The Study on a Real-time Forecasting Model for Short-term Traffic Flow Based on Online Incremental LSVR

Accurate prediction of real-time short-term traffic flow is essential in Intelligent Transportation System (ITS) and is the foundation of urban traffic control and guidance. This paper proposes a new method Online-learning Lagrange Support Vector Regression (Online-learning LSVR), which was adopted in real-time prediction for short-term traffic flow. The algorithm of Online-learning LSVR proposed in this paper and conventional LSVR were compared and evaluated using an urban traffic flow data set. Experimental results were compared showing the improved LSVR not only realized online-learning but also used fewer Support Vectors (SVs) and much less forecasting time without decreasing prediction accuracy.