Interest in neural networks, a concept first described in 1943, has grown rapidly with the development and availability of computer processing power. A large number of different neural network paradigms have been developed particularly for the solution of prediction and classification problems across a wide range of subjects. In traffic engineering neural networks have been applied primarily for traffic pattern recognition alongside image processing.

Recently the possible application of neural networks to traffic modelling has been considered. Fix and Armstrong demonstrated that a neural network could be trained to "drive" a car within a simulation model. This suggests that neural networks could be developed and trained to provide a more realistic and accurate representation of driver behaviour and response in road traffic situations.

Typically a road traffic simulation model would use car following and gap acceptance theory or empirical equations to determine driver decisions. In practice a driver would make decisions based on his perception of the immediate driving environment. In order to examine the potential of a "neural driver" a training model has been developed for the movement of traffic along a straight uninterrupted section of motorway.

In the paper the advantages and disadvantages of this approach to traffic representation are considered, for this particular model and more generally. Particular attention is given to the difficulties of training the model and to subsequent calibration and validation.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £34 +P&P)