Forecasting Approach for Short-term Traffic Flow based on Principal Component Analysis and Combined Neural Network

A combination approach based on Principal Component Analysis (PCA) and Combined Neural Network (CNN) is presented for short-term traffic flow forecasting. The historical data of the forecasted traffic volume and interrelated volumes have been processed by PCA. The results of PCA form the input data for CNN. It not only reduces the dimension of input variables and the size of CNN, but also reserves the main information of the original variables and eliminates relativity among them. An example for explanation of validity is given. The forecast results show that this approach is better than the typical Error Back-Propagation neural network (BP NN) with the same data.