A noise-immune Kalman filter for short-term traffic flow forecasting

Abstract

This paper formulates the traffic flow forecasting task by introducing a maximum correntropy deduced Kalman filter. The traditional Kalman filter is based on minimum mean square error, which performs well under Gaussian noises. However, the real traffic flow data are fulfilled with non-Gaussian noises. The traditional Kalman filter may rot under this situation. The Kalman filter deduced by maximum correntropy criteria is insensitive to non-Gaussian noises, meanwhile retains the optimal state mean and covariance propagation of the traditional Kalman filter. To achieve this, a fix-point algorithm is embedded to update the posterior estimations of maximum correntropy deduced Kalman filter. Extensive experiments on four benchmark datasets demonstrate the outperformance of this model for traffic flow forecasting.