Prediction of Short-Term Traffic Flow Based on Similarity

Abstract

To improve the precision of short-term traffic flow prediction and to enhance the accuracy of programming as well as of traffic flow management, a novel short-term traffic flow prediction method based on similarity is proposed in this study. The similarity observed at a single point on the California expressway is examined, and the similarity on the same day for four adjacent weeks is higher than that on four adjacent days. The wavelet neural network (WNN) is established on this basis; moreover, the traffic flow data regarding the same day for four adjacent weeks and regarding the four adjacent days are divided into two types. Then, more than 200 groups of data are used to train the WNN and to predict the traffic flow on the same day. Results indicate that the mean values of the mean relative estimation error (MRE), mean square percentage error (MSPE), and equalization coefficient (EC) as predicted by the first methoo are 8.55%, 1.32%, and 0.951 6 respectively; the corresponding mean values obtained with the second method are 13.80%, 3.71%, and 0.916 8. The MRE and MSPE values generated with the first method are lower than those obtained with the second method; by contrast, the EC value of the first method is higher than that of the second method. This finding suggests that the prediction accuracy of the first method is higher than that of the second method. Accordingly, the effectiveness of the proposed method is verified.