Research On Short-term Trffic Status Prediction Of Freeway Network

With the development of our society,the freeways have been an important guaranty for rapid economic development of a nation.Meanwhile,the traffic safety and congestion of the freeways are also becoming worldwide problems.The intelligent control and real-time guidance of intelligent transportation system(ITS)can effectively alleviate the traffic congestion,reduce the environmental pollution and provide more secure traffic conditions.However,the premise and key to achieve these goals is to predict the short-time traffic status accurately.Therefore,the thesis researches on the problem of short-term traffic status prediction of freeway network and divides it into two sub-problems,which are short-term traffic parameters prediction using historical traffic data and traffic status identification using predicted traffic data.Thus,the short-term traffic status can be predicted indirectly.The thesis firstly researches on the issue of short-term traffic parameters prediction.The previous works applied a number of models and algorithms based on time series prediction and machine learning to short-term traffic flow prediction,but these static models require the length of the input history data to be predefined and static and cannot automatically determine the optimal time lags,which leads to the unsatisfying results.To overcome the shortage,a short-time traffic parameters prediction model based on long short-term memory recurrent neural network(LSTM RNN)is proposed in this thesis,which takes the advantages of the three multiplicative units in the memory block to determine the optimal time lags dynamically,which leads to better prediction results.Dataset from Caltrans Performance Measurement System(PeMS)is used for building the model and the comparison results between LSTM RNN and several classical prediction models show that the proposed model achieves a higher accuracy,generalizes well and is capable to memorize long historical data.On the basis of traffic parameters prediction model,the thesis researches on the issue of traffic status identification.Due to the fact that traffic data are commonly seriously affected by abnormal noise,the methods in previous works based on traffic flow theory and pattern recognition cannot effectively resist the noise,which results in undesirable identifications.Therefore,a traffic status identification model based on possibilistic fuzzy c-means(PFCM)clustering is proposed in the thesis.The model can identify the noise effectively by means of the typicality matrix introduced in PFCM.Dataset from PeMS is used to build the model and make comparison with classic FCM identification model in order to verify the identification accuracy and the capability of noise-resistance of the proposed model.The short-term traffic parameters prediction model based on LSTM RNN and the traffic status identification model based on PFCM constitute the freeway network short-term traffic status prediction model,which has higher prediction accuracy and better anti-noise ability and thus provide reliable technological guarantee for intelligent control and real-time guidance system which helps to relieve the congestion of freeways.