Applications of Artificial Neural Networks in Transportation

Abstract

In the 1990s over fifty papers were published applying neural network models to transportation problems (Dougherty, 1995). Most of these papers were concerned with road transportation. Yang et al. (1992) and Dougherty and Joint (1992) modeled the driver’s behavior when making strategic and instinctive decisions. These authors use neural networks to analyze data gathered in the following way: volunteer drivers who took part in the experiments chose the routes by comparing the values of different criteria. The collected data were used to train neural networks. The trained networks successfully reproduced the drivers’ decisions. When the developed networks were applied to new data, they produced good results more quickly and accurately than alternative techniques like logit models. Many authors have studied the simulation of the driver’s behavior while driving a vehicle. For example, Hunt and Lyons (1994) use neural networks to model the driver’s behavior while changing the speed and lane on a highway. A developed neural network is regarded as a “neuro driver” who maneuvers the chosen vehicle following input signals (relative position of the vehicle surrounded by other vehicles in particular time intervals) from the near vicinity. The basic disadvantage of this procedure is the problem of collecting real data necessary to train the network. If the data are recorded from one driver only, the neural network displays driving characteristics similar to that driver. A survey should be conducted to determine the set of representative drivers and their driving should be recorded.