Study Of Short-Term Traffic Flow Forecasting Based Markov Random Fields

In recent years,with the rapid development of urbanization and transportation construction,people pay more attention to intelligent transportation systems and has been considered that it's the most powerful means to solve urban traffic problems.Traffic flow forecasting is a key component of the intelligent transportation system.Accurate real-time traffic flow forecasting can implement route guidance and alleviating traffic pressure.Therefore,a forecasting model of short-term traffic flow based on a Markov random field is proposed in this paper.The main research contents are as follows:The temporal-spatial relationship and periodic performance of traffic flow are determined by analyzing the actual traffic flow data.In view of the data loss and abnormality caused by the failure of the acquisition equipment in the original data,the mean interpolation method and the moving average model are used in this paper to deal with the missing and abnormalities of the dataset,and this method can improve the accuracy of the prediction.In addition,a space-time Markov random field model of the intersection is constructed in this paper by analyzing the spatiotemporal and periodic characteristics of the intersection.In order to determine the joint probability distribution of the model,this paper uses the processing of the factor graph to obtain the concrete expression of the joint probability distribution.To solve this model,a method of maximum likelihood estimation is used to obtain the parameters of the model.And then a method based on markov random field to predict the traffic flow is proposed in this paper.Due to the extremely large traffic data,it is difficult to store these datasets,and the time complexity of the training model becomes very high.In addition,the development of the social economy has caused changes in traffic patterns.In this situation,it is difficult for the previous model to adapt to the changes in new traffic patterns.Aiming at this problem,an online learning method based on Markov random field is designed in this paper to implement the prediction of traffic flow.This online learning approach can adapt to changes in traffic patterns.Of course,it also reduces the time complexity of model training and the difficulty of data storage.The simulation data and British highway traffic dataset are used to verify the model and the absolute error(MAE)and average relative error(MAPE)evaluation criteria are applied to prove the effectiveness of the model.Since the traffic flow data of non-working days and working days are different,the datasets used to traffic forecasting are divided into two cases:non-working days and non-working days.The prediction algorithm is compared with other algorithms to prove the robustness and effectiveness of the proposed algorithm.