Short-term Traffic Flow Prediction Based On Nonlinear Filters

With the increase of car ownership,some problems such as traffic congestion ensue.And intelligent transportation systems play an important role in alleviating traffic congestion with the development of technology.Fast and accurate traffic flow prediction is a fundamental and important part of intelligent transportation systems.Because the Volterra adaptive filter can predict chaotic time series,traffic flow is a chaotic time series,which can be predicted by Volterra adaptive filter.The main work of the article are follows:(1)After data preprocessing such as data repair and normalization of the actual traffic flow,according to the theories and methods of phase space reconstruction,the autocorrelation method and the false neighbor method are used to calculate the delay time and the embedding dimension,respectively.(2)According to the calculated delay time and the embedded dimension,the input vector and output value of the system can be constructed.The second or third-order truncated Volterra structure is used to construct the Volterra filter because of the nonlinear fitting characteristics of the Volterra series.The most commonly used least mean square algorithm(LMS)is used to predict the actual traffic flow,and has a good performance.(3)The maximum correntropy criterion algorithm(MCC)based on the correntropy has been derived because of the abnormal data in data collecting in this article.The robustness of the algorithm has been analyzed according to the formula in theory.Moreover,three noises with values of 1,10,100 are added to the traffic flow data to simulate the actual situation.The experiments show that MCC algorithm has a better performance than LMS algorithm in accuracy and robustness.The generalized maximum entropy algorithm(GMCC)has been presented in order to further improve the prediction accuracy and maintain the robustness.The experiments show that GMCC algorithm has a better performance than MCC algorithm.(4)The experiments show that MCC algorithm and GMCC algorithm have some advantages than LMS algorithm in accuracy and robustness.However,the curves of the two algorithms are unstable and the convergence speed is slow.Therefore,the minimum error entropy(MEE)algorithm has been presented in this article.The experiments show that the learning curve of the MEE algorithm is more stable,the convergence speed is faster can be achieved.