Short-term Traffic Flow Prediction Based On Massive Bayonet Data

With the rapid increase of vehicle ownership,the contradiction between supply and demand of limited road resources and increasing vehicle ownership is becoming more and more acute.Traffic control and management is an important means to deal with urban transportation system problems.Traffic volume prediction is one of the most significant techniques for traffic management.How to estimate and predict traffic volume accurately and efficiently has always been the key point in the field of intelligent transportation research.Generally,traditional method obtains the traffic volume information from fixed detector,which has the defects of high maintenance cost and low accuracy.The bayonet data can accurately identify each vehicle's information and could obtain the traffic volume accurately with easy maintenance and strong applicability.At the same time,the travel route of vehicle could be extracted through the information of license plate.The time and space relationship of the traffic flow of different bayonet can be obtained,so that the traffic volume forecast is more accurate.The real time traffic flow status is analyzed and evaluated by using both real time data and history data.It can dynamically grasp the real time change characteristics of urban traffic,predict the characteristics of road traffic in advance,and optimize the dynamic scheduling of road resources.Based on the massive data mining analysis,the traffic volume is effectively represented in this thesis.Firstly,this paper puts forward the concept of the bayonet path.Through the Apriori association rule,we carry out the statistical analysis of the mass over the vehicle track on the card level,and characterize the spatial and temporal relationship between the traffic volume and the traffic volume between the sections of the section and the traffic volume prediction of the traditional single cross section,and turn them into the decision tree back of the time series of multiple card ports.The problem is to improve the scientificity and accuracy of traffic volume prediction.On the problem of traffic time series prediction,the information is identified and extracted by the way of data reconstruction.This paper puts forward the method of K nearest neighbor based on Apriori association rules,random forest based on Apriori association rules and gradient lifting regression tree based on Apriori association rules.Prediction accuracy.The paper combines the traffic problems with the new computer technology,and uses the Spark distributed computing framework,which has high efficiency,good usability and strong generality.It stores,statistics and real-time analysis the data of massive card crossing over the car,and realizes the distributed algorithm of short-term traffic volume prediction.Finally,by analyzing the data of car crossing in a region of Suzhou,the paper validates the effectiveness of the random forest based on Apriori association rules and the gradient lifting tree regression based on Apriori association rules in traffic volume prediction.