Highway Traffic Performance Evaluation Based On Big Data

With the rapid development of economy and the continuous improvement of urban infrastructure,people have more willingness to travel.The emergence of various traffic tools allows people to travel freely.However,as the amount of transportation grow up significantly,traffic problems make the costs that people travel increased greatly.This in turn inhibits people's travel willingness.Therefore,real-time traffic status and evolution trend become the key information that people want to be informed before they plan to go out.This paper mainly studies the two subjects: real-time traffic status assessment and short-term traffic running parameters prediction.Firstly,by using traffic data,real-time traffic status assessment model is established.Then,this paper establishes the model to forecast traffic running parameters.This paper studies real-time traffic state evaluation firstly.Traditional traffic status assessment models could be divided into two categories:index calculation or machine learning algorithms.However,Index calculation does not apply to all nodes.Machine learning algorithms adapts to different nodes very well,but the evaluation results lack traffic theory basis.Thus,this paper manage to establish such model.Technical approaches used in this paper are as follows: firstly,traffic evaluation principles are discussed,then characteristics of traffic historical data are explored.On that basis,Gaussian Mixture Model(GMM)which takes less computation cost and has traffic theory basis is designed to evaluate real-time traffic performance.Datasets from PeMS system are used to test the performance of this model,and it turns out that Gaussian Mixture Model has a good assessment accuracy and applicability.On the basis of real-time traffic status assessment,this paper explores the traffic parameter prediction problem.Popular traffic parameters prediction models are either based on time sequence model or machine learning algorithms,these methods have two main problems.Firstly,they lacks feedback link in training process.Secondly,these model could not choose the best structure automatically in different datasets.Therefore,the predication accuracy is not good and they could not be used in different situation.This paper aims to make up a short-term traffic flow prediction model based on Deep Belief Networks(DBN)which is a kind of neural networks.This model uses multiple layers of RBM reducing the dimension of datasets,and adds several BP layers to analyze the structure of neural networks to select the best one.Datasets from PeMS system are used to test the performance of this model,and it turns out that DBN has a good predication accuracy and applicability.Gaussian Mixture Model and Deep Belief Network constitutes the traffic performance evaluation system.Gaussian Mixture Model is mainly used in real-time traffic status evaluation,it has a good assessment accuracy and applicability.Deep Belief Network is used in traffic parameter prediction,and it could achieve both higher accuracy and good expansibility.