Travel Time Prediction Method Based on Improved Kalman Filter

Abstract

Travel time prediction as an important foundation to traffic management and route guidance, is an important content of intelligent traffic travel information system. Based on the analysis of various travel time prediction methods, in this paper, travel time prediction model is developed by using Kalman filter. For standard Kalman filter method, the system noise covariance and the observation noise covariance matrices are assumed as constant, which may compromise the accuracy of forecasting results. An improved noise covariance adjustment method based on the innovation vector is proposed in this paper. This method can adjust the Kalman filter noise statistical information online according to the real-time data. Simulation experiments show that the improved Kalman filter method can generate a mean absolute relative error under 10%, speed up the convergence, make the prediction results more accurate, and well meet other requirements of the traffic guidance system, when comparing with the standard Kalman filter method.