Kalman Filter Using SOV Model with Maximum Versoria Criterion for Short-Term Traffic Flow Forecasting

Abstract

This paper proposes a prediction method by combining second-order Volterra (SOV) model and Kalman filter to further improve prediction accuracy of the traditional Kalman model in short-term traffic flow forecasting. Nonlinear relationship may exist in traffic flow data, but the traditional Kalman model cannot deal with this problem. Due to the second-order Volterra (SOV) filter can deal with a general class of nonlinear systems, the traditional Kalman combines with second-order Volterra model, named SOV-KF model, is presented. Furthermore, since the Gaussian assumption is not always be fulfilled in the traffic flow data and traditional minimum mean square error (MMSE) criterion do not perform well under non-Gaussian noises. By introducing maximum Versoria criterion, another prediction method called SOV-MVKF model is also proposed. Simulation results show that the SOV-KF model and SOV-MVKF model provide higher prediction accuracy compared to traditional Kalman model.