Real-time traffic flow forecasting model and parameter selection based on ε-SVR

Accurate traffic flow forecasting is key to the development of intelligent transportation systems (ITS). The support vector regression (SVR) method is employed for traffic flow forecasting and the comparative results between SVR and BP model using real traffic data of SCOOT system in Dalian city is also presented in this paper. Since support vector machines have better generalization performance and can guarantee global minima for given training data, it is believed that SVR will perform well for real-time traffic flow forecasting. However, the good generalization performance of SVR highly depends on good parameter selection (PS). This paper describes simple yet practical approach to SVR parameter selection directly from the training data. Experimental and analytical results demonstrate the feasibility of applying SVR to traffic flow forecasting and prove that the SVRpsilas parameter selection can better satisfy real-time demand of traffic flow forecasting and has good practicability.