Neural network structure optimization and its application for passenger flow predicting of comprehensive transportation between cities

The modeling and predicting for passenger flow of comprehensive transportation between cities are studied Passenger flow for different transportation mode is concerned with both social and economic characteristics of passengers, and also concerned with personal preference of every passenger. This is a modeling and predicting problem for complex systems. First, a 3-layer neural network is structured according to Kolmogorov theorem, which is a nonlinear system model with p input variables and n output variables. Second, the optimizing objective function is built with AIC criterion based on Darwin principle that is struggle for existence and survival of the fittest. The third, both the neural network structure and its parameters are obtained simultaneously using Genetic Algorithms, in which the fitness is taken as 1/AIC and both dynamic adaptive crossover rate and mutation rate are used. Therefore, the 3-layer neural network with p:m:n structure is gotten, which represents passenger flow prediction model for comprehensive transportation system between cities. Finally, the computation example shows that the higher prediction precision and faster convergence speed can be obtained using the model in the paper.