Department of Systems and Control

Jožef Stefan Institute

System Identification Based on Gaussian Process Model for Traffic Control Applications (bilateral project with Czech republic)

Principal investigator:

Prof. dr. Juš Kocijan

Duration:

January 1, 2010 - December 31, 2011

Funding:

ARRS - Slovenian Research Agency

Abstract:

Many engineering systems can be characterised as complex since they have a nonlinear behaviour incorporating a stochastic uncertainty. Urban traffic systems or pollution propagation models are typical representatives of such complex systems. Modern traffic management systems, that handle urban traffic organisation, traffic safety or environmental impacts are based on models describing the current traffic situation in the region to be managed.

Queue length is one, but not the only, variable that frequently needs to be modelled for developing an efficient traffic management systems due to the insufficient number of available sensors. The models of these traffic-related variables (like queue length, delay or stop-time) can be of various kinds, first principles models, statistic models or black-box models trying to predict the variables of interest based on known training data. The same applies for pollution models, where pollutant dispersion in the environment is typically assessed by the means of first principles models.

The main drawback of most of the above mentioned methods is that they try to describe the dynamic process of the variables of interest in a deterministic way or they ignore the uncertainty component in the measured data and model representation. Such things cannot be neglected in the case of modelling of complex systems as they can often lead to unreliable predictions.

One of the most appropriate methods for modelling of systems that are complex and contain a great deal of uncertainty is based on the application of Gaussian processes. Gaussian process models provide a Bayesian probabilistic non-parametric modelling approach for black-box identification of nonlinear stochastic systems. The Gaussian processes can highlight areas of the input space where prediction quality is poor, due to the lack of data or to its complexity, by indicating the higher variance around the predicted mean.

The aim of this project is to show the benefits of predicting the modelled uncertain variables in urban traffic or pollution models using a probabilistic nonlinear Gaussian process models and solving some issues that are related to the application of this method.

Publications:

PETELIN, Dejan, ŠINDELÁŘ, Jan, PŘIKRYL, Jan, KOCIJAN, Juš. Financial modeling using Gaussian process models. IDAACS'11: proceedings of the 6th IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, 15-17 September, 2011, Prague, Czech Republic. 2 Vol, Piscataway: IEEE, 2011, vol. 1, 672-677.

PŘIKRYL, Jan, KOCIJAN, Juš. Stochastic analysis of a queue length model using a graphics processing unit. Transactions on Transport Sciences, 2012, vol. 5, no. 2.

PŘIKRYL, Jan, KOCIJAN, Juš. An empirical model of occupancy-queue relation. Proceedings of the 12th IFAC Symposium on Transportation Systems, Redondo Beach, CA, IFAC, 2009, 456-461.

KOCIJAN, Juš, PŘIKRYL, Jan. Soft sensor for faulty measurements detection and reconstruction in urban traffic. Proceedings 15th IEEE Mediterranian Electromechanical Conference, MELECON 2010, 25-28 April, 2010, Valletta, Malta, [Piscataway]: IEEE, 2010, 172-176.