Short Term Traffic Flow Prediction Based On Relevant Vector Machine

The problem of road traffic jam is one of the key factors that restrict the healthy development of urban economy.Real-time and accurate short-term traffic flow prediction is the premise of intelligent transportation systems such as traffic signal control,traffic distribution,route guidance,automatic navigation,accident detection.Traffic flow system is a complex non-linear system,its characteristics of high time variability,randomness,complexity and uncertainty add difficulties to traffic flow prediction,especially short-term traffic flow prediction,because of the influence of road facilities,weather conditions and many other external and human factors,the difficulty and complexity of prediction is higher than medium and long-term traffic flow prediction.Based on the in-depth analysis of the characteristics of traffic flow,the key factors affecting the accuracy of short-term traffic flow prediction and the existing short-term traffic flow prediction models,the short-term traffic flow prediction with high real-time requirements can be regarded as a regression estimation problem in the case of small samples.In view of the outstanding advantages of relevance vector machine in the modeling of small sample regression estimation,this paper establishes a short-term traffic flow prediction model based on relevance vector machine,and through the application of machine learning,statistical learning and swarm intelligence optimization,around the parameters optimization,sample size selection and kernel function construction of the short-term traffic flow prediction model of relevance vector machine The key problems are studied,trying to do some exploratory research on the unsolved problems in the short-term traffic flow prediction modeling.Around these themes,the main research results and innovations of this paper are as follows:1?The framework of short-term traffic flow prediction model based on relevance vector machine is established.Taking traffic flow parameters such as traffic flow,occupancy and average speed as input and traffic flow value in corresponding time period as output,the construction and selection strategy of kernel function in the model are carried out by using Shepard interpolation,paired t-test and clough Tocher refinement,and the parameters of kernel function in relevance vector regression estimation are optimized by combining chaos theory and simulated annealing algorithm.Experimental results on real vehicle flow data set show that the new model achieves better prediction and generalization performance than the classical model.2?A short-term traffic flow prediction model with upstream and downstream road awareness is proposed.The factors such as traffic flow,average speed and lane occupancy are considered in the model.Through the fusion of semantic understanding and binary feature representation,taking the historical traffic flow of the target section,the history and current traffic flow parameters of the upstream and downstream sections collected by the traffic detector as the input and the traffic flow of the corresponding period as the output,a progressive learning method is established.The results show that the RVR model with the perception ability of the upstream and downstream sections can be used in the similar training and prediction time In order to get better prediction performance than the general model.3?A short-term traffic flow prediction model with seasonal index regulation is proposed.The new model uses the DH seasonal index adjustment method to describe the peak time,holidays,rest days,weather changes and other factors,and uses the RBF seasonal core and linear seasonal core to analyze the non-linear data model involved in traffic flow prediction.The actual data prediction results show that the faster fluctuation of traffic flow will lead to greater prediction error value.The seasonal adjustment adopted in this paper It is helpful to deal with the problem of data prediction with seasonal trend.4?The statistical law of the influence of sample size on prediction performance in the new model is explored.Considering the balance between the sample size and the prediction accuracy in the model,the real-time performance of traffic flow prediction is improved.The statistical methods such as Mann Kendall method and non parametric method proposed by Sen are used to detect the variation trend of the prediction value of the short-term traffic flow prediction model of relevance vector machine when the sample size is different.The experimental results show that the relevance vector machine traffic flow prediction model is less sensitive to the sample size,and can get better prediction results than the traditional method when the sample size is small.