Information Center, open seven days a week, is located at 1 Santa Barbara Street and can be reached by phone at +1 (805) 965–3021. Santa Barbara in June is usually dry with daytime temperatures of 70 to 80 o F (21 to 27 o C). A light jacket or sweater is advisable for the evenings. Suntan lotion is recommended for sunny days. While early morning

7 clouds are frequent in Santa Barbara, afternoons tend to be sunny.

Downtown Santa Barbara

Santa Barbara Mission

Al a

me

N

da P adr eS

M

iss io n

Courthouse

err a
Sa lin as

St
Paseo Nuevo

ate

Ca rri llo

La g Ga una rd en

M

ilp as

101

Freeway

Zoo East Cabrillo East beach SYSID 2000

Bird refuge

Ca Bat sti h llo

W es tC ab ril lo

Stearns wharf

State Street is Santa Barbara’s main street and one place to go for a wide range of shopping and restaurants. Another collection of restaurants can be found on Cabrillo Boulevard close to Stern’s wharf. You will also ?nd places to dine on the wharf and across the harbor.

Social Program
A number of social events are planned for the Symposium. Welcoming Reception Plaza del Sol Tuesday, June 20 6.00 – 8.00 PM

An informal reception (casual dress and cash bar) will be held for Symposium attendees and their accompanying guests. A ticket for one free drink is included in the registration package. Companions Orientation Fiesta room (3rd ?oor) Wednesday, June 21 9.00 – 10.00 AM

This informal meeting will provide companions with an opportunity to become better acquainted and to learn about sightseeing and shopping in the Santa Barbara area. A member of the hotel sta? will make a brief presentation in this ocean view suite. Symposium Banquet Plaza del Sol Thursday, June 22 7.00 – 9.00 PM

The banquet will be held at the conference site. Banquet tickets should be ordered via the Symposium Registration form. The cost is $50 per person. And who knows; it is possible that Hollywood movie stars and other celebrities will attend. Don’t miss this gala event.

8 Winery Tour Saturday, June 24 10:00 AM - 3:00 PM

With over 40 wineries, Santa Barbara County is an ideal place to experience excellent wines. This guided tour is scheduled for the day after the Symposium. It will feature a winery tour and tasting of outstanding local wines. Enjoy the scenic bus ride from Santa Barbara over the San Marcos Pass into the beautiful Santa Ynez Valley. The cost is $45 per person including bus transportation, wine tasting fees, and a picnic lunch. Because space is limited to 24 people, reservations are required and should be made early at the Symposium web site: http//www.ece.ucsb/ccec/SYSID2000/. Other Events There are many other events of potential interest to the conference attendees. One example is the “Nite Moves” running race series. This is a 5 kilometer road race, held on at 6.00pm on Wednesday evenings during the summer at Ledbetter Park on West Cabrillo Blvd. All ability levels are welcome, from walkers to international athletes. The cost is approximately $20 and includes dinner, beer, and live music on the beach.

IFAC Copyright Conditions
The material submitted for presentation at an IFAC meeting (congress, symposium, conference, workshop) must be original, not published or being considered elsewhere. All papers accepted for presentation will appear in the Preprints of the meeting and will be distributed to the participants. Papers duly presented will be archived and o?ered for sale, in the form of Proceedings, by Elsevier Science Ltd., Oxford, UK. The papers which have been presented will be further screened for possible publication in the IFAC Journals Automatica and Control Engineering Practice, or in other IFAC a?liated journals. The abstracts of all papers presented will be recorded in Control Engineering Practice. Copyright of material presented at an IFAC meeting is held by IFAC. Authors will be sent a copyright transfer form. Automatica, Control Engineering Practice and, after these, IFAC a?liated journals have priority access to all contributions presented. However, if the author is not contacted by an editor of these journals, within three months after the meetings, he/she is free to re-submit the material for publication elsewhere. In this case, the paper must carry a reference to the IFAC meeting where it was originally presented.

Plenary Lectures
Plenary lecture 1
David Luenberger, Stanford University, California, System Identi?cation in Modern Finance Abstract. Modern ?nance theory has revolutionized the way people think about investments, and has spawned a new generation of analytically based investment methods. However, outstanding theories have not always gained great practical acceptance. Some elements of modern ?nance, such as option pricing theory, have had great practical success, while other equally profound theories, such as portfolio design, have been less widely used. In this talk we argue that the degree of practical success enjoyed by an element of ?nancial theory is greatly in?uenced by the nature of the system identi?cation problems required for implementation of the theory. We shall discuss system identi?cation issues associated with options theory, portfolio design, technical

9 trading, and valuation of publicly traded ?rms; and we shall argue that success in some of these areas is inherently limited by the di?culty of identi?cation. Biography. David G. Luenberger was born on September 16, 1937. He graduated from Caltech in 1959 with a BS in EE, and from Stanford University in 1961 and 1963 with MS and Ph.D. degrees in EE. His Ph.D. dissertation introduced the concept of observers for dynamic systems. He joined the Stanford faculty in 1963. He has carried out research in control theory, optimization and mathematical programming, economics, and ?nance. In the course of this research he has authored over 80 technical publications and 5 textbooks and supervised 43 Ph.D. dissertations. He was a founding member of the Department of Engineering-Economic Systems at Stanford and served as its chairman from 1980 to 1991. He has been active in several application areas, serving as a consultant to several companies; and in 1971–72 he was technical assistant to the President’s Science Advisor in Washington, DC. He has been an associate editor of several journals and served as chairman of the Society of Economic Dynamics and Control. Prof. Luenberger was awarded the Bode Prize of the IEEE Control Systems Society, the Oldenburger Medal of the ASME, and the Expository Writing Award of INFORMS. His current research is in the area of Investment Science with a particular focus on the application of ?nance and systems methods to business issues.

Plenary lecture 2
Jorma Rissanen, IBM Research, San Jose, California, Complexity and Information in Data. Abstract. We describe recent results in model building and selection based on a formal de?nition of complexity and information in a given data sequence. This permits a decomposition of data into the information bearing part and noise and a similar parameter free su?cient statistics factorization of a model which is universal for a parametric class. Biography. Jorma Rissanen was born October 20, 1932 in Finland. He received the Licentiate and the Doctor of Technology degrees in control theory and mathematics from the Technical University of Helsinki in 1960 and 1965, respectively. He has been with IBM Research in San Jose since 1966, except for the academic year 1973–74, when he held the chair of control theory in Link¨ping University, Sweden. o He has done research in control, prediction, and system theories, relation theory, numerical mathematics, information and coding theory, probability theory and statistics. He received in 1998 an IEEE Information Society Golden Jubilee Award for Technological Innovation for the invention of arithmetic coding, the IEEE 1993 Richard W. Hamming medal “For fundamental contributions to information theory, statistical inference, control theory, and the theory of complexity”, IBM Corporate Award in 1991 for the MDL/PMDL principles and stochastic complexity, IBM Outstanding Innovation Award in 1988 for work in statistical inference, information theory, and the theory of complexity, Best Paper Award from IEEE Information Theory Group in 1986, Best Paper Award from IFAC in 1981. He received an Honorary Doctorate from the Technical University of Tampere, Finland, in 1992. He is a Visiting Professor at the University of London, Royal Holloway. He is also an Advisory Editor of the Journal of Statistical Planning and Inference, a Member of the Advisory Board of the IEICE Transactions on

10 Fundamentals of Electronics, Communications and Computer Sciences (Japan), an Associate Editor of the IMA Journal of Mathematical Control and Information, and a former Associate Editor on source coding of the IEEE Transactions on Information Theory.

Plenary lecture 3
Tomas McKelvey, Link¨ping University, Sweden, o Frequency Domain Identi?cation Abstract. Techniques to identify parametric transfer functions from noisy frequency domain data are considered. A maximum-likelihood estimation method is presented which in parallel with the system transfer function also estimates a parametric noise transfer function. This leads to a consistent and e?cient estimator. It is shown how the discrete Fourier transform can be applied to generate frequency domain data from sampled time domain data. For the ?nite data case the exact frequency domain expressions are derived relating the transfer function with the discrete Fourier transformed data for both continuous and discrete time systems. Biography. Tomas McKelvey is an Associate Professor of Automatic Control in the Department of Electrical Engineering at Link¨ping University, Sweden. He was born o 1966 in Lund, Sweden. He received his M.Sc. degree in Electrical Engineering from Lund Institute of Technology, Sweden in 1991 and his Ph.D. degree in Automatic Control from Link¨ping University, Sweden in 1995. He is currently holding a longer o visiting position with the University of Newcastle, Australia. His main scienti?c interests are system identi?cation, time series analysis and signal processing.

Symposium site ?oor plan
Third Floor
Elevator Anacapa Patio Fiesta Anacapa Santa Cruz Santa Rosa San Miguel Entrance RR RR Front Desk Lobby Cafe Los Arcos Maxi's Elevator Barra Los Arcos North San Rafael Entrance Santa Ynez Patio Patio

Grand Ballroom San Rafael Santa Ynez

Sierra Madre
South

Plaza Del Sol
RR

RR

Lobby Level

Maxi's Lounge

11

SYSID 2000: Program summary
Tuesday 20th June 2000 6.00 to 8.00 pm Welcome reception (Plaza del Sol )

Wednesday 21st June 2000 8.30 to 9.30 am 10.00 am to 12.00 1.30 to 3.30 pm 4.00 to 6.00 pm Identi?cation for Robust Control
(Sierra Madre South)

Plenary: David Luenberger (San Rafael ) Neural Networks
(San Miguel )

Subspace Methods I
(Santa Rosa)

Biological Applications
(Santa Cruz )

Model Error Modeling
(San Miguel )

Continuous Time Methods
(Sierra Madre South)

Subspace Methods II
(Santa Rosa)

Environmental Modeling
(Santa Cruz )

Time Varying Systems
(Sierra Madre North)

Identi?cation for Control
(San Miguel )

Continuous Time Identi?cation
(Sierra Madre South)

Blind ID & Equalization
(Santa Rosa)

Model Selection
(Santa Cruz )

Thursday 22nd June 2000 8.30 to 9.30 am 10.00 am to 12.00 1.30 to 3.30 pm 4.00 to 6.00 pm 7.00 to 9.00 pm Adaptive Control I
(San Miguel )

Plenary: Jorma Rissanen (San Rafael ) Process Applications I
(Santa Rosa)

Orthogonal Basis Functions
(Sierra Madre South)

Fault Detection
(Santa Cruz )

Software Demos I
(Sierra Madre North)

Adaptive Control II
(Santa Ynez )

Process Applications II
(Santa Ynez )

Automotive & Mechanical Appl.
(Santa Ynez )

Identi?cation Applications
(Santa Ynez )

Software Demos II
(Sierra Madre North)

Closed-loop Identi?cation
(San Miguel )

Recursive Identi?cation
(Santa Rosa)

Nonlinear ID Bases
(Santa Cruz )

Education
(Sierra Madre South)

Software Demons III
(Sierra Madre North)

Banquet (Plaza del Sol )

Friday 23rd June 2000 8.30 to 9.30 am 10.00 am to 12.00 1.30 to 3.30 pm 4.00 to 6.00 pm Bounded Error Estimation I
(Sierra Madre South)

Plenary: Tomas McKelvey (San Rafael ) Parameter Estimation
(San Miguel )

Nonlinear Identi?cation I
(Santa Rosa)

Distributed Parameter Syst.
(Santa Cruz )

Bounded Error Estimation II
(Sierra Madre South)

Estimation and Filtering
(San Miguel )

Nonlinear Identi?cation II
(Santa Rosa)

Input Sequences
(Santa Cruz )

Set Membership Identi?cation
(Sierra Madre South)

Stochastic Est. & Control
(San Miguel )

Nonlinear Identi?cation III
(Santa Rosa)

Experiment design
(Santa Cruz )

12

SYSID 2000 Technical Program Schedule
Note: The page numbers associated with each title refer to the abstracts included in this booklet. An asterisk, ?, indicates that no abstract was received.

Wednesday morning:
10.00 am to 12.00 noon 11.00 am Set membership identi?cation for H∞ robust control design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

Plenary Lecture 1
8.30 to 9.30 am

San Rafael

Mario Milanese Michele Taragna

Politecnico di Torino, Italy Politecnico di Torino, Italy

System identi?cation in modern ?nance . . . . . . . . . . . . . . . . . 8 David G. Luenberger Stanford University, USA

11.20 am Closed-loop model validation using coprime factor uncertainty models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 Raymond A. de Callafon Univ. California, San Diego Paul M.J. Van den Hof Delft University of Technology

WeAM1

Identi?cation for Robust Control Sierra Madre South
Invited Session Organizers: M. Gevers CESAME, Louvain-la-Neuve, Belgium M. Milanese Politecnico di Torino, Italy Chairs: M. Gevers M. Milanese CESAME, Louvain-la-Neuve, Belgium Politecnico di Torino, Italy

11.40am Panel discussion

WeAM2

Neural Networks
Chairs: S. Toepfer J.B. Gomm

San Miguel

Tech. Universitat Darmstadt, Germany Liverpool John Moores University, UK

10.00 am Model validation for robust control: experiment design issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? M. Gevers X. Bombois B. Codrons F. De Bruyne G. Scorletti 10.20 am Uncertainty model falsi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . ? R. Kosut 10.40 am Model error modeling and control design . . . . . . . . . . . . . . . . . ? L. Ljung

10.00 am Studies on initialization for multilayer networks . . . . . . . . 34 Hiroshi Shiratsuchi Hiromu Gotanda Katuhiro Inoue Kousuke Kumamaru Univ. of the Ryukyus Kinki Univ. in Kyushu Kyushu Institute of Technology Kyushu Institute of Technology

10.20 am On the dynamics of local linear model networks with orthonormal basis functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Oliver Nelles Masayoshi Tomizuka UC Berkeley, ME Dept UC Berkeley, ME Dept.

13

10.40 am Backward centre selection method for on-line adaptation of RBF network models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 J. Barry Gomm Ding Li Yu Liverpool John Moores Univ., UK Liverpool John Moores Univ., UK

10.40 am On the impact of weighting matrices in subspace algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Dietmar Bauer Manfred Deistler Wolfgang Scherrer TU Wien, Austria TU Wien, Austria TU Wien, Austria

11.00 am Comparison of a Hierarchically Constructed Neural Network and a Hierarchical Look-up Table . . . . . . . . . . . . . . 35 Susanne T¨pfer o Oliver Nelles Rolf Isermann. Darmstadt University of Technology University of California at Berkeley Darmstadt University of Technology

11.00 am Weighting in subspace-based system identi?cation . . . . . . . 36 Tony Gustafsson University of California San Diego

11.20 am Optimal weighting for MOESP type of procedures . . . . . . . 36 Dietmar Bauer Tony Gustafsson TU Wien, Austria University of California San Diego

11.20 am An interpolation technique for learning with sparse data . ? J. Van Gorp Y. Rolain 11.40 am Simulation of central pattern generators by using populations of chaotic neurons . . . . . . . . . . . . . . . . . . . . . . . . . 35 P. Arena L. Fortuna D. Garofalo University of Catania, DEES University of Catania DEES University of Catania DEES

11.40 am Subspace algorithms for data from varying sensor locations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Albert Benveniste Mich`le Basseville e Maurice Goursat Laurent Mevel IRISA/INRIA, IRISA/CNRS, INRIA, IRISA/INRIA, France France France France

WeAM4 WeAM3

Biological Applications Santa Rosa
Adaptics Inc., USA Link¨ping Universitet, Sweden o Chairs: E. Walter K. Keesman

Santa Cruz

Subspace Methods I
Chairs: W. Larimore T. McKelvey

L2S CNRS–SUPELEC–UPS, France Wageningen Univ., The Netherlands

10.00 am Error analysis of certain subspace methods . . . . . . . . . . . . . 35 Chiuso Alessandro Picci Giorgio Univ. di Padova Univ. di Padova

10.00 am State and parameter estimation in biotechnical batch reactors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Karel J. Keesman Wageningen Univ., The Netherlands 10.20 am Parameter estimation that improves validity of model for a certain model use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Mario Zec Nadja Hvala Stanko Strmcnik J. Stefan Institute, Slovenia J. Stefan Institute, Slovenia J. Stefan Institute, Slovenia

10.20 am Asymptotic variance analysis of subspace identi?cation methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Magnus Jansson Royal Inst. of Tech. (KTH)

14

10.40 am Recursive identi?cation of activated sludge wastewater treatment plants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Oscar A. Z. Sotomayor Song W. Park Claudio Garcia LSCP-EPUSP, Brazil LSCP-EPUSP, Brazil PTC-EPUSP, Brazil

1.50 pm Model error model from identi?cation in closed-loop . . . . . . ? A. Besan?on-Voda c 2.10 pm On model error modeling in set membership identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 A. Garulli W. Reinelt Universit` di Siena a Link¨ping University o

11.00 am Practical implementation of the optimal experiment design methodology for estimation of microbial heat resistance parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 K.J. Versyck K. Bernaerts J.F. Van Impe BioTeC - KULeuven, Belgium BioTeC - KULeuven, Belgium BioTeC - KULeuven, Belgium

2.30 pm Preliminary test for nonlinear input output relations in SISO systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Torben Knudsen Aalborg University, Denmark

11.20 am System identi?cation of electronic nose data for monitoring cyanobacteria in potable water: black-box modelling . . . . . 38 G.E. Searle J.W. Gardner M.J. Chappell K.R. Godfrey M.J. Chapman University of Warwick, University of Warwick, University of Warwick, University of Warwick, University of Coventry, UK UK UK UK UK

2.50 pm Exact bounds for the frequency response of an uncertain plant with ellipsoidal perturbations . . . . . . . . . . . . . . . . . . . . . 40 G. A. A. A. Chesi Garulli Tesi Vicino Universit` di Siena a Universit` di Siena a Universit` di Firenze a Universit` di Siena a

11.40 am Metabolism and cell cycle modelling by means of neural network based structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Andreia Hanomolo Philippe Bogaerts Raymond Hanus University Libre de Bruxelles University Libre de Bruxelles University Libre de Bruxelles

3.10 pm A new orthogonal basis functions for error quanti?cation in system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 R. Ouvrard P. Coirault T. Poinot J.C. Trigeassou L.A.I.I.-E.S.I.P. L.A.I.I.-E.S.I.P. L.A.I.I.-E.S.I.P. L.A.I.I.-E.S.I.P. France France France France

Wednesday midday:
1.30 pm to 3.30 pm

WeMD2

Continuous Time Methods
Chairs: T. S¨derstr¨m o o P. Gawthrop

Sierra Madre South

WeMD1

Uppsala Universitet, Sweden University of Glasgow, Scotland, UK

Model Error Modeling and Model Validation San Miguel
Chairs: F. Hadaegh A. Garulli Jet Propulsion Laboratory, USA Universit` di Siena, Italy a

1.30 pm Box-Jenkins continuous-time modeling . . . . . . . . . . . . . . . . . 40 Rik Pintelon Johan Schoukens Yves Rolain Vrije Universiteit Brussel Vrije Universiteit Brussel Vrije Universiteit Brussel

1.30 pm A nonparametric approach to model error modeling . . . . . 39 Anders Stenman Fredrik Tj¨rnstr¨m a o Link¨pings universitet, Sweden o Link¨pings universitet, Sweden o

15

1.50 pm Identi?cation methods of dynamic systems in presence of input noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Umberto Soverini Torsten Soderstrom University of Bologna Uppsala University

2.10 pm On state estimation of unknown systems based on subspace state-space system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . 42 Hiroshi Oku Hidenori Kimura University of Tokyo University of Tokyo

2.10 pm Frequential moments : application tolinear system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 E. Etien J.D Gabano J.C Trigeassou LAII LAII LAII

2.30 pm Subspace system identi?cation of multiple time series and di?erent sampling rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? D.D. Ruscio 2.50 pm On the rank de?ciency of the least squares residuals in subspace identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? T.V. Gestel B. De Moor P.V. Overschee

2.30 pm On the statistical accuracy of the empirical transfer function estimator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 P.M.T. Broersen Delft Univ. of Technology

2.50 pm Approaches for identifying continuous-time AR processes from unevenly sampled data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 E. K. Larsson T. S¨derstr¨m o o Uppsala University, Sweden Uppsala University, Sweden

WeMD4

Environmental Modeling

Santa Cruz

Invited Session Organizer: I. Mareels University of Melbourne, Australia Chairs: I. Mareels P. Young University of Melbourne, Australia Lancaster University, UK

3.10 pm Transfer function and frequency response estimation using resonant ?lters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Peter J. Gawthrop Liuping Wang University of Glasgow The University of Newcastle

1.30 pm Linear material ?ow models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 Andreas Gleiss Manfred Deistler Thomas Matyus Technical University, Vienna Technical University, Vienna Technical University, Vienna

WeMD3

Subspace Methods: II
Chairs: B. Ninness J. Maciejowski

Santa Rosa

University of Newcastle, Australia Cambridge University, UK

1.50 pm Time variable and state dependent parameter estimation . ? P. Young

1.30 pm Subspace identi?cation method for combined deterministic-stochastic bilinear systems . . . . . . . . . . . . . . . . 42 Huixin Chen Jan M Maciejowski University of Cambridge, U.K University of Cambridge, U.K.

2.10 pm System identi?cation of an open water channel . . . . . . . . . 43 Erik Weyer University of Melbourne, Australia

1.50 pm Identi?cation of colinear and cointegrated multivariable systems using canonical variate analysis . . . . . . . . . . . . . . . . 42 Wallace E. Larimore Adaptics, Inc

2.30 pm Identi?cation of interactions of runo? rainfall state variables for modelling hourly ?ows of a fast ?owing mountain river . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? R. Chaudhuri A. Chaudhuri

16

Wednesday afternoon:
2.50 pm Estimation of pollutant source speci?cations using a conditional deconvolution method . . . . . . . . . . . . . . . . . . . . . . . 43 Delmaire Benjelloun LASL, France LASL, France 4.00 pm to 6.00 pm

WePM1

Identi?cation for Control
WeMD5

San Miguel

Time-varying systems
Chairs: K. Kumamaru L. El Ghaoui

Sierra Madre North

Chairs: X. Bombois CESAME, Louvain-la-Neuve, Belgium A. Besan?on-Voda c ENSIEG, Grenoble, France

Kyushu Institute of Technology, Japan University of California, Berkeley, USA

4.00 pm A nearly interpolatory algorithm for H∞ identi?cation with mixed time and frequency response data . . . . . . . . . . . . ? G. Gu J. Chen 4.20 pm Use of error criteria in identi?cation for control . . . . . . . . 45 Yucai Zhu Eindhoven Univ. Tech. & Tai-Ji Control

1.30 pm Identi?cation of time-varying systems - a survey . . . . . . . . 44 Evgeni G. Kleiman Hannover University, Germany

1.50 pm Identi?cation for a general class of LPV models . . . . . . . . . ? B. Bamieh L. Giarre 2.10 pm Identi?cation of ARX models with time-varying bounded parameters: A semide?nite programming approach . . . . . . . ? L. El Ghaoui University of California, Berkeley, USA Giuseppe Cala?ore Politecnico di Torino, Italy 2.30 pm H∞ parameter identi?cation for in?ight detection of aircraft icing: the time-varying case . . . . . . . . . . . . . . . . . . . . 44 James W. Melody Univ. Illinois at Urbana-Champaign Thomas Hillbrand Darmstadt Univ. Technology Tamer Basar Univ. Illinois at Urbana-Champaign William Perkins Univ. Illinois at Urbana-Champaign 2.50 pm Impulse response model identi?cation in the case of a multi-period ZOH input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Zi-Jiang Yang Yoshihiro Kodama Kyushu Institute of Technology Kyushu Institute of Technology

4.40 pm The problem of constructing strong models with a minimum operator norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? V. Rusanov A. Daneev 5.00 pm Robust identi?cation in H∞ of non-Gaussian systems . . 45 Vojislav Filipovic CNTS,Yugoslavia

5.20 pm Approaches to control-oriented H∞ identi?cation . . . . . . . 45 Mehrzad Namvar Alina Voda Lab. d’Auto. de Grenoble, France

5.40 pm Controller validation for stability and performance based on an uncertainty region designed from an identi?ed model . 46 X. Bombois M. Gevers G. Scorletti B.D.O. Anderson Universit? Catholique de Louvain e Universit? Catholique de Louvain e LAP ISMRA Australian National University

17
WePM2

What has Continuous time System Identi?cation to O?er? Sierra Madre South
Invited Session Organizers: H. Madsen Tech. University of Denmark, Denmark T. S¨derstr¨m o o Uppsala Universitet, Sweden Chairs: H. Madsen T. S¨derstr¨m o o Tech. University of Denmark, Denmark Uppsala Universitet, Sweden

5.00 pm Blind system identi?cation and equalization . . . . . . . . . . . . . ? E-W. Bai 5.20 pm Direction estimation of coherent signals without eigendecomposition and spatial smoothing . . . . . . . . . . . . . . 47 Jingmin Xin YRP Mobile Telecom. Key Tech. Res. Lab Akira Sano Keio University

Panel discussion: 4.00 pm to 6.00 pm. Panelists: Graham Goodwin, Henrik Madsen, Johan Shoukens, Peter Young,

University Newcastle, Australia Tech. Univ. Denmark, Denmark Free Univ. Brussels, Belgium Lancaster University, UK

WePM4

Model Selection
Chairs: A. Stenman K. Nakano

Santa Cruz
Link¨ping Universitet, Sweden o Univ. Electro-Communications, Japan

WePM3

Blind Identi?cation, Equalization and Source 4.00 pm Localization Santa Rosa Information criterion for selection of model with controller
Chairs: B. Wahlberg E.-W. Bai design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? Royal Institute of Technology, Sweden University of Iowa, USA K. Tsumura 4.20 pm System identi?cation using a multi-model approach: model complexity reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 Anass Boukhris Gilles Mourot Jos? Ragot e CRAN, France CRAN, France CRAN, France

4.00 pm Blind estimation and deconvolution of communication channels with unbalanced noise . . . . . . . . . . . . . . . . . . . . . . . . . 46 Paolo Castaldi Roberto Diversi Roberto P. Guidorzi Umberto Soverini Universit` a Universit` a Universit` a Universit` a di di di di Bologna Bologna Bologna Bologna

4.20 pm Adaptive identi?cation and equalization for time-varying channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Takayuki Naito Keio University, Hiromitsu Ohmori Keio University, Akira Sano Keio University, Kohichi Hidaka Tokyo Coll. Aeronautical Eng., Japan Japan Japan Japan

4.40 pm On model structure selection for nonparametric prediction methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 Anders Stenman Link¨pings universitet, Sweden o

5.00 pm Model order selection of N-FIR models by the analysis of variance method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 Ingela Lind Linkoping University, Sweden

4.40 pm Blind localization by subspace method for a scattering model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Eric Ternisien Gilles Roussel Mohammed Benjelloun LASL, France LASL, France LASL, France

5.20 pm Algorithm for three-step estimation of transfer function with unknown delay steps and order . . . . . . . . . . . . . . . . . . . . 48 Mitsuki Mashino Fukuoka Indust. Tech. Center, Japan Fujio Ohkawa Kyushu Institute Technology, Japan Kazushi Nakano Univ. Electro-Communications, Japan Masayoshi Tomizuka Univ. California, Berkeley, USA

18

5.40 pm Automatic model selection for linear time-invariant systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Gyula Simon Johan Schoukens Yves Rolain BUTE, Hungary Vrije Universiteit, Brussel, Belgium Vrije Universiteit, Brussel, Belgium

11.00 am Adapative control of stochastic strict-feedback systems under a risk - sensitive criterion . . . . . . . . . . . . . . . . . . . . . . . . ? G. Arslan T. Basar 11.20 am An estimation algorithm for stable adaptive control of not necessarily minimum phase systems with bounded disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Leonid S. Zhiteckij Inst. Cybernetics, Kiev, Ukraine

Thursday morning:
10.00 am to 12.00 noon

Plenary Lecture 2
8.30 to 9.30 am

San Rafael ThAM2 Process Applications: I
Chairs: B. Juricek V. Verdult

Santa Rosa

Complexity and information in data . . . . . . . . . . . . . . . . . . . . . 9 Jorma Rissanen IBM Research, California, USA

Univ. of California, Santa Barbara, USA University of Twente, The Netherlands

ThAM1

Adaptive Control: I

San Miguel

10.00 am Identi?cation of the Tennessee Eastman challenge process with subspace methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Ben C. Juricek Dale E. Seborg Wallace E. Larimore Univ. California, Santa Barbara Univ. California, Santa Barbara Adaptics, Inc.

Chairs: Y. Miyasato Inst. Statistical Mathematics, Japan T. Basar Univ. of Illinois, Urbana-Champaign, USA

10.00 am General forms of adaptive nonlinear H∞ control for processes with bounded variations of parameters . . . . . . . . 49 Yoshihiko Miyasato Inst. Statistical Math., Japan

10.20 am Nonlinear identi?cation of high purity distillation columns? C.T. Chou H.J.J. Bloemen V.Verdult T.J.J. van den Boom T. Backx M. Verhaegen 10.40 am Qualitative modelling as a key technique for the automatic identi?cation of mathematical models of chemical reaction systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 David Schaich Ralf Becker Rudibert King IPAT, TU Berlin IPAT, TU Berlin IPAT, TU Berlin

10.20 am Robust output tracking for uncertain strict-feedback systems with unknown virtual control coe?cients . . . . . . . . ? G. Arslan T. Basar 10.40 am Partial convergence of coupled time-varying systems . . . . 50 Keum-Shik Hong Kyung-Jinn Yang Pusan National Univ., Korea Pusan National Univ., Korea

19

11.00 am Combining ?rst principles with black-box techniques for reaction systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Libei Chen Yves Hontoir Dexian Huang Jie Zhang Julian Morris Catholic Univ. Louvain, Belgium Solvay s.a., Belgium Tsinghua University, China Univ. Newcastle-upon-Tyne, UK Univ. Newcastle-upon-Tyne, UK

11.00 am Optimal pole locations for Laguerre and two-parameter Kautz models: a survey of known results . . . . . . . . . . . . . . . . . ? T. Oliveira e Silva 11.20 am Practical aspects of using orthonormal system parameterisations in estimation problems . . . . . . . . . . . . . . . 52 Brett Ninness Stuart Gibson Steve Weller University of Newcastle University of Newcastle University of Newcastle

11.20 am On modelling and control of a rotary sugar dryer . . . . . . . 51 Sergio M. Savaresi Robert R. Bitmead Robert D. Peirce Politecnico di Milano University of California San Diego CSR Ltd, Australia

11.40 am Partial realization in generalized bases: algorithm and example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Thomas J. de Hoog Peter S.C. Heuberger Paul M.J. Van den Hof Delft University of Technology Delft University of Technology Delft University of Technology

11.40 am Model-on-Demand identi?cation for control: an experimental study and feasibility analysis for MoD-based predictive control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Martin W. Braun Brian A. McNamara Daniel E. Rivera Anders Stenman Arizona State Arizona State Arizona State Link¨ping o University University University University

ThAM4

Fault Detection and Monitoring
Chairs: M. Basseville K. Kumamaru

Santa Cruz

ThAM3

IRISA/CNRS, France Kyushu Institute of Technology, Japan

Modeling and Identi?cation with Orthogonal Basis Functions Sierra Madre South
Invited Session Organizers: P. Van den Hof Delft Univ. Tech., The Netherlands B. Wahlberg Royal Institute of Technology, Sweden Chairs: P. Van den Hof B. Wahlberg Delft Univ. Tech., The Netherlands Royal Institute of Technology, Sweden

10.00 am A nonlinear adaptive observer based method for fault detection and isolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Qinghua Zhang IRISA-INRIA

10.20 am Frequency domain local tests for change detection. . . . . . . 53 Albert Benveniste Bernard Delyon Mich`le Basseville e IRISA/INRIA, France IRMAR, France IRISA/CNRS, France

10.00 am Modelling and identi?cation with rational orthogonal basis functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Paul Van den Hof Delft Univ. Tech., Netherlands Bo Wahlberg Royal Inst. Technology, Stockholm Peter Heuberger Delft Univ. Tech., Netherlands Brett Ninness Univ. of Newcastle, Australia Jozsef Bokor SZTAKI, Hungarian Acad. Sciences Tom?s Oliveira e Silva a University of Aveiro, Portugal

10.40 am Optimal auxiliary input design for fault detection of systems with model uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . 53 Toshiharu Hatanaka Katsuji Uosaki Tottori Univ., Japan Tottori Univ., Japan

11.00 am Detection and isolation of sensor and process faults in a heat exchanger using a fuzzy-model library . . . . . . . . . . . . . . 53 Karsten Spreitzer Peter Ball TU Darmstadt - IAT TU Darmstadt - IAT

20
ThMD1 11.20 am A geometric based parametrization of the ?lter gains for the fault . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 Ricardo Lopezlena Estrada, Instituto Mexicano del Petr?leo, M?xico o e Juan Carlos Mart? ?nez Garc? ?a, CINVESTAV-IPN, M?xico. e

Adaptive Control: II
Poster session: 1.30pm to 3.30pm

Santa Ynez

ThAM5

Software Demonstration Session I (General) Sierra Madre North
Invited Session Organizer: R. Schumann Fachhochschule Hannover, Germany Chairs: R. Schumann H. Barker Fachhochschule Hannover, Germany Univ. Wales Swansea, U.K.

Direct adaptive control of a linear parabolic system . . . . . 55 ? St?phane Renou e Ecole Polytechnique de Montral ? Michel Perrier Ecole Polytechnique de Montral Denis Dochain Universit? Catholique de Louvain e Sylvain Gendron PAPRICAN

Adaptive position control of a ?exible manipulator using ANNNAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Torsten Knohl Hanqing Zeng Heinz Unbehauen Ruhr-Universit¨t Bochum a Ruhr-Universit¨t Bochum a Ruhr-Universit¨t Bochum a

Software demonstation session: 10.00 am to 12.00 noon Delta modi?cation of self-tuning pole placement PID controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Vladim? Bob?l ir a Brno Univ. Tech., Czech Republic Petr Dost?l a Brno Univ. Tech., Czech Republic Martin Sysel Brno Univ. Tech., Czech Republic

Galois - a program for generating pseudo-random perturbation signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 H. A. Barker Univ. Wales Swansea, U.K.

Tai-ji ID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? Yucai Zhu

Modeling and adaptive control for a two-link ?exible manipulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 A. Krolikowski P. Banaszczyk Poznan University of Technology Poznan University of Technology

VCLab - the virtual control engineering laboratory . . . . . . 54 Christian Schmid Ruhr-Universitaet Bochum, Germany Robust winder control system using adaptive model output following control method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Masanori Takahashi Ariake National Coll. Tech., Yoshinori Kawasaki Ariake National Coll. Tech., Ikuro Mizumoto Kumamoto University, Motoyuki Kuribayashi Mitsui Miike Machinery, Tomohiro Yasukouchi Mitsui Miike Machinery, Japan Japan Japan Japan Japan

WINROSA 2.0 and DORA for Windows 6.3 . . . . . . . . . . . 54 Peter Krause Angelika Krone Timo Slawinski Rainer Knicker University University University University of of of of Dortmund Dortmund Dortmund Dortmund

Thursday midday:
1.30 pm to 3.30 pm

Adaptive control mixer method for nonlinear control recon?guration: a case study . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Zhenyu Yang Mogens Blanke Aalborg University, Denmark Aalborg University, Denmark

21
ThMD2 ThMD3

Process Applications: II
Poster session: 1.30pm to 3.30pm

Santa Ynez

Automotive and Mechanical Applications Santa Ynez
Poster session: 1.30pm to 3.30pm

System identi?cation and physical parameter estimation of anti-vibration units in semiconductor exposure apparatus56 Shuichi Adachi Hiroyuki Takanashi Hiroaki Kato Takehiko Mayama Shinji Wakui Utsunomiya University Utsunomiya University CANON Inc CANON Inc CANON Inc Continuous time hybrid identi?cation approach for on line tool wear estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 F. J. Carrillo F. Rotella M. Zadshakoyan LGP ENI Tarbes, France LGP ENI Tarbes, France LGP ENI Tarbes, France

Experimental comparison of continuous-time model identi?cation methods on a thermal process . . . . . . . . . . . . . 57 M. Mensler H. Garnier E. Huselstein Kyushu University, Japan Ctr. Recherche Auto. de Nancy, France Ctr. Recherche Auto. de Nancy, France

Low order H∞ control design for a piezo-based milli-actuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 Max Rotunno Raymond A. de Callafon Univ. California, San Diego Univ. California, San Diego

Robust Smith predictor design via uncertainty quanti?cation: application to a reclaimer . . . . . . . . . . . . . . . 57 Keum-Shik Hong Dong-Hunn Kang Jeom-Goo Kim Pusan National University, Korea Pusan National University, Korea Pusan National University, Korea

A new fuzzy logic approach to EGR control . . . . . . . . . . . . . 59 Karsten Spreitzer TU Darmstadt - IAT

Local linear model tree (LOLIMOT) for nonlinear system identi?cation of a turbocharger with variable turbine geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 Jochen Scha?nit Oliver Nelles Rolf Isermann Wolfram Schmid Darmstadt University of Technology Darmstadt University of Technology Darmstadt University of Technology DaimlerChrysler AG

Nonlinear system identi?cation of rapid thermal processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Caizhong Tian Takao Fujii Osaka university Osaka university

Neuro-based modularized modeling and its application to deaerator with level control systems . . . . . . . . . . . . . . . . . . . . 58 Yukihiro Toyoda Bailey Japan Co., Ltd., Kazushi Nakano Univ. Electro-Communications, Takami Matsuo Oita University, Kohji Higuchi Univ. Electro-Communications, Japan Japan Japan Japan

Identi?cation of spark ignition engine models based on neural network via experimental design techniques . . . . . . 59 Ivan Arsie Fabrizio Marotta Cesare Pianese Gianfranco Rizzo University University University University of of of of Salerno, Salerno, Salerno, Salerno, Italy Italy Italy Italy

Corrosion prediction in pulp and paper industry with neuro-fuzzy technique. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 Maide Bucolo Universit` degli Studi di Catania, Italy a Luigi Fortuna Universit` degli Studi di Catania, Italy a Martin Nelke Management Intell. Tech., Germany Alessandro Rizzo Univ. degli Studi di Catania, Italy Tatiana Sciacca Univ. degli Studi di Catania, Italy

Preserving stability/performance when facing an unknown time-delay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 S. Diop Supelec, France I. Kolmanovsky Ford Research Lab., Michigan P. Moraal Ford Forshugszentrum Aachen, Germany M. van Nieuwstadt Ford Research Lab., Michigan

22
ThMD4

Identi?cation Applications
Poster session: 1.30pm to 3.30pm

Santa Ynez
Robust autonomous robot tracking using interval analysis62 M. Kie?er L. Jaulin E. Walter D. Meizel L2S CNRS–SUPELEC–UPS, L2S CNRS–SUPELEC–UPS, L2S CNRS–SUPELEC–UPS, HEUDIASYC CNRS–UTC, France France France France

An adaptive clustering method of the cross-sectional mean void fraction signals of gas-liquid two-phase ?ow . . . . . . . . 60 Katsuhiro Inoue Kyushu Inst. Technology, Japan Kousuke Kumamaru Kyushu Inst. Technology, Japan Kotohiko Sekoguchi Polytechnic Coll., Kagawa, Japan

Parameter identi?cation of a substitution model for a ?exible link . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Dirk Nissing Jan Polzer University of Duisburg, Germany University of Duisburg, Germany

Realisation of hierarchical look-up tables on low-cost hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Susanne T¨pfer. o Darmstadt University of Technology

ThMD5

Software Demonstration Session II (Matlab) Sierra Madre North
Invited Session Organizer: R. Schumann Fachhochschule Hannover, Germany Chairs: R. Schumann M. Flores Fachhochschule Hannover, Germany EngineSoft

Exponential ARX model based generalized predictive control for thermal power plants . . . . . . . . . . . . . . . . . . . . . . . . 61 H. Peng T. Ozaki Y. Toyoda K. Oda Institute of Statistical Mathematics, Institute of Statistical Mathematics, Bailey Japan Co. Ltd., Bailey Japan Co. Ltd., Japan Japan Japan Japan

Software demonstation session: 1.30 pm to 3.30 pm

Nonlinear system identi?cation of a closed-loop lean combustion process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 Sergio M. Savaresi Robert R. Bitmead Wayne J. Dunstan Politecnico di Milano University of California San Diego University of California San Diego

PIDtune: a graphical package for integrated system identi?cation and PID controller design . . . . . . . . . . . . . . . . 63 Melvin E. Flores Daniel E. Rivera EngineSoft Arizona State University

Identi?cation and analysis of standby power supply batteries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 Steven Schooling Control Systems Centre, UMIST, UK Peter Wellstead Control Systems Centre, UMIST, UK

Matlab-toolbox ICAI: identifying standardized process models for industrial control design . . . . . . . . . . . . . . . . . . . . . 63 Ste?en K¨rner o DaimlerChrysler, Germany Reimar Schumann Fachhochschule Hannover, Germany Birga Syska Fachhochschule Hannover, Germany

Identi?cation of viscoelastic materials . . . . . . . . . . . . . . . . . . 62 M. Mossberg L. Hillstrom T. S¨derstr¨m o o Uppsala University Uppsala University Uppsala University The ADAPTx software for automated multivariable system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Wallace E. Larimore Adaptics, Inc

Dynamic modeling of moisture in paper machine grade changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? P. Viitam¨ki a

Extension for the frequency domain system identi?cation toolbox: graphical user interface, objects, improved

23
numerical stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Istvan Kollar Budapest Univ. Tech. & Econ., Hungary Johan Schoukens Vrije Universiteit Brussel, Belgium Rik Pintelon Vrije Universiteit Brussel, Belgium Gyula Simon Budapest Univ. Tech. & Econ., Hungary Gyula Roman Budapest Univ. Tech. & Econ., Hungary 5.20 pm A class of output inter-sampling approaches to closed-loop identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 Lianming Sun Yasuhiro Miyake Hiromitsu Ohmori Akira Sano Keio Keio Keio Keio University University University University

System identi?cation toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? L. Ljung 5.40 pm Closed loop performance enhancement in H2 , avoiding system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 Sylvain Prudhomme Bernard Gimonet ONERA-DCSD ONERA-DCSD

Thursday afternoon:
4.00 pm to 6.00 pm

ThPM2 ThPM1

Recursive Identi?cation San Miguel
Chairs: B. Wahlberg P. Albertos

Santa Rosa

Closed-Loop Identi?cation
Chairs: H. Hjalmarsson J. Schoukens

Royal Institute of Technology, Sweden Vrije Universiteit Brussel, Belgium

Royal Institute of Technology, Sweden UPV Valencia, Spain

4.00 pm Probability density function of indirect non-parametric transfer function estimates for plants in closed-loop . . . . . 64 W. P. Heath University of Newcastle, Australia

4.00 pm Recursive identi?cation under scarce measurements . . . . . 66 Roberto Sanchis Pedro Albertos Univ. Jaume I. Castellon. Spain UPV Valencia Spain

4.20 pm Comparison of di?erent closed-loop parametrization schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 L. Keviczky Cs. Banyasz Hungarian Academy of Science Hungarian Academy of Science

4.20 pm Hierarchical identi?cation and its convergence for the transfer function matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 F. Ding J. B. Yang Y. M. Xu 4.40 pm Batch scheme recursive parameter estimation of gradually time-varying nonlinear systems . . . . . . . . . . . . . . . . . . . . . . . . . 67 Sequare Daniel-Berhe Univ. California, Los Angeles Heinz Unbehauen Ruhr-University Bochum, Germany

4.40 pm Closed-loop identi?cation with an unstable or nonminimum phase controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 Beno? Codrons Univ. catholique de Louvain, Belgium ?t Brian D.O. Anderson Australian National University Michel Gevers Univ. catholique de Louvain, Belgium

5.00 pm Direct closed loop identi?cation of reduced order controllers (input matching) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 I.D. Landau A.Karimi Laboratoire d’Automatique de Grenoble Sharif University, Teheran, Iran

24

5.00 pm A new recursive scheme of extended instrumental variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? K.R. Chernyshov 5.20 pm Optimal identi?cation of continuous systems and a new fast algorithm for on-line mode . . . . . . . . . . . . . . . . . . . . . . . . . 67 Witold Byrski Stanislaw Fuksa Univ. Mining and Metallurgy, Poland Univ. Mining and Metallurgy, Poland

5.20 pm Continuous-time identi?cation of nonlinear systems using radial basis function network model and genetic algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 Tomohiro Hachino Izuru Karube Yoshitaka Minari Hitoshi Takata Kagoshima University, Kyushu Inst. Technology, Kagoshima University, Kagoshima University, Japan Japan Japan Japan

5.40 pm A genetic algorithm approach to identi?cation of nonlinear polynomial models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 Zi-Jiang Yang Tadasuke Fujimoto Katsumi Kumamaru Kyushu Institute of Technology Kyushu Institute of Technology Kyushu Institute of Technology

5.40 pm Some implementation aspects of sliding window least squares algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 Qinghua Zhang IRISA-INRIA

ThPM4 ThPM3

Bases for Nonlinear System Identi?cation Santa Cruz
Chairs: K. Poolla D. Eisenman University of California, Berkeley, USA Jet Propulsion Laboratory, USA

Education and Training in System Identi?cation Sierra Madre South
Invited Session Organizers: R. Guidorzi Universit` di Bologna, Italy a D. Rivera Arizona State University, USA Chairs: R. Guidorzi D. Rivera Universit` di Bologna, Italy a Arizona State University, USA

4.00 pm Globally constrained local function approximations via hierarchical modeling; a framework for system modeling under partial information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 ¨ Henrik Ojelund Technical University of Denmark Payman Sadegh Technical University of Denmark 4.20 pm Nonlinear system identi?cation via wavelet expansions . . . ? M. Pawlak Z. Hasiewicz 4.40 pm Reconsideration of dead time measurement by wavelet from phase property of frequenct response . . . . . . . . . . . . . . . . . . . . 68 Tetsuya Tabaru Seiichi Shin University of Tokyo University of Tokyo

4.00 pm Towards a web-based study support environment to improve study e?ectiveness in teaching automatic control courses . ? G. Copinga M. Verhaegen M. Van den Ven 4.20 pm A modular approach in designing an environment for teaching system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . 69 Roberto P. Guidorzi Umberto Soverini Paolo Castaldi Roberto Diversi Universit` a Universit` a Universit` a Universit` a di di di di Bologna Bologna Bologna Bologna

5.00 pm Complex-valued minimal radial basis function neural network for nonlinear system identi?cation . . . . . . . . . . . . . 68 Deng Jianping N Sundararajan P Saratchandran Nanyang Technological University Nanyang Technological University Nanyang Technological University

4.40 pm Teaching system identi?cation to a variety of audiences . . ? L. Ljung

25

5.00 pm Beyond step testing and process reaction curves: introducing meaningful system identi?cation concepts in the undergraduate chemical engineering curriculum . . . . . 70 D.E. Rivera M.E. Flores Arizona State University Arizona State University

The CONTSID toolbox: a matlab toolbox for continuous-time system identi?cation . . . . . . . . . . . . . . . . . . . 71 H. Garnier M. Mensler Ctr. Recherche Auto. de Nancy, France Kyushu University, Japan

5.20 pm Learning system identi?cation using multimedia . . . . . . . . 70 Christian Schmid Ruhr-Universitaet Bochum, Germany 5.40 pm Should control be the only motivation, and MATLAB the only tool? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 E. Walter L2S CNRS–SUPELEC–UPS, France

CLOSID - a matlab toolbox for closed-loop system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Paul Van den Hof Raymond de Callafon Edwin van Donkelaar Delft Univ. Tech., Netherlands Univ. California at San Diego Delft Univ. Tech., Netherlands

FIT, ?ltering and identi?cation tool . . . . . . . . . . . . . . . . . . . . . ? O. Moseler M. Vogt

ThPM5

Software Demonstration Session III (Matlab) Sierra Madre North
Invited Session Organizer: R. Schumann Fachhochschule Hannover, Germany Chairs: R. Schumann O. Ravn Fachhochschule Hannover, Germany Technical University of Denmark

Friday morning:
10.00 am to 12.00 noon

Plenary Lecture 3
8.30 to 9.30 am

San Rafael

Software demonstation session: 4.00 pm to 6.00 pm

Frequency domain identi?cation . . . . . . . . . . . . . . . . . . . . . . . . 10 Tomas McKelvey Link¨ping Universitet, Sweden o The Adaptive Blockset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 Ole Ravn Technical University of Denmark FrAM1 Grid-based look-up table optimization toolbox . . . . . . . . . . . 71 Oliver Nelles Alexander Fink Darmstadt Univ. Tech., Germany Darmstadt Univ. Tech., Germany

Bounded-Error Parameter and State Estimation I Sierra Madre South
Invited Session Organizer: E. Walter L2S CNRS–SUPELEC–UPS, France Chairs: E. Walter E.-W. Bai L2S CNRS–SUPELEC–UPS, France University of Iowa, USA

Local linear model trees (LOLIMOT) toolbox for nonlinear system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 Oliver Nelles Alexander Fink Rolf Isermann Darmstadt Univ. Tech., Germany Darmstadt Univ. Tech., Germany Darmstadt Univ. Tech., Germany

10.00 am Set-membership binormalized data-reusing algorithms . . . 72 Paulo S. R. Diniz Stefan Werner Univ. Federal do Rio de Janeiro Helsinki University of Technology

26

10.20 am Convergence analysis of the quasi-OBE algorithm and performance implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 J.R. Deller, Jr S. Gollamudi S. Nagaraj Y.F. Huang Michigan State University, University of Notre Dame, University of Notre Dame, University of Notre Dame, USA USA USA USA

10.20 pm Spectral based parameter estimation in nonlinear stochastic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Ola Markusson Automatic Control (KTH) Sweden H? akan Hjalmarsson Automatic Control (KTH) Sweden 10.40 am Multi-objective identi?cation of FIR models . . . . . . . . . . . . 74 Tor A Johansen Norwegian Univ. Science & Tech.

10.40 am Variable gain parameter estimation algorithms for fast tracking and smooth steady state . . . . . . . . . . . . . . . . . . . . . . . . ? Er-Wei Bai Y-F. Huang 11.00 am Optimal set-membership ?ltering using an adaptive minimax algorithm with automatic bound tuning . . . . . . . . 73 S. Gollamudi S. Nagaraj Y. F. Huang University of Notre Dame University of Notre Dame University of Notre Dame

11.00 am On the estimation of optimal weights for instrumental variable system identi?cation methods . . . . . . . . . . . . . . . . . . 74 Petre Stoica Magnus Jansson Uppsala University Royal Inst. of Tech. (KTH)

11.20 am L2 model reduction and variance reduction . . . . . . . . . . . . . 74 Fredrik Tj¨rnstr¨m a o Lennart Ljung Link¨ping University o Link¨ping University o

11.20 am Self-tuning control by model unfalsi?cation . . . . . . . . . . . . . 73 S. M. Veres H. Xia University of Southampton University of Strathclyde

11.40 am Extended kalman ?ltering and weighted least squares dynamic identi?cation of robots . . . . . . . . . . . . . . . . . . . . . . . . 74 M. Gautier Ph. Poignet IRCyN, Univ. Nantes, France LVR, IUT Bourges, France

11.40 am Robust estimation for uncertain models in a data fusion scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Ali H. Sayed Tareq Y. Al-Na?ouri Thomas Kailath UCLA Stanford University Stanford University

FrAM3

Nonlinear System Identi?cation: I

Santa Rosa

Chairs: I. Mareels University of Melbourne, Australia A. Fink Darmstadt Universitat Technology, Germany

FrAM2

Parameter Estimation
Chairs: H. Hjalmarsson B. Wahlberg

San Miguel

Royal Institute of Technology, Sweden Royal Institute of Technology, Sweden

10.00 am Adaptive identi?cation of plants with nonlinear static gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 F. Giri F.Z. Chaoui Y. Rochdi LAP, ISMRA LAP, ISMRA LA2I, EMI

10.00 am Mis?t versus latency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Philippe Lemmerling Bart De Moor Katholieke Universiteit Leuven Katholieke Universiteit Leuven

10.20 am A nonlinear black box identi?cation procedure; illustrated on a distillation column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? F. Declerq R. De Keyser

27

10.40 am A nonlinear grey-box example using a stepwise system identi?cation approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 Jonas Sjoberg Chalmers University Sweden

10.40 am Reduced order models for di?usion systems via collocation methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Bharath Bhikkaji Torsten S¨derstr¨m o o Uppsala University Uppsala University.

11.00 am Non-linear system identi?cation using Volterra series expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 Hassouna Samira Coirault Patrick Poinot Thierry LAII-ESIP, France LAII-ESIP, France LAII-ESIP, France

11.00 am Parameter estimation of fractional systems: application to the modeling of a lead-acid battery . . . . . . . . . . . . . . . . . . . . . 77 Jun Lin Thierry Poinot Jean-Claude Trigeassou Rgis Ouvrard LAII, LAII, LAII, LAII, France France France France

11.20 am Nonparametric identi?cation of generalized Hammerstein models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 M. Pawlak R.K. Pearson B.A. Ogunnaike F.J. Doyle III University of Manitoba, Canada ETH Zuerich, Switzerland DuPont Expt. Station, Delaware University of Delaware, Wilmington

11.20 am Non integer model from modal decomposition for time domain system identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Olivier Cois Univ. Bordeaux I/ENSERB, France Alain, Oustaloup Univ. Bordeaux I/ENSERB, France Eric, Battaglia Univ. Bordeaux I/ENSERB, France Jean-Luc, Battaglia LEPT, ENSAM, France

11.40 am Quantifying the accuracy of Hammerstein model estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Brett Ninness Stuart Gibson University of Newcastle University of Newcastle

Friday midday:
1.30 pm to 3.30 pm

FrAM4

Distributed Parameter Systems
Chairs: T. S¨derstr¨m o o L. Autrique

Santa Cruz

FrMD1

Uppsala Universitet, Sweden IMP-CNRS, Univ. Perpignan, France

Bounded-Error Parameter and State Estimation II Sierra Madre South
Invited Session Organizer: E. Walter L2S CNRS–SUPELEC–UPS, France Chairs: E. Walter J. Norton L2S CNRS–SUPELEC–UPS, France University of Birmingham, UK

10.00 am Parameter estimation and model approximation for di?usion models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Torsten S¨derstr¨m o o Susanne Remle Uppsala University, Sweden Uppsala University, Sweden

10.20 am Optimal sensor strategy for parametric identi?cation of a thermal system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 L Autrique C Chaussavoine JP Leyris A Ferriere IMP-CNRS Univ. Perpignan, France IMP-CNRS Univ. Perpignan, France IMP-CNRS Univ. Perpignan, France IMP-CNRS, France

1.30 pm Updating of bounding ellipsoids cut simultaneously by two non-parallel linear bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 P.F.Weston J.P.Norton U. of Birmingham, UK U. of Birmingham, UK

1.50 pm Some signal-processing applications of set solutions . . . . . 78 D. Joachim J.R. Deller, Jr Sanders, A Lockheed-Martin Co., USA Michigan State University, USA

28

2.10 pm Set-membership estimation with the trace criterion made simpler than with the determinant criterion . . . . . . . . . . . . 78 C. Durieu E. Walter B. Polyak LESiR CNRS–ENS Cachan France L2S CNRS–SUPELEC–UPS France Institute of Control Science, Russia

2.10 pm Restricted structure linear estimators for multiple mdoel and adaptive systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 M.J. Grimble University of Strathclyde

2.30 pm Bounds on trajectories in systems with parametric and observation perturbations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? D. Rokiyanskii A. Kinev F. Chernousko 2.50 pm Constructing and updating navigation maps with uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 Mauro Di Marco Andrea Garulli Domenico Prattichizzo Antonio Vicino University University University University of of of of Siena, Siena, Siena, Siena, Italy Italy Italy Italy

2.30 pm Design and application of digital FIR di?erentiators using modulating functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Armin Wolfram Olaf Moseler Darmstadt University of Technology Darmstadt University of Technology

2.50 pm Concurrent rotor time constant and ?ux estimation of IM using nonlinear observers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Juan Carlos G?mez Univ. Nacional Rosario, Argentina o Trist?n P?rez a e University of Newcastle, Australia

FrMD3

Nonlinear System Identi?cation: II
Chairs: I. Mareels B. David

Santa Rosa

3.10 pm Dynamic programming algorithms for guaranteed state estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? A. Kurzhanski T. Fiippova

University of Melbourne, Australia Universit? catholique de Louvain, Belgium e

1.30 pm Modelling non-linear process dynamics using partially recurrent neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 A. Hamid Adom J. Barry Gomm David Williams Liverpool John Moores Univ., UK Liverpool John Moores Univ., UK Liverpool John Moores Univ., UK

FrMD2

Estimation and Filtering
Chairs: P. Broersen M. Grimble

San Miguel

Delft Univ. Technology, The Netherlands University of Strathclyde, Scotland, UK

1.50 pm Quasi-ARMAX modeling approaches to identi?cation and prediction of nonlinear systems . . . . . . . . . . . . . . . . . . . . . . . . . 80 Jinglu Hu Kousuke Kumamaru Kotaro Hirasawa Kyushu University, Japan Kyushu Inst. Technology, Japan Kyushu University, Japan

1.30 pm Improving the e?ciency of reduced statistics ARMA estimators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 P.M.T. Broersen S. de Waele Delft Univ. of Technology Delft Univ. of Technology

2.10 pm Robust real-time identi?cation based on an optimal residuals nonlinear transformation . . . . . . . . . . . . . . . . . . . . . 81 D. Wang G. W. Barton J. A. Romagnoli University of Sydney University of Sydney University of Sydney

1.50 pm A semide?nite programming approach to ARMA estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 T. McKelvey P. Stoica J. Mari Link¨ping Univeristy, Sweden o Uppsala University, Sweden ADtranz, Sweden

29

2.30 pm Estimating and testing an exponential-a?ne term structure model by nonlinear ?ltering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 Mikkel Baadsgaard Jan Nygaard Nielsen Henrik Madsen Tech. Univ. of Denmark Tech. Univ. of Denmark Tech. Univ. of Denmark

2.30 pm Sequences with uniformly decaying auto-correlation coe?cients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 S. R. Venkatesh M. A. Dahleh MIT MIT

2.50 pm Estimating physical parameters of nonlinear systems using bond graph models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 Peter J Gawthrop University of Glasgow

2.50 pm Set-theoretic input sequence design for orthonormal model identi?cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 R.K. Pearson P.H. Menold F.J. Kraus ETH Zuerich, Switzerland Universit¨t Stuttgart, Germany a ETH Zuerich, Switzerland

3.10 pm Parameter estimation in nonlinear systems with auto and crosscorrelated noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 B. David G. Bastin Univ. catholique de Louvain, Belgium Univ. catholique de Louvain, Belgium

Friday afternoon:
4.00 pm to 6.00 pm

FrMD4

FrPM1

Set Membership Identi?cationSierra Madre South Input Sequences in Linear and Nonlinear Identi?cation Santa Cruz Chairs:
Invited Session Organizers: F. Allgower Universit¨t Stuttgart, Germany a D. Rivera Arizona State University, USA Chairs: F. Allgower D. Rivera Universit¨t Stuttgart, Germany a Arizona State University, USA E. Walter M. Milanese L2S CNRS–SUPELEC–UPS, France Politecnico di Torino, Italy

4.00 pm A probabilistic approach to model set identi?cation . . . . . .83 Tomoki Miyazato Shinji Hara Tokyo University of Technology Tokyo Institute of Technology

1.30 pm Optimal input desgin using linear matrix inequalities . . . 82 Kristian Lindqvist H? akan Hjalmarsson Royal Inst. Tech., Sweden Royal Inst. Tech., Sweden

4.20 pm Worst-case ?1 identi?cation based on correlation analysis83 Hiroaki Fukushima Toshiharu Sugie Kyoto University Kyoto University

1.50 pm On consistency and model validation for systems with parameter uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 S. Gugercin A. C. Antoulas Rice University Rice University

4.40 pm Improved bounding of transfer-function parameters by use of prior bounds on time-domain behaviour . . . . . . . . . . . . . . 84 H.Messaoud J.P.Norton Ecole Nat. d’Ing. de Monastir, Tunisia U. of Birmingham, UK

2.10 pm Design of minimum crest factor multisinusoidal signals for plant-friendly identi?cation of nonlinear process systems 82 Martin W. Braun Ra?l Ortiz-Mojica u Daniel E. Rivera Arizona State University Arizona State University Arizona State University

5.00 pm Nonparametric identi?cation of multivariate linear uncertain-stochastic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 A.R. Pankov Inst. Information Transmisson Prob. E.N. Platonov Inst. Information Transmisson Prob. K.V. Siemenikhin Inst. Information Transmisson Prob.

30

5.20 pm An arti?cial neural network implementation of the second Fogel-Huang algorithm for parameter set estimation . . . . 84 X.-F. Sun Y.-Z. Fan Beijing Univ Aero & Astro, P.R.China Beijing Univ Aero & Astro, P.R.China

FrPM3

Nonlinear System Identi?cation: III Santa Rosa
Chairs: B. Codrons K. Poolla Universit? catholique de Louvain, Belgium e University of California, Berkeley, USA

5.40 pm Computing uncertainty regions with simultaneous con?dence degree using bootstrap . . . . . . . . . . . . . . . . . . . . . . . 84 Fredrik Tj¨rnstr¨m a o Link¨ping University o

4.00 pm Frequency response function measurements in the presence of nonlinear distortions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 J. Schoukens R. Pintelon Y. Rolain T. Dobrowiecki VUB VUB VUB Budapest Univ. Tech. & Economics

FrPM2

Stochastic Estimation and Control

San Miguel

Chairs: K. Kumamaru Kyushu Institute of Technology, Japan H. Madsen Tech. University of Denmark, Denmark 4.00 pm Estimating functions for discretely observed di?usions with measurement noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 Jan Nygaard Nielsen Kim Nols?e Henrik Madsen Tech. Univ. of Denmark Tech. Univ. of Denmark Tech. Univ. of Denmark

4.20 pm Second-order vs. ?rst-order algorithms: new perspectives . ? H. Hyotyniemi 4.40 pm Nonlinear system identi?cation with multilevel perturbation signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 H. A. Barker K. R. Godfrey A. J. Tucker University of Wales Swansea, U.K University of Warwick, U.K University of Warwick, U.K.

4.20 pm Stochastic approximation type nonlinear ?lter with randomly varying truncations . . . . . . . . . . . . . . . . . . . . . . . . . . 85 Katsuji Uosaki Tottori Univ., Japan Han-Fu Chen Chinese Acad. Science, P.R.China Toshiharu Hatanaka Tottori Univ., Japan 4.40 pm Application of multiple tree stochastic theory on estimating signal over network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? J. Sembiring K. Akizuki 5.00 pm Identi?cation of stochastic N-L-N systems using monotone correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ? K.R. Chernyshov 5.20 pm Filtering, prediction, and smoothing with Gaussian sum representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 Miroslav Simandl Univ. West Bohemia, Czech Republic Jakub Kralovec Univ. West Bohemia, Czech Republic

5.00 pm On the prediction error of regularized models . . . . . . . . . . . 86 Andreas Poncet ABB Corporate Research, Switzerland James L. Massey ETH Z¨rich u 5.20 pm Precursors of bifurcations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Arthur J. Krener University of California, Davis

5.40 pm NLMIMO, non-linear multi-imput-output toolbox . . . . . . . . ? E. Bertolissi M. Birattari A. Duchateau H. Bersini

31
FrPM4

Experiment design
Chairs: G. Burdick M. Braun

Santa Cruz 5.00 pm
Jet Propulsion Laboratory, USA Arizona State University, USA Steady state optimal test signal design for constrained multivariable systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Qiang Zhan Christos Georgakis Lehigh University Lehigh University

4.00 pm A two-step approach to control-relevant design of test input signals for iterative system identi?cation . . . . . . . . . . . . . . . 87 Yudi Samyudia Jay H. Lee Delft University of Technology Purdue University

5.20 pm Accuracy of autotuning identi?cation methods and achievable closed loop performance . . . . . . . . . . . . . . . . . . . . . 88 Claudio Scali Gabriele Marchetti Daniele Semino University of Pisa, Italy University of Pisa, Italy University of Pisa, Italy

4.20 pm On closed loop identi?cation for PI controllers iterative tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 P?ricles Rezende Barros Univ. Fed. da Para? e ?ba, Brazil Gustavo Henrique Machado de Arruda Univ. Fed. da Para? ?ba, Brazil 4.40 pm Quality criteria and design of biased identi?cation experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Tomasz Twardowski AGH, Measurement Div., Poland

5.40 pm Design of identi?cation experiments for step response models: open loop case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Dianne Smektala Honeywell Hi-Spec Solutions, Canada William Cluett University of Toronto, Canada Liuping Wang University of Newcastle, Australia

32

33

SYSID 2000 Abstracts
Wednesday morning:
10.00 am to 12.00 noon

Plenary Lecture 1
8.30 to 9.30 am

San Rafael Set membership identi?cation for H∞ robust control design
Mario Milanese Michele Taragna Politecnico di Torino, Italy Politecnico di Torino, Italy

11.00 am

System identi?cation in modern ?nance David G. Luenberger Stanford University, USA

Abstract:
WeAM1

Identi?cation for Robust Control Sierra Madre South
Invited Session Organizers: M. Gevers CESAME, Louvain-la-Neuve, Belgium M. Milanese Politecnico di Torino, Italy Chairs: M. Gevers M. Milanese CESAME, Louvain-la-Neuve, Belgium Politecnico di Torino, Italy

10.00 am Model validation for robust control: experiment design issues M. Gevers X. Bombois B. Codrons F. De Bruyne G. Scorletti

The Set Membership H-in?nity identi?cation of LTI discrete-time SISO systems from noise corrupted measurements is considered. A test is given for validating the assumptions on system and noise. The main focus of the paper is on the trade-o? between identi?ed model set complexity and optimality properties, and on the non-conservativeness in the identi?cation error evaluation. By using a weighted H-in?nity norm, the identi?cation error measures the degradation of the closed-loop performance that is underlying the control design, due to the mismatch between the actual plant and the model. Thus, optimality properties of the identi?ed model set re?ect in minimality of conservativeness of guaranteed closed-loop performances. Identi?cation of the optimal model set is a NP-hard problem and alternative algorithms are investigated, simpler to be computed, at the expense of identi?cation accuracy degradation. This is measured by the suboptimality level of the identi?ed model set, i.e. the ratio between the achieved identi?cation error and the minimal one. A method is given to evaluate upper and lower bounds on this suboptimality level. Reduced order model sets that tightly include the optimal one are derived.

10.20 am Uncertainty model falsi?cation R. Kosut

11.20 am Closed-loop model validation using coprime factor uncertainty models Raymond A. de Callafon Univ. California, San Diego Paul M.J. Van den Hof Delft University of Technology

Abstract:
10.40 am Model error modeling and control design L. Ljung

Model (in)validation techniques are used to bridge the gap between models used in robust control synthesis and uncertainty models obtained from identi?cation experiments. In most applications the aim is to design a robust controller and therefore it is valuable to

34
validate or invalidate an uncertainty model in view of this application by considering a closed-loop model validation technique. In this paper a model validation approach is presented that generalizes the (in)validation of possibly unstable models on the basis of closed-loop experiments with a stabilizing, but possibly unstable, controller. The approach is presented in a robust control framework with an uncertainty model described with coprime factor perturbations. It is shown that this approach yields an a?ne expression of the uncertainty model in all possible transfer functions that can be measured via a closed-loop experiments, which facilitates the optimization involved with a model invalidation.

10.20 am On the dynamics of local linear model networks with orthonormal basis functions Oliver Nelles Masayoshi Tomizuka UC Berkeley, ME Dept UC Berkeley, ME Dept.

Abstract:

11.40am Panel discussion

WeAM2

Neural Networks
Chairs: S. Toepfer J.B. Gomm

San Miguel

This paper deals with local linear model networks for nonlinear system identi?cation. It compares the standard nonlinear autoregressive with exogenous input (NARX) model structure with the new nonlinear orthonormal basis functions (NOBF) model structure. In particular, the dynamics of interpolated local ARX and OBF models are studied and signi?cant advantages for the NOBF approach are pointed out. Furthermore, the dynamic e?ect of scheduling variables in local linear model networks based on OBFs is investigated. It is concluded that the NOBF approach is a promising alternative for nonlinear system identi?cation and deserves more attention in the future.

Tech. Universitat Darmstadt, Germany Liverpool John Moores University, UK

10.40 am Backward centre selection method for on-line adaptation of RBF network models J. Barry Gomm Ding Li Yu Liverpool John Moores Univ., UK Liverpool John Moores Univ., UK

10.00 am Studies on initialization for multilayer networks Hiroshi Shiratsuchi Hiromu Gotanda Katuhiro Inoue Kousuke Kumamaru Univ. of the Ryukyus Kinki Univ. in Kyushu Kyushu Institute of Technology Kyushu Institute of Technology

Abstract:

Abstract:

This paper proposes an initialization of back propagation (BP) networks with one hidden layer trained for pattern classi?cation problems: weights of hidden units are initialized so that the hyperplanes should pass through the center of input pattern set, and those of output units are initialized to zero. At the same time, we suggest that if there exist two or more hidden layers, their initial bias weights except for the ?rst hidden layer are given by zero so that the hyperplanes should pass through the center of input domain. From simulation results for MONK’s problem, Iris classi?cation problem and Sonar target identi?cation problem, it is veri?ed that the proposed initialization is e?ective.

An approach to tracking process time variations using an adaptive radial basis function network model is described. The method is based on a numerically robust recursive algorithm for updating the network output layer weights. It is shown how network centres contributing least to the network output can be found and removed from model calculations. Thus, both the structure and weights of the network are adaptive. An illustrative example is given, to demonstrate the e?ectiveness of the algorithm and illustrate its performance, in an application to modelling a real chemical process. Results show more accurate model predictions compared to using a network with only weight updating.

35
11.00 am Comparison of a Hierarchically Constructed Neural Network and a Hierarchical Look-up Table Susanne T¨pfer o Oliver Nelles Rolf Isermann. Darmstadt University of Technology University of California at Berkeley Darmstadt University of Technology presented where biological neural networks can be e?ciently built and simulated, at the aim to study spatio-temporal organization arising from populations of chaotic neurons. Some models of Central Pattern Generators are also presented with application to the generation of insect walking patterns, implemented on a robotic structure directly driven by the simulator.

Abstract:

Static and dynamic nonlinear models of real processes are required in many disciplines, i.e. for system analysis, controller design, model-based control, process optimization, supervision and fault diagnosis. According to the considered application, models have to predict or simulate nonlinear characteristics of processes or control strategies in o?-line or real-time (on-line) operation. Two black-box models, LOLIMOT and hierarchical look-up tables, are presented which are well suited to deal with identi?cation of complex nonlinear systems. LOLIMOT is a local linear neuro-fuzzy model which preferably can be used for o?-line simulation. On the other hand hierarchical look-up tables represent an extension of classical look-up tables and allow the on-line simulation of the identi?ed nonlinear systems under time-critical conditions. Even though both models are based on completely di?erent approximation principles they share a recursive partitioning approach for solving the structure identi?cation problem. Both hierarchical construction algorithms lead to recursive partitioning of the input space in accordance with the complexity of the underlying nonlinear process. Because of this inherent structure identi?cation feature the curse of dimensionality can be avoided. Advantages of the divide-and-conquer principle are especially short training times and parsimonious models.

WeAM3

Subspace Methods I
Chairs: W. Larimore T. McKelvey

Santa Rosa
Adaptics Inc., USA Link¨ping Universitet, Sweden o

10.00 am Error analysis of certain subspace methods Chiuso Alessandro Picci Giorgio Univ. di Padova Univ. di Padova

Abstract:

Some subspace methods make use of an oblique projection to determine the observability matrix and the system matrices (A, B, C, D). There is experimental evidence (Chiuso and Picci 1999, Kawauchi et al. 1999) that these methods perform poorly in certain situations. In this paper we present an error analysis of oblique projections and show that in some situations small errors in the subspace along which the projection is done may lead to large errors in the estimate. This may happen for instance in the N4SID algorithm. Asymptotically optimal weighting strategies for the estimation of the observability matrix have been derived in the literature (see Gustafsson) which however make use of information on the TRUE system. It seems desirable (at least in a ?rst stage) to choose system-independent weighting matrices. The orthogonal projection approach (MOESP, CVA) has been found to be more robust in simulations as well as asymptotically suboptimal (see Gustafsson). We shall provide further evidence that sometimes the orthogonal projection weighting (MOESP) should be recommended; in fact, it can be shown that the worst-case ”signal to noise” ratio on the rows of the estimated observability matrix is larger for orthogonal projections than for oblique projections. A variance analysis is presented and simulations are included which compare oblique projection based algorithms with the orthogonal projection.

11.20 am An interpolation technique for learning with sparse data J. Van Gorp Y. Rolain

11.40 am Simulation of central pattern generators by using populations of chaotic neurons P. Arena L. Fortuna D. Garofalo University of Catania, DEES University of Catania DEES University of Catania DEES

Abstract:

In this paper a simulation environment is

36
numerically. On the other hand it is also interesting from a theoretical point of view, as it adds to the knowledge of the e?ects of the choice of the weighting matrices, showing that several choices lead to the same asymptotic distributions of the estimates. It is also shown, that the distribution of the noise does not in?uence the asymptotic variance. Keywords: subspace methods, asymptotic properties, linear systems, state space models References Bauer, D., M. Deistler and W. Scherrer (1999). Consistency and asymptotic normality of some subspace algorithms for systems without observed inputs. Automatica 35, 1243-1254.

10.20 am Asymptotic variance analysis of subspace identi?cation methods Magnus Jansson Royal Inst. of Tech. (KTH)

Abstract:

The class of subspace algorithms for system identi?cation is an interesting complement to the maximum likelihood or prediction error methods, especially for multivariable systems. The statistical analysis of the subspace methods is di?cult since the estimates depend on the data in a rather complicated manner. Previous results include proofs of generic consistency and asymptotic normality of the estimates. However, no explicit transparent expression for the covariance matrix of the limiting distribution has so far been reported because of the aforementioned di?culties. The main objective of this paper is to provide a methodology that simpli?es the asymptotic analysis of subspace based estimation algorithms. The basic idea is illustrated by deriving the asymptotic covariance matrix corresponding to the estimates of the state-space matrices (or the transfer function estimate) of a quite general subspace algorithm.

11.00 am Weighting in subspace-based system identi?cation Tony Gustafsson University of California San Diego

Abstract:

10.40 am On the impact of weighting matrices in subspace algorithms Dietmar Bauer Manfred Deistler Wolfgang Scherrer TU Wien, Austria TU Wien, Austria TU Wien, Austria

Subspace-based methods for system identi?cation are often based on an estimate of the range space of the extended observability matrix. It is thus of great interest to investigate, and also optimize, the accuracy of the estimated subspace. Especially, the in?uence of certain weighting matrices is an unresolved issue. Here, an asymptotic analysis of the accuracy of the estimated subspace is presented. The main result of the analysis is that statistically sound weighting matrices are found.

Abstract:

Subspace algorithms are used for the estimation of linear, time invariant, discrete time, state space systems. Due to their numerical properties they are an alternative to the more classical maximum likelihood or prediction error methods. In this paper some results concerning the asymptotic distribution of the estimates of the transfer function obtained by using subspace methods in the case of no exogenous inputs are given. The discussion centers on the estimation of the transfer function rather than on the estimation of the system matrices, since the transfer function is a system invariant. Therefore the accuracy of the transfer function estimate can be directly used to evaluate the relative e?ciency of di?erent methods. The present paper uses the techniques and the asymptotic expressions derived in Bauer et al. (1999). In particular the e?ects of certain weighting matrices used in the algorithms are investigated. This on the one hand leads to a signi?cant decrease in the number of calculations needed to obtain the asymptotic variance

11.20 am Optimal weighting for MOESP type of procedures Dietmar Bauer Tony Gustafsson TU Wien, Austria University of California San Diego

Abstract:

The asymptotic distribution for the MOESP type of subspace methods has been investigated recently in a number of papers. Consistency and asymptotic normality have been derived, however not much knowledge about the e?ect of the user choices on the asymptotic accuracy has been gained. In this paper a new instrumental variables interpretation of these algorithms is given. This leads to a lower bound on the achievable estimation accuracy, which is di?erent from the Cram?r e Rao lower bound, and in one example shown in the paper this lower bound is higher than the Cram?r Rao bound, e

37
and thus adds something to the knowledge, i.e. in such a situation one can show, that irrespective on how one of the weighting matrices is chosen, there will always be a gap to the Cram?r Rao bound and thus in this case the MOESP e type of subspace prodcedure is suboptimal. However the lower bound sometimes is lower than the CR bound, showing that the lower bound cannot be obtained for any weighting scheme. The derived lower bound depends on the choice of the row weighting. This is in accordance to the fact, that the choice of a certain row weighting does in?uence the asymptotic accuracy of the system matrix estimates, although it does not a?ect the accuracy of the estimated poles. Some facts about the asymptotic covariance matrix are derived. Keywords: Subspace Methods, Asymptotic Properties, Identi?cation, Linear Systems, State Space Models, Discrete-time Systems K. Keesman Wageningen Univ., The Netherlands

10.00 am State and parameter estimation in biotechnical batch reactors Karel J. Keesman Wageningen Univ., The Netherlands

Abstract:

11.40 am Subspace algorithms for data from varying sensor locations. Albert Benveniste Mich`le Basseville e Maurice Goursat Laurent Mevel IRISA/INRIA, France IRISA/CNRS, France INRIA, France IRISA/INRIA, France

In this paper the problem of state and parameter estimation in biotechnical batch reactors is considered. Models describing the biotechnical process behaviour are usually non-linear with time-varying parameters. Hence, the resulting large dimensions of the augmented state vector, roughly ? 7, in recursive estimation is a crucial problem. However, by decomposition techniques on the basis of singular perturbation analysis or batch phase analysis one is able to reduce the dimension of the augmented state vector. Furthermore, prior knowledge of parameters and initial states is essential. It is therefore shown how these initial values can be e?ectively obtained from the data. The approach will be demonstrated by two examples, a wastewater sludge treatment and a beer fermentation process, using real data.

Abstract:

We investigate the use of subspace identi?cation algorithms for output-only identi?cation of the eigenstructure of a linear MIMO system. We focus on the following situation, typical in vibration-based structural analysis. Several successive data sets are recorded, with sensors at di?erent locations in the structure. Some of the sensors, called the reference sensors, are kept ?xed, while the other ones are moved for the di?erent records. This emulates a situation with hundreds of sensors available, while in fact only, say, ten are actually at hand. One additional di?culty is that the input, besides being not observed, is turbulent in nature and nonstationary. The purpose of this paper is to show how subspace methods can be adapted to such a situation.

10.20 am Parameter estimation that improves validity of model for a certain model use Mario Zec Nadja Hvala Stanko Strmcnik J. Stefan Institute, Slovenia J. Stefan Institute, Slovenia J. Stefan Institute, Slovenia

Abstract:

WeAM4

Biological Applications
Chairs: E. Walter

Santa Cruz

L2S CNRS–SUPELEC–UPS, France

The paper considers the problem of parameter estimation in the case that model parameters are not uniquely identi?able from plant measurements. In such a case, biased estimates of parameters values are obtained, which a?ects the results when the model is used for a certain purpose. If uncertainty of parameter estimates is taken into account, also the model results spread over a certain region instead of being a single value. The aim of simulation study presented in the paper is to ?nd out whether the estimation of model parameters can be improved in such a way that small enough range of model results is obtained. The results of the study indicate that from plant measurements available for the estimation of model parameters, it is possible to extract data that is important for the estimation of model parameters relative to a certain model use. If this data is improved by a proper measurement campaign (e.g. proper choice of measured variables, better accuracy, higher measurement frequency) it is to expect that a valid model for a certain

38
model use is obtained. Simulation study is performed for a simple activated sludge model from wastewater treatment, while the estimation of model parameters is done by Monte Carlo simulation. inactivation model is discussed. Traditionally, the parameters are obtained by applying (linear or nonlinear) regression procedures on data which are collected in a time-consuming and labor-intensive series of inactivation experiments at constant temperatures (so-called static experiments). Opposed to the static approach, in this contribution, the temperature is considered as a time-varying (dynamic) control input for identi?cation experiments. The temperature-time pro?le is optimized with respect to a Fisher information matrix based objective function. During this design on simulation level, parameter values estimated from preliminary experiments are considered as the nominal values. The optimal dynamic temperature pro?le obtained as such has been implemented in practice. The quality of the resulting parameter estimates is assessed by the construction of the joint con?dence region and by evaluation of the Fisher information matrix elements.

10.40 am Recursive identi?cation of activated sludge wastewater treatment plants Oscar A. Z. Sotomayor Song W. Park Claudio Garcia LSCP-EPUSP, Brazil LSCP-EPUSP, Brazil PTC-EPUSP, Brazil

Abstract:

The activated sludge wastewater treatment plants are complex systems, incorporating a large number of biological, physical-chemical and biochemical processes that are di?cult to supervise and control. The activated sludge processes are characterized by sti? dynamics, nonlinearities, time-varying, multivariable with many cross-couplings, large perturbations in ?ow and load, uncertainties concerning the composition of the incoming wastewater, a low level of stability punctuated by abrupt failures and a sensitive community of microorganisms. With the growing complexity of these processes, there is now a crescent need for analytical tools, which can examine its dynamics. This was highlighted in a report published by the European Commission Program COST Action 682 Integrated Wastewater Management, that indicated ”Identi?cation of the Dynamics Processes in WWTP” as an important topic for future research. This paper demonstrates the application, through simulations, of the continuous-discrete extended Kalman ?lter in the iterative adjustments of unknown parameters of reduced order models corresponding to the nitrate and dissolved oxygen concentration dynamics in activated sludge processes. The obtained models are applied to compare the process response and show a good ?tting to real data. They could be used to implement advanced control strategies.

11.20 am System identi?cation of electronic nose data for monitoring cyanobacteria in potable water: black-box modelling G.E. Searle J.W. Gardner M.J. Chappell K.R. Godfrey M.J. Chapman University of Warwick, University of Warwick, University of Warwick, University of Warwick, University of Coventry, UK UK UK UK UK

Abstract:

11.00 am Practical implementation of the optimal experiment design methodology for estimation of microbial heat resistance parameters K.J. Versyck K. Bernaerts J.F. Van Impe BioTeC - KULeuven, Belgium BioTeC - KULeuven, Belgium BioTeC - KULeuven, Belgium

Linear black-box modelling techniques are applied to data collected from an electronic nose experiment. FIR, ARX, MAX and ARMAX inverse models of the nose system are used to discriminate between two di?erent strains of cyanobacteria (blue-green algae), one toxic, the other not. Models for both pre-processed (static) and raw (dynamic) data are proposed, and successful classi?cation rates of 100% and 99.3% are obtained, respectively. These results are compared favourably with similar results obtained elsewhere using neural networks on the same data. The reasons for, and e?ects of, normalisation of the data are considered. The e?ects of the inclusion of additional model inputs (namely, the temperature and humidity in the sensor chamber) are also investigated.

11.40 am Metabolism and cell cycle modelling by means of neural network based structures Andreia Hanomolo Philippe Bogaerts Raymond Hanus University Libre de Bruxelles University Libre de Bruxelles University Libre de Bruxelles

Abstract:

In this paper, the design and the implementation of an experiment aimed at optimal estimation of the microbial kinetic parameters in a thermal

39 Abstract:
Due to their importance to the pharmaceutical industry and implicitly to the health, the animal cell cultures and the research carried out over their behaviour have gained recently an increasing attention. The paper presents a case study concerning the application of the neural networks in the modelling of hte main parameters of an animal cell culture: the metabolic components and the cell cycle phases. For the metabolism a hybrid (neural-classical) structure is described for the building of a continuous simulator capable to reconstruct the trajectory of the main components from the initial conditions. A special attention is paid to the choice of the cost function and to the ”amount” of the classical contribution to the hybrid model. For the cell cycle the outputs of the above continuous simulator are used and a neural network black box approach is employed in order to obtain the evolution of the three phases. The structures were tested on two batch animal cell cultures for which few and asynchronous measurements are available, namely one culture growing on microcarriers, the other in suspension. method that estimates the frequency response of the model error by an automatic procedure. A bene?t with this approach is that the tuning can be done locally, i.e., that di?erent resolutions can be used in di?erent frequency bands. The ideas are based on local polynomial regression and utilize a statistical criterion for selecting the optimal resolution.

1.50 pm Model error model from identi?cation in closed-loop A. Besan?on-Voda c

2.10 pm On model error modeling in set membership identi?cation A. Garulli W. Reinelt Universit` di Siena a Link¨ping University o

Wednesday midday:
1.30 pm to 3.30 pm

Abstract:

WeMD1

Model Error Modeling and Model Validation San Miguel
Chairs: F. Hadaegh A. Garulli Jet Propulsion Laboratory, USA Universit` di Siena, Italy a

A recent perspective on model error modeling is applied to set membership identi?cation techniques in order to highlight the separation between unmodeled dynamics and noise. Model validation issues are also easily addressed in the proposed framework. The computation of the minimum noise bound for which a nominal model is not falsi?ed by i/o data, can be used as a rationale for selecting an appropriate model class. Uncertainty is evaluated in terms of the frequency response, so that it can be handled by H-in?nity control techniques.

1.30 pm A nonparametric approach to model error modeling Anders Stenman Fredrik Tj¨rnstr¨m a o Link¨pings universitet, Sweden o Link¨pings universitet, Sweden o

2.30 pm Preliminary test for nonlinear input output relations in SISO systems Torben Knudsen Aalborg University, Denmark

Abstract:

Abstract:

To validate an estimated model and evaluate its reliability is an important part of the system identi?cation process. Recent work on model validation has shown that the use of explicit model error models provide a better way of visualizing the possible de?ciencies of the nominal model. Previous contributions have mainly focused on parametric black-box models for estimating the error model. However, this requires that a correct model order for the error model has to be selected. Here we suggest an adaptive and nonparametric frequency-domain

Deriving nonlinear models from experimental input/output data are much more complicated than deriving linear models. Consequently, if there is doubt about the existence or signi?cance of nonlinear e?ects a good test for linearity would be very useful. It should of cause be a preliminary test i.e. applicable before candidate models are derived. Existing tests of this kind are based on the Subba-Rao method for time series and therefore they do not take advantage of the properties for linear systems with known

40
input signal. Further the existing methods su?ers from lack of robustness and do not have well de?ned statistical properties. This paper therefore develops two new test based on the basic properties for linear systems namely superposition and the sine input/output relation. Simulation analysis shows that the existing method is only reliable for additive white Gaussian noise but for e.g. colored Gaussian noise it is absolutely unreliable as false alarm probabilities can be as high as 50% when they should be 5%. The new methods turned out to be superior to the existing method. Especially the sine method whish is reliable also for colored, heavy tailed or skew distributed noise is a signi?cant improvement to the existing method.

Abstract:

This paper describes a new method for quantifying errors in continuous time system identi?cation. The purpose is to estimate a global model constituted of a nominal model and a model of bias error. The nominal model is a transfer function that represents the slow modes and the steady state of the system; i.e. it represents the system behavior in low frequencies. The bias error model represents the rapid modes unmodelled by the nominal model. In steady state, the bias error model has a null value, i.e. a behavior essentially situated in high frequencies. The key idea is that the bias error model is orthogonal basis model using particular band-pass or high-pass orthogonal functions. The bias error model gives an information about the true bias error, and provides, for example, a frequency-domain bound. An application to a chemical reaction column is used to illustrate this new approach.

2.50 pm Exact bounds for the frequency response of an uncertain plant with ellipsoidal perturbations G. A. A. A. Chesi Garulli Tesi Vicino Universit` di Siena a Universit` di Siena a Universit` di Firenze a Universit` di Siena a

WeMD2

Continuous Time Methods
Chairs: T. S¨derstr¨m o o P. Gawthrop

Sierra Madre South

Abstract:

This paper deals with the frequency domain properties of an ellipsoidal family of rational functions, i.e. a family of rational functions whose coe?cients depend a?nely on an ellipsoidal parameter set. The problems considered are relevant to several recently developed techniques in the identi?cation for control research area. The frequency plots of such a family are characterized and an e?cient algorithm for computing the envelope of the Bode plots is devised. In particular, it is shown that the extremal values of the magnitude and phase of the value set at each frequency, which are in general non-convex optimization problems, can be computed via the solution of a sequence of Linear Matrix Inequalities (LMIs).

Uppsala Universitet, Sweden University of Glasgow, Scotland, UK

1.30 pm Box-Jenkins continuous-time modeling Rik Pintelon Johan Schoukens Yves Rolain Vrije Universiteit Brussel Vrije Universiteit Brussel Vrije Universiteit Brussel

Abstract:

3.10 pm A new orthogonal basis functions for error quanti?cation in system identi?cation R. Ouvrard P. Coirault T. Poinot J.C. Trigeassou L.A.I.I.-E.S.I.P. L.A.I.I.-E.S.I.P. L.A.I.I.-E.S.I.P. L.A.I.I.-E.S.I.P. France France France France

This paper treats the identi?cation of continuous-time models using arbitrary band-limited excitation signals. A modeling approach is presented that has the two following advantages: 1) asymptotically (the amount of data tends to in?nity) there is no approximation error over the complete frequency band from DC to Nyquist, 2) it allows to identify general parametric noise models. The key idea is to combine a continuous-time plant model with a discrete-time noise model (= hybrid Box-Jenkins model structure).

41

1.50 pm Identi?cation methods of dynamic systems in presence of input noise Umberto Soverini Torsten Soderstrom University of Bologna Uppsala University

2.30 pm On the statistical accuracy of the empirical transfer function estimator P.M.T. Broersen Delft Univ. of Technology

Abstract: Abstract:
In many practical situations the observed input–output data of an identi?ed process are corrupted by noise both on the input and output. It is well–known that in this case many identi?cation methods give biased results. In this paper the performance of three di?erent techniques which can be used for the identi?cation of errors–in–variables models are analyzed on the basis of a simulated system already proposed in the literature. More precisely, the considered methods are the Instrumental Variable (IV) method, the Frisch Scheme (FR) approach and the Joint–Output (JO) method. The comparative analysis con?rms a well–known fact that the IV method yields, in general, very inaccurate parameter estimates in spite of the low computational load required. On the contrary, the JO approach gives a very good accuracy of the parameter estimates, even in presence of high amounts of noise on the data. Its main drawback is the extensive requirements on computer time. In this respect, the FR approach might be the method to be preferred since its behaviour is characterized by a good compromise between the required computational load and the accuracy in the estimates. The simulation results show that the FR method has good statistical properties and is robust with respect to violation of the assumptions.

At its introduction, the accuracy of the Empirical Transfer Function Estimator ETFE has been described for periodical input and for a realization of a stationary stochastic process as input. Later, an in?nite value for the variance of the ETFE has been derived by treating the input as a stationary stochastic process. The results for the three di?erent types of input are compared. An example clari?es the experimental conditions for each type of input. It shows under which conditions the theory of a single realization of a stochastic process applies and when a full stochastic treatment is necessary. This has consequences for the method of averaging that has to be applied to improve the accuracy.

2.50 pm Approaches for identifying continuous-time AR processes from unevenly sampled data E. K. Larsson T. S¨derstr¨m o o Uppsala University, Sweden Uppsala University, Sweden

Abstract:

2.10 pm Frequential moments : application tolinear system identi?cation E. Etien J.D Gabano J.C Trigeassou LAII LAII LAII

When identifying a continuous-time AR process from discrete-time data, an obvious approach is to replace the derivative operator in the continuous-time model by an approximation. In some cases, a linear regression model can then be formulated. The well-known least squares method would be very desirable to apply, since it enjoy good numerical properties and low computational complexity, in particular for fast or nonuniform sampling. The focus of this paper is the latter, i.e. nonuniform sampling. Two consistent least squares schemes for the case of unevenly sampled data are presented. The precise choice of derivative approximation turns out to be crucial.

Abstract:

A new class of linear system invariant is presented. An estimation of frequential moments is performed by a least squares algorithm. The linears relationships between moments and transfer function parameters are also used to estimate local frequency models.

3.10 pm Transfer function and frequency response estimation using resonant ?lters Peter J. Gawthrop Liuping Wang University of Glasgow The University of Newcastle

Abstract:

A resonant ?lter approach is proposed for

42
direct identi?cation of continuous-time transfer functions from input-output data when the input contains periodic components. The asymptotic properties of the method are analysed; in particular the noise reduction properties are emphasised. Some illustrative simulations are provided. measurement series are available for observing basically the same process with slight changes. Examples include ?nancial series from the same economic sector, high dimensional series in the measurement of sheet forming processes such as sheet aluminum or paper machines, and highly instrumented vibrating structures. Basically, some of the measurements are providing information already contained in other measurements. By explicitly dealing with the possible rank de?ciency of the measurements, a reduced rank process can be determined that can have far fewer parameters to estimate. Since modeling accuracy as measured by variance is proportional to the number of estimated parameters, the resulting models can be considerably more accurate. Maximum likelihood tests of hypotheses are constructed to determine the best reduced rank structure by optimal statistical tests.

WeMD3

Subspace Methods: II
Chairs: B. Ninness J. Maciejowski

Santa Rosa

University of Newcastle, Australia Cambridge University, UK

1.30 pm Subspace identi?cation method for combined deterministic-stochastic bilinear systems Huixin Chen Jan M Maciejowski University of Cambridge, U.K University of Cambridge, U.K.

2.10 pm On state estimation of unknown systems based on subspace state-space system identi?cation Hiroshi Oku Hidenori Kimura University of Tokyo University of Tokyo

Abstract:

In this paper, a ‘four-block’ subspace system identi?cation method for combined deterministic-stochastic bilinear systems is developed. Estimation of state sequences, followed by estimation of system matrices, is the central component of subspace identi?cation methods. The prominent di?erence of our new approach is a ‘four-block’ arrangement of data matrices which leads to a linearization of the system state equation, when written in block form. A major advantage of this approach, over a previous bilinear subspace algorithm, is that the measured input is not restricted to be white. We show that, providing a certain data-dependent eigenvalue condition is met, our algorithm provides asymptotically unbiased estimates, and we indicate the rate at which the bias decreases. Simulation results show that this algorithm requires a smaller sample size than earlier algorithms (for comparable performance) and that the computational complexity is signi?cantly lower.

Abstract:

1.50 pm Identi?cation of colinear and cointegrated multivariable systems using canonical variate analysis Wallace E. Larimore Adaptics, Inc

In this paper, we consider a state estimation problem of a MIMO model with which a MIMO system is identi?ed by using Subspace State-Space System IDenti?cation (4SID). This problem is very important because the knowledge of the present state of a model can be a clue to design an adaptive control law for a system. First, we review the recursive 4SID algorithm based on the matrix inversion lemma shown in our previous paper. Then, we propose a scheme for estimating the state of models obtained by the recursive 4SID algorithm. A major advantage of 4SID schemes is that we can select the order of an identi?ed model as the number of dominant singular values of a certain matrix which consists of I/O data. However, it becomes a bottleneck on considering the recursive state estimation because the dimension of state space changes from one time to another during the identi?cation proceeds. To circumvent this problem, we introduce a new kind of state (named “quasi-state”) which is still meaningful over di?erent update steps. Finally, we illustrate the usefulness of our idea by considering a tracking control problem.

Abstract:

This paper addresses the problem of identi?cation of a multivariable system where the output time series may be colinear and perhaps integrating processes. Such processes can occur when numerous

2.30 pm Subspace system identi?cation of multiple time series and di?erent sampling rates D.D. Ruscio

43 Abstract:
2.50 pm On the rank de?ciency of the least squares residuals in subspace identi?cation T.V. Gestel B. De Moor P.V. Overschee In this paper we derive models of the water level in an irrigation channel from system identi?cation experiments. We present the complete system identi?cation procedure from experiment design to model validation, taking into account prior physical information and that the intended use of the models is prediction and control. It is shown that a ?rst order linear model captures the main trends in the data well, and together with prior information this model is probably su?cient for design of standard controllers. It is also shown that a higher order nonlinear model gives remarkably accurate predictions.

WeMD4

Environmental Modeling

Santa Cruz

Invited Session Organizer: I. Mareels University of Melbourne, Australia Chairs: I. Mareels P. Young University of Melbourne, Australia Lancaster University, UK

2.30 pm Identi?cation of interactions of runo? rainfall state variables for modelling hourly ?ows of a fast ?owing mountain river R. Chaudhuri A. Chaudhuri

1.30 pm Linear material ?ow models Andreas Gleiss Manfred Deistler Thomas Matyus Technical University, Vienna Technical University, Vienna Technical University, Vienna

2.50 pm Estimation of pollutant source speci?cations using a conditional deconvolution method Delmaire Benjelloun LASL, France LASL, France

Abstract:

This paper presents static and dynamic linear models for material ?ow systems. Based on the a priori knowledge given e.g. by mass balances these models are highly structured in a natural way. In the static case this structure is exploited for the reconciliation of the ?ow measurements, the estimation of the parameters and for simulation. In the dynamic case graph theoretic conditions are provided for stability, reachability and identi?ability of three types of stock dynamics.

Abstract:

1.50 pm Time variable and state dependent parameter estimation P. Young

2.10 pm System identi?cation of an open water channel Erik Weyer University of Melbourne, Australia

In this paper, we will consider the general problem of the one dimensional pollutant scattering in the air or in water described by a parabolic partial di?erential equation. Using a Fourier series expansion of the input-output variables, we show that the pollutant ?ow restoration is a classical deconvolution problem if the source position is known. Secondly, we suppose the case of two possible sources one of whom being operating. This can be split into two separated parts : a deconvolution step followed by an hypothesis testing method. Unfortunately, in the presence of noise, the detection step can be a?ected by the deconvolution step leading to false detection. To overcome this inconvenient, we propose an algorithm which thresholds the sensor measures and each conditional impulse response. This approach enables to minimize the number of parameters to be estimated and makes the detection reliable. The main interest lies in the joint estimation of the position and the source temporal ?ow from a conditional deconvolution point of view.

44
WeMD5 detecting an aircraft icing event when only a noisy state

Time-varying systems
Chairs: K. Kumamaru L. El Ghaoui

Sierra Madre North measurement is available. While previous studies have

Kyushu Institute of Technology, Japan University of California, Berkeley, USA

1.30 pm Identi?cation of time-varying systems - a survey Evgeni G. Kleiman Hannover University, Germany

addressed identi?cation during a pilot-induced maneuver, taking advantage of the excitation due to input, this paper addresses identi?cation during cruise, where excitation is provided only by unknown disturbances due to turbulence. Simulation results show that for moderate turbulence levels, the H∞ algorithm provides a timely and unambiguous icing indication.

Abstract:

This survey covers the publications issued in 1993-1997 and partly in 1990-1992 and concerns the methods of identi?cation and parameter esti- mation for time-varying systems. Consideration is given to the identi?cation methods and algorithms of linear and nonlinear systems, distributed para- meter and delay systems, input signals in dynami- cal systems.

2.50 pm Impulse response model identi?cation in the case of a multi-period ZOH input Zi-Jiang Yang Yoshihiro Kodama Kyushu Institute of Technology Kyushu Institute of Technology

1.50 pm Identi?cation for a general class of LPV models B. Bamieh L. Giarre

2.10 pm Identi?cation of ARX models with time-varying bounded parameters: A semide?nite programming approach L. El Ghaoui University of California, Berkeley, USA Giuseppe Cala?ore Politecnico di Torino, Italy

This paper considers the problem of impulse response identi?cation for a linear sampled-data system where the input signal is held constantly within a multiple of the sampling period of the output signal. To improve identi?cation accuracy, this paper proposes a new identi?cation approach by employing the Haar scaling and wavelet functions. Based on the close relation between piecewise-constant approximation with Haar scaling and wavelet functions, a hierarchical identi?cation procedure is proposed which identi?es the system impulse response from a coarse resolution level to a ?ne resolution level successively. At each resolution level, the BIC is utilized to determine the length of the decomposed impulse response in the corresponding subspace Since the identi?ed impulse response model is not smooth when it is represented by some Haar scaling functions of di?erent widths, we can replace each Haar scaling function in the preidenti?ed impulse response model by a Gaussian basis function with corresponding position and width. Then an improved identi?cation method is also proposed to achieve smooth continuous-time impulse response model from sampled data.

Abstract:

2.30 pm H∞ parameter identi?cation for in?ight detection of aircraft icing: the time-varying case James W. Melody Univ. Illinois at Urbana-Champaign Thomas Hillbrand Darmstadt Univ. Technology Tamer Basar Univ. Illinois at Urbana-Champaign William Perkins Univ. Illinois at Urbana-Champaign

Wednesday afternoon:
4.00 pm to 6.00 pm

WePM1

Abstract:

In?ight parameter identi?cation of aircraft ?ight dynamics is considered in this paper in the context of an ice management system. In particular, an H∞ parameter identi?cation algorithm is evaluated in terms of

Identi?cation for Control
Chairs: X. Bombois

San Miguel

CESAME, Louvain-la-Neuve, Belgium

45
A. Besan?on-Voda c ENSIEG, Grenoble, France of SISO stochastic discrete-time linear time-invariant systems. In an e?ort to decrease the undermodeling-induced component of the estimation error,model structure allowing for the encoding of prior knowladge of pole position, have been introduced. For that in model is introduced the set of linear ?lters. Choice of ?lters can give Laguerre basis, Kautz basis or rational wavelet basis. The identi?cation algorithm is two-stage nonlinear algorithm. The ?rst stage involves taking the inverse discrete Fourier transform and multiplication by suitable window function. The second stage involves ?nding the best analytic approximation using Nehary theorem to the function obtained at stage one. The disturbance is stochastic and a priori is known class of distributions to which real noise belongs. That is much more unrestrictive assumption in comparison with exact knowladge of noise distribution as usualy is supposed. The algorithm yields both an identi?ed model and explicit H-inf norm error bound. The H-inf norm of error bound depends from noise uncertainty (class of distributions). Finally, the conditions in terms of properties of convolutive window function and stochastic noise are speci?ed under which algorithm is convergent.

4.00 pm A nearly interpolatory algorithm for H∞ identi?cation with mixed time and frequency response data G. Gu J. Chen

4.20 pm Use of error criteria in identi?cation for control Yucai Zhu Eindhoven Univ. Tech. & Tai-Ji Control

Abstract:

This work discusses the use of error criteria in parameter estimation and in model order selection where the purpose of identi?cation is control. It will be shown that in general a model estimated using output-error criterion is not optimal according to output-error criterion. Prediction-error criterion will perform better in parameter estimation. This fact implies that an output-error model is not optimal for simulation. For model order selection, on the other hand, prediction-error criterion is not control relevant; and output-error will do a better job. Therefore, for control relevant identi?cation, it is suggested that prediction error criterion is used for parameter estimation and output error criterion for model order selection. Simulations will be used to illustrate the ideas. Keywords: Identi?cation for control, error criteria, parameter estimation, order selection

5.20 pm Approaches to control-oriented H∞ identi?cation Mehrzad Namvar Alina Voda Lab. d’Auto. de Grenoble, France

4.40 pm The problem of constructing strong models with a minimum operator norm V. Rusanov A. Daneev

This paper presents deterministic approches to control-oriented H∞ identi?cation of linear systems using time domain measurements. In the open-loop approach, we identify the models with the minimum additive uncertainty in H∞ norm and then constant or frequency dependent bounds for the additive modeling error, in the same optimization procedure are obtained. In the closed-loop approach, the models are identi?ed in such a way that the multiplicative error in the closed-loop sensitivity functions are minimized. The achieved models will then be suitable for iterative identi?cation-control design. All these problems are solved via LMI based optimization techniques which allow to transform di?erent identi?cation objectives into LMI constraints. Finally the applicability of the proposed methods is illustrated by a numerical example.

Abstract:

5.00 pm Robust identi?cation in H∞ of non-Gaussian systems Vojislav Filipovic CNTS,Yugoslavia

Abstract:

This paper considers identi?cation in H-inf

46

5.40 pm Controller validation for stability and performance based on an uncertainty region designed from an identi?ed model X. Bombois M. Gevers G. Scorletti B.D.O. Anderson Universit? Catholique de Louvain e Universit? Catholique de Louvain e LAP ISMRA Australian National University

WePM3

Blind Identi?cation, Equalization and Source Localization Santa Rosa
Chairs: B. Wahlberg E.-W. Bai Royal Institute of Technology, Sweden University of Iowa, USA

Abstract:

This paper focuses on the validation (for stability and for performance) of a controller that has been designed from an unbiased model of the true system, identi?ed either in open-loop or in closed-loop using a prediction error framework. A controller is said to be validated for stability if it stabilizes all models de?ned by an ellipsoidal parametric uncertainty set containing the true system with some prescribed probability. Such uncertainty set is computed from the covariance matrix of the parameters of the identi?ed model. Our ?rst contribution is to design the general LFT framework for robustness stability analysis linking the controller to be validated with such parametric uncertainty region resulting from prediction error identi?cation. This leads us to a necessary and su?cient condition for the robust stabilization of all plants in this nonstandard uncertainty region. Our second contribution is to show that we can compute the worst case performance of a given controller over all systems in such uncertainty set described by ellipsoidal regions in parameter space, by recasting the problem as an LMI-based optimization problem, for which the exact solution can be computed. A controller is then said to be validated for performance if the worst case performance is better than some threshold value.

4.00 pm Blind estimation and deconvolution of communication channels with unbalanced noise Paolo Castaldi Roberto Diversi Roberto P. Guidorzi Umberto Soverini Universit` a Universit` a Universit` a Universit` a di di di di Bologna Bologna Bologna Bologna

Abstract:

WePM2

What has Continuous time System Identi?cation to O?er? Sierra Madre South
Invited Session Organizers: H. Madsen Tech. University of Denmark, Denmark T. S¨derstr¨m o o Uppsala Universitet, Sweden Chairs: H. Madsen T. S¨derstr¨m o o Tech. University of Denmark, Denmark Uppsala Universitet, Sweden

Recent years have seen an increasing interest in so called ”blind identi?cation” problems; these problems consider the estimation of the channel parameters without knowing its input and/or the reconstruction of the input sequence without knowing the channel characteristics. Problems of this type arise in many ?elds and are of particular interest in modern communications and signal processing. Two di?erent approaches can be used for solving the blind identi?cation problem: methods based on higher-order statistics of the processes and approaches based on second-order statistics. In this paper a new blind identi?cation method, belonging to the latter class, is presented. It can be considered as the development of the techniques proposed for identifying errors-in-variables models. By using the structural shift properties of the generalized Sylvester matrix of the system, an estimate of the covariance matrix of the unknown source is obtained and a suitable identi?cation criterion for the estimation of the channel coe?cients introduced. A reconstruction of the input sequence is then obtained by means of a standard LS equalizer. Unlike several other blind identi?cation approaches, this method allows to deal with output measurements a?ected by di?erent amounts of noise on the channels. Numerical simulations, performed on a system taken from the literature, show that the method can be advantageously used also in presence of poor signal-to-noise ratios.

Panel discussion: 4.00 pm to 6.00 pm. Panelists: Graham Goodwin, Henrik Madsen, Johan Shoukens, Peter Young, University of Newcastle, Australia Tech. Univ. Denmark, Denmark Free Univ. Brussels, Belgium Lancaster University, UK

47
4.20 pm Adaptive identi?cation and equalization for time-varying channel Takayuki Naito Keio University, Hiromitsu Ohmori Keio University, Akira Sano Keio University, Kohichi Hidaka Tokyo Coll. Aeronautical Eng., Japan Japan Japan Japan estimation.

5.00 pm Blind system identi?cation and equalization E-W. Bai

Abstract:

A new deterministic adaptive identi?cation and equalization approach is investigated based on an internal model principle to take account of time-varying dynamics of the channel parameters. The proposed adaptive algorithm incorporates an internal model for compensating the dynamics of the time-varying parameters of the FIR channel model. The stability of the adaptation can be assured by designing the parameter adjusting dynamics in the adaptive algorithm by the help of the small gain theorem and passivity theorem. Its e?ectiveness of the proposed algorithms is examined and compared to ordinary LMS approaches in numerical simulation of adaptive identi?cation and equalization of a fading channel with parameters changing with a ?rst-order function of time.

5.20 pm Direction estimation of coherent signals without eigendecomposition and spatial smoothing Jingmin Xin YRP Mobile Telecom. Key Tech. Res. Lab Akira Sano Keio University

Abstract:

4.40 pm Blind localization by subspace method for a scattering model Eric Ternisien Gilles Roussel Mohammed Benjelloun LASL, France LASL, France LASL, France

Abstract:

We are interested in the parameters estimation of a dispersive system modeled by a physical equation in the case of a ?nite area. Furthermore, in this area, a temporally and spatially unknown source emits a pollutant of which the variations are recorded thanks to a multisensor network located in the monitored area. The problem is then to localize the source and to identify the propagation parameters using only the sensors signals. The source-multisensor approach permits to consider the problem from a system theory point of view, and hence to transform the physical model in a state-space model where the state variables correspond to the local concentration estimates. Moreover, the blind aspect of the problem requires the use of speci?c methods such as subspace one. Observed signals cross-correlation is also used to estimate the time-delay between two sensors, and thus to brought information about the source location. So, we propose an identi?cation and localization method based on a subspace approach combinated with a priori information about the form of the state-space matrices, and with time-delays

In array signal processing, a major problem is the estimation of the directions-of-arrival (DOA) of the signals impinging on an array of sensors. Although the spatial smoothing based eigenstructure methods can be used to resolve the problem of direction estimation of coherent signals, these methods are computationally expensive, so more e?cient algorithms are desired in some applications such as mobile communication systems where multipath propagation is often encountered. Therefore the purpose of this paper is to investigate a new approach for DOA estimation of the coherent narrow-band signals in communication systems without eigendecomposition and spatial smoothing. By exploiting the inherently temporal properties exhibited by most communication signals, in this paper, we propose a new cyclic approach for estimating the directions of coherent narrow-band signals without eigendecomposition and spatial smoothing processes. In which, the cyclostationarity and subarray scheme are combined to decorrelate the signal coherency and to suppress the noise and interfering signals, while the noise subspace is estimated from the resulting cyclic correlation matrix by a linear operation. As a result, the proposed approach has the advantages of computational simplicity and robust detection capability. The performance of the presented approach is veri?ed through numerical examples

48
WePM4 local polynomial modeling ideas. This has several

Model Selection
Chairs: A. Stenman K. Nakano

Santa Cruz implications on the choice of model structure, which is
Link¨ping Universitet, Sweden o Univ. Electro-Communications, Japan

4.00 pm Information criterion for selection of model with controller design K. Tsumura

discussed at length in the paper. It is concluded that the NARX structure should be considered as the default choice in the local polynomial context. Furthermore, it is shown that the predictions in some situations can be enhanced by tuning other parameters that are special for the nonparametric case. The usefulness of the method is illustrated in numerical simulations. For the chosen application it is shown that the prediction errors are in order of magnitude directly comparable to more established modeling tools such as arti?cial neural nets.

4.20 pm System identi?cation using a multi-model approach: model complexity reduction Anass Boukhris Gilles Mourot Jos? Ragot e CRAN, France CRAN, France CRAN, France

5.00 pm Model order selection of N-FIR models by the analysis of variance method Ingela Lind Linkoping University, Sweden

Abstract:

Abstract:

A multi-model provides a powerful tool for non-linear dynamic systems identi?cation. Theoretical studies have shown that any continuous function on a compact domain can be approximated arbitrarily well by a fuzzy model. However, due to the curse of dimensionality, building a multi-model is a di?cult task in practice without some precautions. Hence, only the simplest structure ?tting the training data must be chosen in order to keep good generalisation capabilities of the model. In this paper, parsimony is obtained by removing redundant and/or useless rules. The proposed approach has been successfully illustrated by simulation examples from literature.

Identi?cation of nonlinear ?nite impulse response (N-FIR) models is studied. In particular the selection of model structure, i.e., to ?nd the contributing input time lags, is examined. A common method, exhaustive search among models with all possible combinations of the input time lags, has some undesired drawbacks, as a tendency that the minimization algorithm gets stuck in local minima and heavy computations. To avoid these drawbacks it is needed to know the model structure prior to identifying a model. In this contribution it is shown by experiment that a statistical method, the analysis of variance, is a good alternative to exhaustive search in the identi?cation of the structure of nonlinear FIR-models. It is possible to reduce the risks of getting an erroneous model structure due to the non-convexity of the minimization problems, reduce the computation time needed and also get a good estimate of how far the ?t of the desired model can be enhanced.

4.40 pm On model structure selection for nonparametric prediction methods Anders Stenman Link¨pings universitet, Sweden o

Abstract:

In this paper we continue to explore identi?cation of nonlinear systems using the previously proposed concept of model-on-demand. The idea is to estimate the process dynamics locally and on-line using process data stored in a database, and has in earlier contributions proven to be capable to produce results comparable to (or better than) other nonlinear black-box approaches. The modeling part of the method is based on

5.20 pm Algorithm for three-step estimation of transfer function with unknown delay steps and order Mitsuki Mashino Fukuoka Indust. Tech. Center, Japan Fujio Ohkawa Kyushu Institute Technology, Japan Kazushi Nakano Univ. Electro-Communications, Japan Masayoshi Tomizuka Univ. California, Berkeley, USA

49 Abstract:
For the problem of consistently estimating the transfer function in the presence of colored output noise, a three-step estimation procedure has been previously developed. This procedure exploits the input/output correlation information with respect to the correlation time in the ?rst and second steps by setting the model order higher than the actual order. The parameters of the higher order model are transformed to those of the actual order model in the third step by the compressive parameter transformation yielding the bias-compensated least squares (BCLS) estimates. In this paper, the three-step estimation procedure is extended to deal with the transfer function with unknown delay steps and order. The auto-correlation function of the error between the process output and model output is utilized for evaluating the model ?tness. The validity of the proposed method is demonstrated through a simulation using a sample set of data in MATLAB. 5.40 pm Automatic model selection for linear time-invariant systems Gyula Simon Johan Schoukens Yves Rolain BUTE, Hungary Vrije Universiteit, Brussel, Belgium Vrije Universiteit, Brussel, Belgium

ThAM1

Adaptive Control: I

San Miguel

Chairs: Y. Miyasato Inst. Statistical Mathematics, Japan T. Basar Univ. of Illinois, Urbana-Champaign, USA

10.00 am General forms of adaptive nonlinear H∞ control for processes with bounded variations of parameters Yoshihiko Miyasato Inst. Statistical Math., Japan

Abstract:

Abstract:

A completely automatic identi?cation system is described which uses the latest results of the frequency domain identi?cation approach to provide an easy-to-use and reliable tool to the inexperienced users. The proposed system accepts periodic measurement data and performs the whole identi?cation procedure automatically from the data-preprocessing step to model and parameter selection. The validation phase provides information on the results at the level of the non-expert user. The performance of the system is illustrated by results obtained from real measurement data.

A new class of adaptive nonlinear H∞ control systems for processes with bounded variations of parameters, is proposed in this manuscript. Those control schemes are derived as solutions of particular nonlinear H∞ control problems, where unknown system parameters are regarded as exogenous disturbances to the processes, and thus, in the resulting control systems, the L2 gains from system parameters to generalized outputs are made less than the prescribed positive constants. It is shown that the proposed control strategy can be applied to any time-varying (and even time-invariant) systems, and the resulting control systems are bounded for arbitrarily large but bounded variations of time-varying parameters. Also, the control schemes are shown to be sub-optimal to some H∞ cost functionals (or certain di?erential games), when the high-frequency gains are time-invariant. Even if that condition does not hold, the boundedness of overall systems is assured and the L2 gains from system parameters to generalized outputs are prescribed explicitly. Not only the general forms of the control schemes, but also two explicit descriptions of those general ones, are provided.

Thursday morning:
10.00 am to 12.00 noon

10.20 am Robust output tracking for uncertain strict-feedback systems with unknown virtual control coe?cients G. Arslan T. Basar

Plenary Lecture 2
8.30 to 9.30 am

San Rafael
10.40 am Partial convergence of coupled time-varying systems Keum-Shik Hong Kyung-Jinn Yang Pusan National Univ., Korea Pusan National Univ., Korea

Complexity and information in data Jorma Rissanen IBM Research, California, USA

50 Abstract:
Asymptotic behavior of a partial state of coupled nonautonomous ordinary and/or partial di?erential equations is investigated. The system assumes the existence and uniqueness of the solution and the existence of a Lyapunov function candidate. Although ordinary Lyapunov analysis can not guarantee the asymptotic convergence of the state, the analysis in this paper assures the asymptotic convergence to zero of the partial state which remains in the time derivative of the Lyapunov function candidate. The results are applied to adaptive systems. unstable zero and pole was simulated. Simulation results showing that the proposed controller copes usefully with such an unstable and nonminimum phase plant are given.

ThAM2

Process Applications: I
Chairs: B. Juricek V. Verdult

Santa Rosa

Univ. of California, Santa Barbara, USA University of Twente, The Netherlands

11.00 am Adapative control of stochastic strict-feedback systems under a risk - sensitive criterion G. Arslan T. Basar

10.00 am Identi?cation of the Tennessee Eastman challenge process with subspace methods Ben C. Juricek Dale E. Seborg Wallace E. Larimore Univ. California, Santa Barbara Univ. California, Santa Barbara Adaptics, Inc.

11.20 am An estimation algorithm for stable adaptive control of not necessarily minimum phase systems with bounded disturbances Leonid S. Zhiteckij Inst. Cybernetics, Kiev, Ukraine

Abstract:

Abstract:

This paper deals with indirect adaptive control of possibly nonminimum phase discrete-time linear system subject to bounded disturbances. The plant to be controlled needs to be only controllable. It is assumed that the bounds on the disturbances are known. No a priori information about the bounds of plant parameters is required. To overcome the singularities that may appear in the adaptive control law, a new way is proposed using a parameter estimates modi?cation. It is based upon set-membership identi?cation concept. In contrast to other algorithms, the algorithm advanced in this paper yields a set of admissible estimates. Finding suitable modi?ed estimates reduces to maximizing the absolute value of the determinant of the associated Sylvester resultant matrix. A key feature of the adaptive scheme is that the maximization is over the only one variable that is scalar. From a practical point of view, it is attractive that the proposed estimation technique requires small computational e?ort for its implementation. The control law is derived utilizing the standard pole placement approach. The convergence and stability properties of the adaptive controller are established. To demonstrate some features of the control scheme, the adaptive closed-loop system containing a plant whose transfer function has one

The Tennessee Eastman Challenge Process is a realistic simulation of a real chemical process that has been widely used in process control studies. In this case study, several identi?cation methods are examined and used to develop models that contain seven inputs and ten outputs. ARX and FIR models are identi?ed using reduced-rank regression techniques (PLS and CCR) and state-space models identi?ed with predictive error methods and subspace algorithms. For a variety of reasons, the only successful models are the state-space models produced by two popular subspace algorithms, N4SID and Canonical Variate Analysis (CVA). The CVA model is the most accurate. Important issues for identifying the Tennessee Eastman Challenge Process and comparisons between the subspace algorithms are also discussed.

10.20 am Nonlinear identi?cation of high purity distillation columns C.T. Chou H.J.J. Bloemen V.Verdult T.J.J. van den Boom T. Backx M. Verhaegen

51
scheme for an industrial reactive distillation column. 10.40 am Qualitative modelling as a key technique for the automatic identi?cation of mathematical models of chemical reaction systems David Schaich Ralf Becker Rudibert King IPAT, TU Berlin IPAT, TU Berlin IPAT, TU Berlin

11.20 am On modelling and control of a rotary sugar dryer Sergio M. Savaresi Robert R. Bitmead Robert D. Peirce Politecnico di Milano University of California San Diego CSR Ltd, Australia

Abstract: Abstract:
Mathematical modelling and identi?cation of systems, where the structure of some equations is unknown, is a di?cult and time-consuming task. As this is usually done manually by a human expert, it is associated with terms such as intuition. However, an experienced modeller will have a clear procedure to deduce a model. In a ?rst step he or she will analyse experimental data on a qualitative level. Only those models are then tested in a quantitative identi?cation which pass the qualitative check. In this contribution this e?cient procedure is translated into methods. A tool, TAM-C, is introduced which automatically ?nds structures and parameters of formal kinetics. The main part of the contribution is dedicated to the qualitative preselection of possible model candidates and the application to an industrial process.

This paper deals with the problem of ?tting a set of data collected on a rotary sugar dryer, by means of a ?rst-principles mathematical model. Thanks to the highly-constrained strcture of the model it is discovered that the sugar dryer is characterised by two di?erent working conditions: the ”standard-mode” (characterised by a non-zero sugar moisture content), and the ”overdried-mode” (namely a condition where the sugar moisture content is almost nil, and the evaporative phenomenon becomes negligible). On the basis of this two-stage behaviour, an accurate ?t between the model and the measured data can be achieved, and an innovative control strategy can be drawn.

11.00 am Combining ?rst principles with black-box techniques for reaction systems Libei Chen Yves Hontoir Dexian Huang Jie Zhang Julian Morris Catholic Univ. Louvain, Belgium Solvay s.a., Belgium Tsinghua University, China Univ. Newcastle-upon-Tyne, UK Univ. Newcastle-upon-Tyne, UK

11.40 am Model-on-Demand identi?cation for control: an experimental study and feasibility analysis for MoD-based predictive control Martin W. Braun Brian A. McNamara Daniel E. Rivera Anders Stenman Arizona State Arizona State Arizona State Link¨ping o University University University University

Abstract:

Abstract:

A control oriented hybrid model structure combining ?rst principles models with standard black-box techniques for modelling nonlinear dynamics of reaction systems is presented in this paper. The approach is formulated in a general framework for continuous stirred tank reactors and analyzed in details through a case study of a reactive distillation column. The approach is based on easily established mass balance equations, the stoichiometry of the system as well as model reduction techniques. The choice of combined inputs and the model structure is motivated by some general control objectives for this class of systems. A progressive identi?cation of this model structure can be performed when a dominant part exists. The application to real process data is presented. This model structure has been successfully used in an IMC

An experimental study of “Model-on-Demand” (MoD) identi?cation is made on a pilot-scale brine-water mixing tank. MoD estimation is compared against semi-physical modeling techniques using identi?cation data generated from a systematically designed m-level Pseudo-Random Sequence (PRS) input. The estimated models are the basis for evaluating the usefulness of MoD-based Model Predictive Control (MPC). For this application, MoD-MPC is shown to provide better performance at high bandwidths compared to a linear MPC controller.

52

ThAM3

Modeling and Identi?cation with Orthogonal Basis Functions Sierra Madre South
Invited Session Organizers: P. Van den Hof Delft Univ. Tech., The Netherlands B. Wahlberg Royal Institute of Technology, Sweden Chairs: P. Van den Hof B. Wahlberg Delft Univ. Tech., The Netherlands Royal Institute of Technology, Sweden

11.20 am Practical aspects of using orthonormal system parameterisations in estimation problems Brett Ninness Stuart Gibson Steve Weller University of Newcastle University of Newcastle University of Newcastle

Abstract:

10.00 am Modelling and identi?cation with rational orthogonal basis functions Paul Van den Hof Delft Univ. Tech., Netherlands Bo Wahlberg Royal Inst. Technology, Stockholm Peter Heuberger Delft Univ. Tech., Netherlands Brett Ninness Univ. of Newcastle, Australia Jozsef Bokor SZTAKI, Hungarian Acad. Sciences Tom?s Oliveira e Silva a University of Aveiro, Portugal

The last decade has seen a resurgence of interest in the use of ortho-normalised parameterisations for system estimation applications. This has produced a large body of theoretical work that has sought to provide sound scienti?c underpinnings. The paper here has a di?erent focus in that emphasis is placed on practical dividends associated with orthonormal parameterisations.

11.40 am Partial realization in generalized bases: algorithm and example Thomas J. de Hoog Peter S.C. Heuberger Paul M.J. Van den Hof Delft University of Technology Delft University of Technology Delft University of Technology

Abstract:

Decomposing dynamical systems in terms of orthogonal expansions enables the modelling/approximation of a system with a ?nite length expansion. By ?exibly tuning the basis functions to underlying system characteristics, the rate of convergence of these expansions can be drastically increased, leading to highly accurate models (small bias) being represented by few parameters (small variance). Additionally algorithmic and numerical aspects are favourable. A recently developed general theory for basis construction will be presented, that is a generalization of the classical Laguerre theory. The basis functions are applied in problems of identi?cation, approximation, realization, uncertainty modelling, and adaptive ?ltering, particularly exploiting the property that basis function models are linearly parametrized. Besides powerful algorithms, they also provide useful analysis tools for understanding the underlying identi?cation/approximation algorithms.

Abstract:

A solution is presented for the problem of realizing a minimal state-space model of an LTI discrete-time system from a partial expansion in terms of generalized orthonormal basis functions, also known as Hambo basis functions. For the solution of the minimal partial realization problem fruitful use is made of the Hambo operator transform theory that underlies the basis function expansion. The resulting realization algorithm can be applied in an approximative sense, for instance for the computation of a low order model from a large basis function expansion that is obtained in an identi?cation experiment. This approach is illustrated with an example.

11.00 am Optimal pole locations for Laguerre and two-parameter Kautz models: a survey of known results T. Oliveira e Silva

53
auxiliary input is designed to enlarge the distance measured by the Kullback discrimination information measure between the system models corresponding to the normal and the fault modes using the exact knowledge of system models. However, in practice, the designer hardly knows the exact models. Hence, the optimal auxiliary input should be designed for the case of the system models with uncertainty. Here, the auxiliary input is designed to maximize the distance for the worst combination of system models under the model uncertainty. Numerical simulation results indicate that the proposed auxiliary input in fault detection using the backward sequential probability ratio test reduces the mean detection time without making much e?ects on original system behavior and false alarm rate.

ThAM4

Fault Detection and Monitoring
Chairs: M. Basseville K. Kumamaru

Santa Cruz

IRISA/CNRS, France Kyushu Institute of Technology, Japan

10.00 am A nonlinear adaptive observer based method for fault detection and isolation Qinghua Zhang IRISA-INRIA

Abstract:

In this paper we propose a new method for fault detection and isolation based on adaptive observers. Each of the adaptive observers estimates the state variables and one of the parameters of the monitored system. The algorithms for both residual generation and evaluation are presented. The proposed method is applicable to nonlinear dynamic systems for which the required adaptive observers can be designed.

11.00 am Detection and isolation of sensor and process faults in a heat exchanger using a fuzzy-model library Karsten Spreitzer Peter Ball TU Darmstadt - IAT TU Darmstadt - IAT

Abstract:

10.20 am Frequency domain local tests for change detection. Albert Benveniste Bernard Delyon Mich`le Basseville e IRISA/INRIA, France IRMAR, France IRISA/CNRS, France

Abstract:

We propose frequency domain tests for early detection of slight changes in input/output transfer functions. The experimental setup consists in recording successive pairs of i/o spectra. We propose a frequency domain test based on LeCam’s local approach, by extending the techniques of two of the authors to nonparametric frequency domain. The target application which motivates this study is quality control in mechanics.

10.40 am Optimal auxiliary input design for fault detection of systems with model uncertainty Toshiharu Hatanaka Katsuji Uosaki Tottori Univ., Japan Tottori Univ., Japan

This contribution presents an approach to model-based fault detection and isolation (FDI) of sensor and process faults for nonlinear processes. The examined approach decomposes the process into subprocesses and identi?es a nonlinear model for each. Takagi-Sugeno type fuzzy models generated using the LOLIMOT algorithm are used. The algorithm is based on the idea of approximating a nonlinear dynamic discrete function by piece-wise linear models. Each local model is valid in a subregion of the input space. The validity is expressed by a membership function continuously de?ned over the whole input space. The model output is calculated as the weighted sum of all local models. The model library is used to generate several estimates for process outputs and states. Comparing these estimates with measured signals leads to residuals which indicate the state of the system and provide information about the source of possible faults. Implementation of the models either as a parallel model or as a series-parallel model lead to di?erent FDI results. In this contribution this di?erent sensitivity is investigated also. The applicability of the multi-model based FDI is illustrated on an industrial scale thermal plant. Seven di?erent process faults and eight di?erent sensor faults can be detected.

Abstract:

Introduction of an auxiliary input is known to be useful to detect the system fault quickly. Such an

54
11.20 am A geometric based parametrization of the ?lter gains for the fault Ricardo Lopezlena Estrada, Instituto Mexicano del Petr?leo, M?xico o e Juan Carlos Mart? ?nez Garc? ?a, CINVESTAV-IPN, M?xico. e three stages. In the ?rst stage, the user selects an appropriate Galois ?eld, with up to 128 elements. In the second stage, the user generates a suitable maximum-length sequence in the ?eld, with a period up to 20000. In the ?nal stage, the user generates the required pseudo-random signal by converting the ?eld elements into a suitable set of signal levels. Comprehensive help and viewing facilities are provided at every stage. Sets of pre-computed conversions are also available to the user in the ?nal stage to facilitate the generation of pseudo-random signals with desirable harmonic properties for the purposes of system identi?cation.

Abstract:

This paper concerns a methodology for the synthesis of Fault Detection and Isolation (FDI) ?lters. Our methodology is based on a geometric based parametrization of the feedback gain matrices which solve the observer-based FDI problem, i.e., the so-called Beard Jones Detection Fault Problem (BJDFP). We also consider here the so-called Restricted Diagonal Detection Fault Problem (BJDFP). Our proposed methodology is based on the characterization of the ?xed dynamics of the observer solutions. We illustrate the methodology with some academic examples.

Tai-ji ID Yucai Zhu

VCLab - the virtual control engineering laboratory ThAM5 Christian Schmid Ruhr-Universitaet Bochum, Germany

Software Demonstration Session I (General) Sierra Madre North
Invited Session Organizer: R. Schumann Fachhochschule Hannover, Germany Chairs: R. Schumann H. Barker Fachhochschule Hannover, Germany Univ. Wales Swansea, U.K.

Abstract:

Software demonstation session: 10.00 am to 12.00 noon

Galois - a program for generating pseudo-random perturbation signals H. A. Barker Univ. Wales Swansea, U.K.

This contribution presents a remote laboratory approach using virtual reality on the Web. Extensions like plug-ins and Java applets, which use MATLAB/SIMULINK for simulations, are integrated into the Web browser. Interactive animation of control engineering experiments can be performed using VRML models of laboratory plants. The Web-based techniques for realising a virtual laboratory of this kind are introduced and discussed in detail. The central focus of this contribution is put on the technical realisation of the most important elements. Some virtual laboratory examples illustrate the usefulness of this approach.

Abstract:

GALOIS is a program for generating pseudo-random signals based on maximum-length sequences in Galois ?elds. The PC version of the software has been developed with Borland C++ Builder from the original Open Windows version for Sun workstations. GALOIS is based on the properties of the Galois ?elds in which the maximum-length sequence is generated and on the properties of the conversions of the ?eld elements into signal levels through which the pseudo-random signal is generated from the maximum-length sequence. The program runs under a graphical user interface that provides for the generation of pseudo-random signals in

WINROSA 2.0 and DORA for Windows 6.3 Peter Krause Angelika Krone Timo Slawinski Rainer Knicker University University University University of of of of Dortmund Dortmund Dortmund Dortmund

Abstract:

In recent years, many methods have been developed for the di?erent data–mining ?elds. One of the ?elds is fuzzy modeling, which is predominantly applied, if interpretable results are desired. In order to build and test

55
fuzzy models, powerful software tools for fuzzy modeling and simulation are needed. In this paper two programs, WINROSA and DORA for Windows, are presented, which full?l both requirements. WINROSA incorporates the Fuzzy–ROSA method, which is a method for automatic rule generation on the basis of relevance tests. DORA for Windows is a simulation tool with emphasis on fuzzy systems. The features of both programs are presented and it is shown how both programs complement each other. input perturbation.

Adaptive position control of a ?exible manipulator using ANNNAC Torsten Knohl Hanqing Zeng Heinz Unbehauen Ruhr-Universit¨t Bochum a Ruhr-Universit¨t Bochum a Ruhr-Universit¨t Bochum a

Thursday midday:
1.30 pm to 3.30 pm

Abstract:

ThMD1

Adaptive Control: II
Poster session: 1.30pm to 3.30pm

Santa Ynez

Direct adaptive control of a linear parabolic system ? St?phane Renou e Ecole Polytechnique de Montral ? Michel Perrier Ecole Polytechnique de Montral Denis Dochain Universit? Catholique de Louvain e Sylvain Gendron PAPRICAN

This paper is concerned with the position control of a ?exible manupilator using a combination consisting of arti?cial neural networks (ANNs) and a nonlinear adaptive controller (NAC). A class of Hammerstein models is introduced to describe the dynamics of the ?exible manipulator by taking into account the friction and dead-zone nonlinearities in its driving system. For the control of such a Hammerstein system the application of an inverse of the static input nonlinearity is proposed for compensation, whereas a linear adaptive controller is used for the resulting dynamic system. In the proposed new scheme, an ANN is applied instead of a ?xed inverse nonlinearity. The key feature of this approach is that the ANN can describe several types of nonlinear functions without structural changes. To control the linear part of the system, an adaptive LQ controller is installed.

Abstract:

The dynamics of several chemical reactors are described by partial di?erential equations (PDE’s) and are therefore considered as distributed parameter systems. A typical example is the tubular reactor for which mass and energy balances lead to parabolic PDE’s that account for convection, dispersion and conversion occurring in the reactor. The use of a PDE model may be an interesting approach in control design if the PDE model gives a more accurate representation of reality. Various approaches have been considered to use the PDE phenomenological model directly. They can be divided in two groups. The ?rst group is composed of early lumping methods where a preliminary discretization is used. The second group is composed of late lumping methods where the controller design problem is solved directly with the PDE model. This paper presents a late lumping approach using adaptive model reference algorithm for a linear distributed parameter system with input boundary control and output boundary reference. The control and adaptation laws are designed by using a Lyapunov second method approach. This controller is applied to a tubular reactor model with unknown kinetic parameters. Simulation results are shown for set-point changes, variation of kinetic parameters and

Delta modi?cation of self-tuning pole placement PID controllers Vladim? Bob?l ir a Brno Univ. Tech., Czech Republic Petr Dost?l a Brno Univ. Tech., Czech Republic Martin Sysel Brno Univ. Tech., Czech Republic

Abstract:

This contribution presents the application of self-tuning digital PID controllers for process control modelled by δ - models. The process is identi?ed by the regression (ARX) model using the recursive least squares method (RLSM) with applied directional forgetting. The basic RLSM algorithm has been modi?ed for the δ - model structures. Controller synthesis is designed on the basis of the pole placement method (a double real pole and a pair of complex conjugated poles with the same real component has been chosen in the characteristic polynomial). The real component in?uences the speed of the control loop transient and in?uences control output changes, too. By changing the imaginary component the desired overshoot can be in?uences. Two digital PID controller structures have been designed in δ - modi?cation. Control results

56
using digital PID controller on the basic δ - and z - models are compared. By comparing both approaches it is obvious that the controller output u(k) in the δ - model has excellent behaviour (without oscillation), the parameter estimates have excellent convergence. Over a short sampling period the δ - parameter estimates converge to the continuous parameters of the process model and the controller output u(k) converges to the steady state.

Abstract:

In the conventional mine winder control systems, S-curve control and PI control are mainly utilized. Since the control parameters of these control methods are constant, they can not adapt su?ciently to the large change of the wire rope length. In this work, we propose a new robust winder control system using the adaptive model output following control based on Backstepping strategy. Further, it is shown that the e?ectiveness and robustness of the proposed control method are con?rmed through experiments using an experimental winder model with a maximum length of 15 meters of wire rope.

Modeling and adaptive control for a two-link ?exible manipulator A. Krolikowski P. Banaszczyk Poznan University of Technology Poznan University of Technology Adaptive control mixer method for nonlinear control recon?guration: a case study Zhenyu Yang Mogens Blanke Aalborg University, Denmark Aalborg University, Denmark

Abstract:

This paper presents modeling and adaptive control for a two-link ?exible manipulator. Dynamic nonlinear model of an experimental two-link ?exible manipulator is derived on the basis of Lagrange assumed-modes approach. In this approach links are modeled as Euler-Bernoulli beams of uniform density and constant ?exural rigidity, and the de?ection of each link is approximated using the assumed mode shapes for a speci?ed number of elastic modes. A reduced order model comprising two modes for each link is developed. The natural frequencies of vibration are calculated by solving the eigenvalue problem. These are also veri?ed by the FFT analysis performed on the data from a model and from experiments. The obtained results show a good way of modeling of a real manipulator which is suitable for control purpose. To this end, an adaptive low-dimensional controller of saturation type based on the presented model is given and simulations of adaptive control behavior are presented. However, for practical implementation the control algorithm should be modi?ed taking into account that in real system the modal amplitudes are not directly available for measurement. Finally, the measurement and control system set-up is shortly described.

Abstract:

The frequency-domain control mixer method is extended into an adaptive form for application in the nonlinear control recon?guration. Within each local range the control mixer matrix is calculated based on the faulty and nominal linear system models which are derived from on-line linearization of the faulty and ?ctitious nominal nonlinear systems at common state points. In global level, the control mixer matrix need to be updated with respect to the updatings of local linear models which validations are evaluated by their tracking ability to corresponding nonlinear systems. The simulation on a nonlinear ship propulsion system shows the potential of this method in practical application.

ThMD2

Process Applications: II
Poster session: 1.30pm to 3.30pm

Santa Ynez

Robust winder control system using adaptive model output following control method Masanori Takahashi Ariake National Coll. Tech., Yoshinori Kawasaki Ariake National Coll. Tech., Ikuro Mizumoto Kumamoto University, Motoyuki Kuribayashi Mitsui Miike Machinery, Tomohiro Yasukouchi Mitsui Miike Machinery, Japan Japan Japan Japan Japan

System identi?cation and physical parameter estimation of anti-vibration units in semiconductor exposure apparatus Shuichi Adachi Hiroyuki Takanashi Hiroaki Kato Takehiko Mayama Shinji Wakui Utsunomiya University Utsunomiya University CANON Inc CANON Inc CANON Inc

Abstract:

In this paper, system identi?cation and

57
physical parameter estimation of semiconductor exposure apparatus for active vibration control are discussed. Since the semiconductor exposure apparatus has multi-degrees-of-freedom (multi-DOF) motional mechanism, the apparatus must be treated as a Multi-Input-Multi-Output (MIMO) system. In this paper, subspace based state-space system identi?cation (4SID) method is utilized, because it can be applied to an MIMO systems easily. E?ectiveness of the 4SID method to an MIMO system identi?cation problem is examined using experimental data. An estimation method of physical parameters of the semiconductor exposure apparatus based on the identi?ed models is also presented. By evaluating physical parameters of the system, fault detection and diagnosis can be done easily. Inertial elements can be estimated from transfer matrix of the identi?ed state-space model, and sti?ness and damping elements can be estimated by minimizing a cost function which consists of characteristic polynomials of the identi?ed state-space model and the physical model. Furthermore, a system identi?cation device is developed for modeling of the semiconductor exposure apparatus. It measures input and output data for identi?cation and executes system identi?cation algorithms based on MATLAB commands. weaknesses of each method with objective and subjective criteria. It emerges from this table that the linear ?ltering and modulating function based methods, along with two particular integral methods provide the best performances.

Robust Smith predictor design via uncertainty quanti?cation: application to a reclaimer Keum-Shik Hong Dong-Hunn Kang Jeom-Goo Kim Pusan National University, Korea Pusan National University, Korea Pusan National University, Korea

Abstract:

Experimental comparison of continuous-time model identi?cation methods on a thermal process M. Mensler H. Garnier E. Huselstein Kyushu University, Japan Ctr. Recherche Auto. de Nancy, France Ctr. Recherche Auto. de Nancy, France

The Smith predictor is a type of compensator that eliminates a time-delay occurred in industrial processes. Recently, the robustness in of the Smith predictor has been great issue. In this paper, a robust Smith predictor design is investigated. In this paper, the system identi?cation method is used to ?nd out a nominal model and its uncertainty bound of the nominal model is quanti?ed in the frequency domain through the uncertainty quanti?cation method. In addition, a robust criterion for the Smith predictor is also derived. With the obtained nominal model, uncertainty bound and the robust criterion derived, a robust Smith predictor is designed. This designed robust Smith predictor is applied to a reclaimer, which has a large output time-delay and is used in the raw yard of a steel plant. The performance of the designed robust Smith predictor is veri?ed through the experiments and the simulations.

Abstract:

Even if continuous-time model identi?cation methods are the object of growing interest, only few works have been conducted so far in order to evaluate them. In this paper, with the help of the CONTSID toolbox, a comparative study of seventeen continuous-time methods is carried out with real data obtained from a scale-bench hot air?ow device. The compared methods cover a large range of continuous-time methods, from the Linear Filters to Integral methods via Modulating Functions, for the Equation Error structure-based models. A nonlinear programming method based on the Levenberg-Marquardt algorithm is also considered for the Output Error structure-based models. A ?rst order and a second order models are used to identify the process. Moreover, some aspects of robustness of the methods against their design parameters are given. The conclusions are summarized under the form of a chart which exposes clearly and brie?y the forces and the

Nonlinear system identi?cation of rapid thermal processing Caizhong Tian Takao Fujii Osaka university Osaka university

Abstract:

Rapid thermal processing (RTP) is a key technology for single-wafer fabrication operations including defect annealing, oxidation, and chemical vapor deposition. The most common concern in manufacturing is thermal uniformity across a wafer. In this paper, identi?cation of rapid thermal processing is studied for ?nding a more accurate model to predict and control the temperature of semiconductor wafers. A nonlinear Wiener model is ?rst applied to identify the RTP system. A recursive method based on an Extended Kalman Filter is developed to estimate the model parameters and states simultaneously with respect to parameter variation in RTP system under process noise. The identi?cation result shows that the

58
model’s prediction error was greatly reduced as compared to a linear time invariant model. The proposed method can also be easily applied to model-based adaptive control. We hope that these e?orts will be use in obtaining signi?cant improvements in the performance characteristics of semiconductor manufacturing equipment.

Abstract:

Neuro-based modularized modeling and its application to deaerator with level control systems Yukihiro Toyoda Bailey Japan Co., Ltd., Kazushi Nakano Univ. Electro-Communications, Takami Matsuo Oita University, Kohji Higuchi Univ. Electro-Communications, Japan Japan Japan Japan

Abstract:

There is an increase in requests for speci?cations of simulators for checkout of distributed control systems (DCSs) that leads to a great demand for the hardware-in-the-loop (HIL) system based on physical models. Then, the following problems are pointed out:(1) the more computation time is needed for building higher accurate physical models, (2) high-priced interfaces which can process a huge number of input/output (I/O) signals in combination with DCSs are needed. To overcome these problems, it is necessary to develop a more practical and accurate simulator which can be embedded in controllers as DCS components. Furthermore, it is important to append the learning/identi?cation function to the controllers for reducing the tuning task of adjusting physical-model-based output responses with actual ones. In this paper, a two-step procedure for modeling large-scale processes called ”modularized modeling” procedure is presented. First, a low-dimensional modeling of each subsystem in the process is performed by using recurrent neural networks (NNs). Secondly, the NN-based identi?ed subsystems are integrated to complete a whole model. Through an example of modeling a deaerator with level control systems, the validity of our procedure is demonstrated. The obtained model is compact enough to be embedded in each controller and is utilized as the simulator for checkout of DCSs.

The topics of recovering of materials, quality control, and environmental impact have gained a raising attention in industrial activities, in order to enforce the quality parameters imposed by the new laws. The innovative approach of the intelligent methods to control and optimisation of the manufacturing process becomes necessary to cut the industrial costs. This paper deals with the development a Neuro-Fuzzy (NF) system for the control of corrosion potential value (AISI316) in Pulp & Paper production. Pulp & Paper production within the chemical process industry is a well established industrial activity in the European market. One of the problems in a chemical process is the impact of the chemical agents in the production machinery and in the ?nal product. Particularly, in a pulp mill corrosion phenomena may take place during the ozone-bleaching phase. Unexpected corrosion phenomena can actually modify the production equipment, causing sudden changes in the plant parameters. The aim of this work is to realise an expert system for the control of the corrosion phenomena in order to avoid unscheduled downtime, detriment of the product quality, and environmental damages. The performance of the NF expert system has been compared with that of the classical Neural Network MLPs (NN).

ThMD3

Automotive and Mechanical Applications Santa Ynez
Poster session: 1.30pm to 3.30pm

Continuous time hybrid identi?cation approach for on line tool wear estimation F. J. Carrillo F. Rotella M. Zadshakoyan LGP ENI Tarbes, France LGP ENI Tarbes, France LGP ENI Tarbes, France

Abstract:

Corrosion prediction in pulp and paper industry with neuro-fuzzy technique. Maide Bucolo Universit` degli Studi di Catania, Italy a Luigi Fortuna Universit` degli Studi di Catania, Italy a Martin Nelke Management Intell. Tech., Germany Alessandro Rizzo Univ. degli Studi di Catania, Italy Tatiana Sciacca Univ. degli Studi di Catania, Italy

The direct real time measurement of tool wear is an impossible task. This work presents the design of an adaptive observer for on-line ?ank wear estimation and tool life monitoring in machining using cutting force measurements. After a brief introduction resuming previous works in modeling and estimation for on-line ?ank wear, a nonlinear and an approximated linear wear models are described. From a reduced order version (controllable and observable) of the approximated linear wear model a continuous time hybrid identi?cation strategy is proposed. This approach is called hybrid

59
because the process model is continuous but the identi?cation algorithm is discrete with an observation vector composed of a linear combination of input-output (and from input-output process derivatives obtained by low pass ?ltering) measurements. To cope with measurement noise a recursive instrumental variable identi?cation algorithm was used. From the identi?ed parameters and using prior knowledge of the process model an adaptive wear observer (from cutting forces measures) is designed.Simulation results obtained using the nonlinear wear model combined with the adaptive wear observer (from the approximated linear wear model) were realized and shown to be in a good agreement between the computed ?ank wear model and the estimated one. of exhaust gas in the fresh air-fuel-mixture is too high, combustion degrades and even mis?res are possible. This results in erratic engine behavior, reduction of e?ciency and increased pollutant emissions. By analyzing cylinder-pressure valuable information on the combustion process and the quality of combustion can be gained. Therefore cylinder pressure of the engine was investigated in order to ?nd characteristic values which describe the quality of combustion and the engine smoothness. With the help of these characteristic values a fuzzy-control scheme was devised which optimizes EGR and thus minimizes emission of NOX without e?ecting quality of combustion and driveability adversely at the same time. The controller was successfully implemented in a test vehicle and examined under real life operating conditions.

Low order H∞ control design for a piezo-based milli-actuator Max Rotunno Raymond A. de Callafon Univ. California, San Diego Univ. California, San Diego Local linear model tree (LOLIMOT) for nonlinear system identi?cation of a turbocharger with variable turbine geometry Jochen Scha?nit Oliver Nelles Rolf Isermann Wolfram Schmid Darmstadt University of Technology Darmstadt University of Technology Darmstadt University of Technology DaimlerChrysler AG

The aim of the paper is to apply new techniques for low order or low complexity linear control design to servo controllers for milli-actuators in hard disk drives. This is accomplished in a systematic way by using a H∞ -norm based mixed sensitivity design in which the order of the controller is constrained to a desired value. The mixed sensitivity approach allows for an intuitive tradeo? between disturbance rejection and servo tracking in the hard disk dual-stage control design. The proposed method is applied to a prototype piezoelectric milli-actuator and for this case study a comparison is made between the full and low order control designs. With the approach outlined in this paper, it is illustrated that a low order controller for a milli-actuator can be designed with a satisfactory servo performance.

Abstract:

Abstract:

A new fuzzy logic approach to EGR control Karsten Spreitzer TU Darmstadt - IAT

This paper deals with nonlinear dynamic system identi?cation by local basis function networks. A special kind of local basis function network generated by a tree construction algorithm is proposed. This local linear model tree (LOLIMOT) is applied for identi?cation of a truck Diesel engine exhaust turbocharger with variable turbine geometry (VTG). The charging pressure is modelled as the output of a nonlinear second order multiple input system with engine speed, pulse width of injection and VTG control signal as inputs. The LOLIMOT approach was capable to identify the turbocharger with measured signals from engine test stand experiments and with ?fteen local linear models in less than one minute on a Pentium PC.

Abstract:

The aim of this contribution is to show that it is possible to optimize exhaust gas recirculation (EGR) systems on the basis of the engine smoothness. EGR is an e?ective way of reducing nitrogen oxide emissions by mixing exhaust gas with fresh air-fuel-mixture. Nitrogen oxide emissions can be reduced by up to 60 % when EGR is implemented. However, recirculating too much exhaust gas can be hazardous to engine operation. If the amount

Identi?cation of spark ignition engine models based on neural network via experimental design techniques Ivan Arsie Fabrizio Marotta Cesare Pianese Gianfranco Rizzo University University University University of of of of Salerno, Salerno, Salerno, Salerno, Italy Italy Italy Italy

60 Abstract:
An Experimental Design technique applied for the training of Arti?cial Neural Networks is presented. The methodology is oriented to maximize the information achievable during experimental tests by guiding them in a structured way and to avoid the overtraining and the over?tting of the Network. The adopted technique is known as Active Learning and is based on the selection of the most signi?cant training examples from an existing set of experimental data. Each experiment, feeding the Neural Network, is chosen as the one which brings the highest content of information through the maximization of an information functional based on the Shannon’s Entropy concept. The results for the estimation of the Mechanical E?ciency and the Load Torque for a Spark Ignition Engine exhibit high precision levels with a lower number of experiments with respect to the use of conventional strategies for selecting the training data set. Keywords: Automotive Engine Control, Neural Network, Experimental Design, Active Learning. ThMD4

Identi?cation Applications
Poster session: 1.30pm to 3.30pm

Santa Ynez

An adaptive clustering method of the cross-sectional mean void fraction signals of gas-liquid two-phase ?ow Katsuhiro Inoue Kyushu Inst. Technology, Japan Kousuke Kumamaru Kyushu Inst. Technology, Japan Kotohiko Sekoguchi Polytechnic Coll., Kagawa, Japan

Abstract:

Preserving stability/performance when facing an unknown time-delay S. Diop Supelec, France I. Kolmanovsky Ford Research Lab., Michigan P. Moraal Ford Forshugszentrum Aachen, Germany M. van Nieuwstadt Ford Research Lab., Michigan

The adaptive clustering algorithm has been constructed by using the autoregressive model with bias component to analyze the cross-sectional mean void fraction signals in gas-liquid two-phase ?ow. The algorithm was established in recursive form (parameter estimation) with new cluster creation rule. However it was very complicated that the recursive form for parameter estimation based on the mixture probability algorithm with a maximum likelihood estimates. So in this paper, the algorithm based on the mixture probability algorithm is modi?ed and new algorithm based on a stochastic Newton method is proposed. The e?ectiveness of the methods is con?rmed by applying to analyzing the cross-sectional mean void fraction signals, and determination of the new boundaries on a super?cial gas-liquid velocities diagram. The method based on a stochastic Newton method seems to be more e?ective than the method based on the mixture probability algorithm, although the later one is more complete than the former one theoretically. Simpli?cation of the mixture probability algorithm and theoretical rearrangement of the stochastic Newton method are under investigations.

Abstract:

Embedded controllers executed in real-time are, frequently, subject to a time-varying delay induced by task prioritization or communication over prioritized communication networks. Depending on the microprocessor or network load the delay value may vary. The control design that is based on the worst case assumption with respect to the delay may be very conservative and fail to deliver the adequate performance. On the other hand,


更多相关文章:
CFA知识点-Corporate Finance 公司金融
全球最大的 CFA(特许金融分析师)培训中心 CFA 知识点-Corporate Finance 公司金融 Corporate Finance, 中文通常译作公司金融,或者公司理财。在 CFA 一级考试中,它...
通达信财务指标公式
{财务指标公式源码} 年净资产收益率:FINANCE(30)/FINANCE(19)*100*(4/FINANCE(37)),COLORWHITE; 报酬率:FINANCE(34)/FINANCE(33)*100,NODRAW; 净资产收益率...
财务中英文对照
(亏损) Profit from operating activities 营业利 润/经营活动之利润 Finance costs 财务费用/财务成本 Financial result 财务费用 Finance income 财务收益 Change in...
第二章 金融体系预览(英文习题及答案)
3) Which of the following can be described as involving direct finance? A) A corporation's stock is traded in an over-the-counter market. B) People...
The Cost of Capital, Corporation Finance and the Th...
The Cost of Capital, Corporation Finance and the Theory of Investment 全文翻译_金融/投资_经管营销_专业资料。资本成本、公司财务与投资理论 ...
国外会计&金融英文学术期刊排行
Auditing and Finance 1B Rank 1A FINANCE Journal of Finance Journal of Financial Economics Review of Financial Studies Journal of Financial and Quantitative An...
LSE金融相关专业申请(由难到易)
~ Portfolio Management* Quantitative Methods for Finance and Risk Analysis* Real Estate Finance* Valuation and Security Analysis* Law of International Economic...
Finance专业设置及申请技巧
Finance专业设置及申请技巧_企业管理_经管营销_专业资料。多样化的美国金融硕士 一、MBA 下的金融 MBA in Finance 是美国大学金融专业设置的主要形式之一。因为设在...
Finance(国际金融)关键术语名词解释_图文
Finance(国际金融)关键术语名词解释_金融/投资_经管营销_专业资料。Finance(国际金融)关键术语名词解释 Chapter 1 《American Economic Review》 《Journal of Finance...
公司理财知识点整理重点corporate finance
公司理财知识点整理重点corporate finance_经济学_高等教育_教育专区。第一章 1 三大理财活动 总论 2 公司理财的目标(选择判断) 1、利润最大化 缺陷: 没有考虑...
更多相关标签:

All rights reserved Powered by 甜梦文库 9512.net

copyright ©right 2010-2021。
甜梦文库内容来自网络,如有侵犯请联系客服。zhit325@126.com|网站地图