Highway Traffic State Identification And Travel Speed Prediction Based On Key Operating Vehicle Data

Due to the large number of passengers or the transportation of dangerous chemicals,the "two passengers and one danger" vehicles will bring more serious losses in the event of a traffic accident.Due to the large size and low speed of such operating vehicles,it is one of the important factors for traffic congestion.The installation of the Beidou vehicle terminal enables real-time monitoring of key operating vehicles and also provides data for road traffic status identification and traffic flow prediction.Based on above,this paper studies the highway traffic status identification and travel speed prediction,aiming at improving the supervision level of key operating vehicles,providing technical support for reducing traffic accidents and effectively alleviating traffic congestion.The main content is:(1)For the problem of abnormal data and missing data of operating vehicles,anomalous data processing rules were formulated,and a trend-historical missing data filling algorithm was proposed.Through a large number of statistical analysis of operational vehicle data,seven kinds of abnormal data representations are summarized,and corresponding processing rules are proposed according to the characteristics of abnormal data.A trend-history filling algorithm is proposed for the missing of time series data.Taking the data of the operating vehicles in the test section as an example,the validity of the data processing method is verified.(2)Based on the travel speed of the road and the delay of the unit travel time,an FCM highway traffic state recognition algorithm is proposed.The algorithm uses the Euclidean distance as the metric to obtain the clustering center points of five types of traffic states:unobstructed,basically unblocked,generally congested,moderately congested,and severely congested by iterative solution,and the traffic state is identified by the membership degree matrix.The example verification shows that the proposed algorithm can effectively identify the traffic state of the road.(3)Propose a highway speed prediction algorithm for SA-SVR.Using the global search characteristics of the simulated annealing optimization algorithm,the kernel function factor and penalty factor of SVR are optimized to realize the prediction of the travel speed.Taking three experimental sections as an example,it is proved that the optimized algorithm can effectively increase the stability and prediction accuracy of the prediction results.(4)An algorithm for highway travel speed prediction based on LSTM is proposed.The algorithm takes the travel speed of the road as input,and uses the long-term memory ability of the algorithm to predict the future travel speed.Three gradient-decreasing algorithms with adaptive adjustment learning rate are proposed to optimize the model.The experimental section verification shows that the adam gradient descent optimization strategy is the best.Finally,the LSTM with adam gradient and SA-SVR are compared in the absence of traffic accidents and traffic accidents.The results show that the LSTM prediction results are better than SA-SVR under the two scenarios.