This chapter considers an approach to distributed traffic responsive signal control using Learning Classifier Systems (Holland, 1976). The intention is to accommodate realistic kinds of detector data and wide ranges of candidate performance criteria for traffic management in a fully flexible manner. The approach to achieving this is to use evolutionary computing (eg Holland, 1975) and reinforcement learning (eg Sutton and Barto, 1998) with performance fed back from microscopic traffic simulations: this approach has the advantage that it is not specific to any particular objective or form of primary data. The purpose of this work is to develop an approach to distributed optimisation that can achieve good traffic performance flexibly according to any on a range of possible criteria using data from existing traffic detectors. Here each junction in a road network is controlled by a Learning Classifier System using only locally available input and performance data; a multi-agent approach is proposed. Learning Classifier Systems (LCS) can be used for optimisation in a way that offers substantial promise for application in traffic-responsive signal control systems where the way in which the control responds to variations in traffic flows can be adapted according to measured conditions. This is important in order to achieve traffic control that is sufficiently flexible to respond rapidly when traffic conditions change in a fundamental way, as occurs at the start of a peak period, without being unduly sensitive to short-term variations in flow. The expectation is that this will be possible by their use of both reinforcement learning and evolutionary computing techniques. Furthermore, they offer the automated rule development of neural networks together with the transparency of production system rules. The importance of this approach for traffic control is that it offers a means by which signal control strategies can be developed directly according to their performance, evaluated using detailed microscopic simulation as opposed to that estimated from formulae that have been adopted on grounds of analytical convenience. This closed-loop approach to development of control strategies offers several advantages over the use of traditional explicit optimisation formulations. These include flexibility in respect of objectives so that multiple and varying needs can be accommodated, ability to use various different kinds of detector data according to their availability, and freedom from dependence on a single explicit evaluation formula that is intended to embody the whole of a traffic model. This final point has been found to be especially important in recent research work where certain fine details of the models used have been found to have an unexpectedly strong influence on performance. (Author/publisher)
Abstract