Artificial Intelligence-Neural Network Methods

This chapter deals with applications of artificial intelligence and more particularly of artificial neural network (ANN) methods for transport demand. First, the structure of a biological and of an artificial neuron, the fundamental functions of an ANN, analogies between artificial and biological neurons, and the activation function of an ANN are explained. The various types of ANN, algorithms, and software are presented afterward. The structure of an ANN (input layer, output layer, hidden layer), the propagation rule (feedforward, backpropagation, recurrent), the method of supervision of learning (supervised learning, reinforced learning, nonsupervised learning), and the learning rule (error correction, Hebbian, Boltzman, Perceptron) are extensively analyzed. Similarities and differences of ANN and statistical and econometric methods are surveyed: assumptions, mechanism and processes, nonlinearities, collinearities, fluctuations of data. A step-by-step analytical example of an application of ANN for the long-term forecast of air transport demand is scrupulously studied and explained: architecture of the ANN model, single and multiple inputs, data and software, propagation rule (feedforward, recurrent), number of iterations, errors. Another application of ANN for the analysis of rail demand is explained in detail. Applications of ANN for the analysis of other sectors of transport are also discussed: road traffic, road safety, driver behavior, self-driven vehicles, freight transport, maintenance needs, performance and quality of service, effects of unpredicted events, and transport economics.