Dec. 20 2005 4:00PM

Dec. 20 2005 4:20PM

This work is concerned with application of the ongoing research in the area of multiple-participant decision-making. Specifically, decision making (i.e. control) of traffic flow in urban areas is considered. Both areas, i.e. multiple
participant decision making and the problems of traffic control, has already been presented, hence we address the detailed problems of distributed decision making in this domain. The problem of urban traffic control is very suitable for distributed control, since there exist a well justified decompositions of the controlled system. A lot of work on this topic has already been presented in the area of multi-agent approach to traffic control. Recall, that in this domain, the main means of control are the ratios of periods of green light on junctions. Hence, the basic decision-making unit is the controller of green light in a single junction. Notably, this decomposition is used in the current control systems, where the signal plans are developed for a single junction and then synchronized between junctions. The challenge is to achieve this in an automatic (and optimized) way. In order to achieve this goal, we will use the approach developed in the area of multiple-participant decision making, namely fully probabilistic design accompanied by merging of probability distributions. Recall, that the technique of FPD yields optimal decision making strategy from the probabilistic models of the system and its desired behaviour. The task of application of the general theory is than reduced to proper mathematical modelling of the system. Specifically, we need to select the variables of interest, their desired values and their probabilistic modelling. Success of the approach is heavily dependent on correct parameterization of the traffic system. We will present a proposition of such modelling. The probabilistic description of the underlying real system is based on state-space approach. Description of the desired behaviour, i.e. the ideal distributions, are defined on the same state variables and observed variables. Further development of this proposition is dependent on verification by software simulation and comparing with the real system.

International Workshop on Data - Algorithms - Decision Making (Contact person: Martin Janžura)