Fuzzy Neural Network System for Urban Expressway Speed Prediction on Rainy Days

Abstract:

A fuzzy neural network system was developed to improve urban expressway shortterm speed prediction accuracy on rainy days, taking fuzzy influencing factors such as traffic volume, occupancy and precipitation, as well as their nonlinear interaction into account. Based on the traffic flow and weather data of Shanghai, the best model structure was determined and its performance was evaluated against those of the existing autoregressive integrated moving average model, the back propagation neutral network, and the support vector machines model. The results show that the root mean square error and mean absolute percent error of the fuzzy neural network system are 3.05 km?h-1 and 3.95% respectively, which outperform those of the other three prediction models.