Abstract:

In order to realize the real-time traffic flow state estimation of the regional freeway network in cities,a macroscopic traffic flow state space model is constructed.This model helps to estimate the traffic flow states and up- date the model parameters,and it can improve the adaptability and accuracy of the traffic flow model.Then,the SVD (Singular Value Decomposition)-based optimized robust UKF (Unscented Kalman Filter) algorithm is pro- posed.In the algorithm,the singular value decomposition is adopted to replace the Cholesky decomposition,thus solving the problem that the filtering can't continue when the covariance matrix is non-positive.Meanwhile,differ- ent strategies are chosen according to whether the observation covariance matrix is pathological,and both the gain matrix and the observation covariance matrix are adaptively calculated.Furthermore,the error caused by the high nonlinearity of the constructed model is inhibited.Experimental results show that the proposed algorithm can avoid the filtering divergence of the EKF (Extended Kalman Filter) algorithm and can accurately track the trend of the traffic flow,thus improving the stability and precision of the traffic flow state estimation.

Key words: traffic flow state space model, UKF algorithm, singular value decomposition, error resistance factor