Short Term Traffic Flow Prediction Based on Online Learning SVR

Abstract

A number of different forecasting methods have been proposed for traffic flow forecasting including historic method, real-time method, time series analysis, and artificial neural networks (ANN), but accuracy and time efficiency in prediction are a couple of contradictions to be hard to resolve for real-time traffic information prediction. In order to improve time efficiency of prediction, a new short-term traffic flow prediction model and method based on accurate online support vector regression (AOSVR) is proposed in this paper, which can update the prediction function in real time via incremental learning way. A comparison of the performance of AOSVR with ANN, real time, and historic approach is carried out. Experiments results demonstrate that the AOSVR predictor can reduce significantly both relative mean errors and root mean squared errors of predicted travel times. Therefore, AOSVR based traffic flow prediction is applicable and performs well for traffic data analysis.