Traffic Flow Prediction via Weighted Combination of ARIMA and WASDNN Models

The modeling and prediction of traffic flow are important in transportation study and traffic network management. Different from individual models, a hybrid model which is a combination of autoregressive integrated moving average (ARIMA) and weights and structure determination neural network (WASDNN) models, is proposed for modeling and prediction of traffic flow. Compared with individual models like ARIMA and WASDNN, the proposed ARIMA-WASDNN model has better performance in modeling and prediction of traffic flow of Shenzhen city as an example. In addition, prediction results of different activation functions of hidden layer neurons are provided. What is more, the proposed ARIMA-WASDNN model also performs better than the popular long short-term memory (LSTM) model in predicting traffic flow of Shenzhen city. The effectiveness and superiority of ARIMA- WASDNN model are substantiated by results of numerical experiments.