THE USE OF NEURAL NETWORKS TO RECOGNISE AND PREDICT TRAFFIC CONGESTION

This paper shows how a new method of computing called "neuro- computing" (neural networks) can assist in the following types of pattern recognition problem: a) the problem of recognising that the road system is in a particular state of congestion; and b) the problem of real time short term forecasting. Section 2 describes the main features of a neural network approach. Trials of its application to: a) congestion recognition; and b) to short term forecasting of traffic flows on an urban network are presented in sections 3 and 4 respectively. A method for "training" neural networks via a simulation package to infer parameters that are not directly measurable is then suggested. The results indicate that although much work still needs to be done, neural network technology can provide a powerful method of analysing, interpreting and predicting complex data sets.