Short-Term Traffic Flow Prediction and Its Application Based on the Basis-Prediction Model and Local Weighted Partial Least Squares Method

Abstract

Traffic flow forecasting is one of the key issues in smart traffic systems. The changing process of traffic flow involves high randomness, environmental interference and measurement noise, which bring difficulties to accurate traffic flow prediction. Aiming at improving the accuracy of short-term traffic flow prediction, this paper presents a basis-prediction model. A raw traffic flow series can be deemed as summation of a basis series that implies the changing trend of the traffic flow and a deviation series which represents the random interference information involved in the flow. The basis series mainly comprises low-frequency signals and some high-frequency ones compose the deviation series. The basis series and the deviation series can be obtained using wavelet decomposition of the raw traffic flow. The local weighted partial least squares (LW-PLS) method is adopted to predict the basis series and the result is used as the prediction of the raw traffic flow. Real data of traffic flow of Xinbei city, Taiwan province was collected and used for validation of the proposed basis-prediction model. The results show that the use of that model improves the accuracy of short-term traffic flow prediction by about 2% on average.