Department of Systems and Control

Jožef Stefan Institute

Application of Gaussian processes to the modelling and control of complex stochastic systems (bilateral cooperation with Bulgaria)

Principal investigator:

Prof. dr. Juš Kocijan

Duration:

January 1, 2009 - December 31, 2010

Funding:

ARRS - Slovenian Research Agency

Abstract:

Many engineering systems can be characterised as complex since they have a nonlinear behaviour incorporating a stochastic uncertainty. It has been shown that one of the most appropriate methods for modelling of such systems is based on the application of Gaussian processes. The principal objective of this project is the further development of the methods for modeling and optimal control of complex systems based on Gaussian processes and their application to specific stochastic systems. This includes the fulfillment of the following tasks. A Gaussian process model for prediction of ozone concentration in the air in a particular region in Bulgaria will be developed. Gaussian processes will be applied to obtain an updated dynamic model of the pilot gas-liquid separator at the Jozef Stefan Institute (the separator has been reconstructed and experiments have shown that its dynamics has changed). Recently, an approximate multi-parametric Nonlinear Programming approach to explicit solution of stochastic model predictive control (MPC) problems for constrained nonlinear systems based on a Gaussian process model has been developed. However, the off-line computational complexity with this approach tends to increase very rapidly with the number of states. Another purpose of this project is to develop parallel computing algorithms to improve the off-line computational efficiency of the methods for design of explicit stochastic MPC controllers. The developed algorithms will be applied to design an explicit stochastic MPC controller for the pilot gas-liquid separator. The obtained results will be published in prestigious international journals and international conferences.

Publications:

KOCIJAN, Juš, GRANCHAROVA, Alexandra. Gaussian process modelling case study with multiple outputs. Comptes Rendus de l Academie Bulgare des Sciences, 2010, vol. 63, no. 4, 601-607.

GRANCHAROVA, Alexandra, KOCIJAN, Juš, JOHANSEN, Tor Arne. Explicit output-feedback nonlinear predictive control based on black-box models. Engineering applications of artificial intelligence. [Print ed.], 2011, vol. 24, no. 2, 388-397.
PETELIN, Dejan, KOCIJAN, Juš, GRANCHAROVA, Alexandra. On-line Gaussian process model for the prediction of the ozone concentration in the air. Compt. rend. Acad. bulg. Sci, 2011, vol. 64, no. 1, 117-124.

GRANCHAROVA, Alexandra, NEDIALKOV, D., KOCIJAN, Juš, HRISTOVA, H., KRASTEV, A. Application of Gaussian processes to the prediction of ozone concentration in the air of Burgas. IN: International conference Automatics and Informatics '09, Bulgaria, Sofia, 29.09 -4.10.09. Proceedings : John Atanasoff celebration days. Sofia: Union of Automation and Informatics, 2009, pp. IV-17-IV-20.

GRANCHAROVA, Alexandra, KOCIJAN, Juš, KRASTEV, A., HRISTOVA, H. High-order Gaussian process models for prediction of ozone concentration in the air. IN: ŠNOREK, Miroslav (ed.). EUROSIM 2010 : proceedings of the 7th EUROSIM Congress on Modelling and Simulation, September 6-10, 2010, Prague, Czech Republic. Vol. 2, Full papers (CD). Prague: Czech Technical University in Prague, 2010, 8 pages.

GRANCHAROVA, Alexandra, KOCIJAN, Juš, JOHANSEN, Tor Arne. Dual-mode explicit output-feedback predictive control based on neural networks models. IN: 8th IFAC Symposium on Nonlinear Control Systems, September 01-03, 2010, Bologna, Italy. NOLCOS 2010. IFAC, 2010, pp. 545-550.

GRANCHAROVA, Alexandra, KOCIJAN, Juš. A parallel computing algorithm for design of explicit nonlinear model predictive controllers. IN: International conference Automatics and Informatics '10, Bulgaria, Sofia, 3.10 -7.10.10. Proceedings : John Atanasoff celebration days. Sofia: Union of Automation and Informatics, 2010, pp. I-233 - I-236.