Real-time Traffic Flow Forecasting Based on MW-AOSVR

Accurate traffic flow forecasting is the key to the development of intelligent transportation systems (ITS). However, the classical forecasting method using the support vector regression (SVR) based on RBF kernel does not support online learning and has the problems of information loss, long processing time, low robustness and so on. An effective Marr Wavelet kernel which we combine the wavelet theory with AOSVR (MW-AOSVR) to construct for traffic flow forecasting is presented in this paper. The forecasting performance of MW-AOSVR is evaluated by real-time traffic flow data of southbound US 101 Freeway, in Los Angeles, USA and a variety of experiments are carried out. The experimental results demonstrate that the proposed approach with Marr Wavelet kernel provides more optimal performance than that with radial basis function (RBF) kernel and has much more precise forecasting rate and higher efficiency, especially for boundary approximation.