Traffic Volume Forecasting Model of Freeway Toll Stations During Holidays – An SVM Model

Xiaowei Hu
Harbin Institute of Technology, School of Transportation Science and Engineering

Yongzhi Xiao
Harbin Institute of Technology, School of Transportation Science and Engineering

Tianlin Wang
Harbin Institute of Technology, School of Transportation Science and Engineering

Lu Yang
Harbin Institute of Technology, School of Transportation Science and Engineering

Pengcheng Tang
Chinaroads Communications Science & Technology Group CO., Ltd.

Abstract

Support vector machine (SVM) models have good performance in predicting daily traffic volume at toll stations, however, they cannot accurately predict holiday traffic volume. Therefore, an improved SVM model is proposed in this paper. The paper takes a toll station in Heilongjiang, China as an example, and uses the daily traffic volume as the learning set. The current and previous 7-day traffic volumes are used as the dependent and independent variables for model learning, respectively. This paper found that the basic SVM model is not  accurate enough to forecast the traffic volume during holidays. To improve the model accuracy, this paper first used the SVM model to forecast non-holiday traffic volumes, and proposed a prediction method using quarterly conversion coefficients combined with the SVM model to construct an improved SVM model. The result of the prediction showed that the improved SVM model in this paper was able to effectively improve accuracy, making it better than in the basic SVM and GBDT model, thus proving the feasibility of the improved SVM model.