Research On Traffic Flow And Travel Time Short-term Forecasting Method For Freeways Based On Data Driven

Short-term forecasting of traffic flow and travel time is the basis of the implementation of intelligent transportation system.Accurate and effective short-term forecasting of traffic flow and travel time can not only allow travelers to understand the road conditions and arrange travel routes in advance,but also allow relevant traffic management departments understand the trend of traffic changes,do a good job of induction in advance.These can save travel costs,make full use of existing road resources,and reduce traffic congestion.In addition,as the characteristics of the short-term traffic flow on the highway is strong non-linear,uncertain,and time-varying,thus,how to accurately predict the highway traffic flow and travel time in real-time is a challenging task and important research issues.In this paper,based on the highway masscharged data,the short-term forecasting of highway traffic flow and travel time is mainly studied.The main research contents and innovations are as follows:(1)In order to establish a reliable data foundation for the short-term forecasting of highway traffic flow and travel time,this paper analyzes and summarizes the original charge data of the highway and gives the data anomaly identification methods and processing methods.Through analysis,highway charge data is redundant,missing,errors,and other anomalies.For the case of data redundancy,delete redundant data directly.For the case of missing data,the weighted average method and the K nearest neighbor algorithm are used to compensate the missing data based on the continuous missing cycles.For the case of data errors,the methods of repairing error data by road information and deleting data that is not in the interval [?-2?,?(10)2?]are used according to whether the mileage data error or the travel time error,where ? is the sample mean and ? is the standard deviation.(2)In order to improve accuracy and stability of the short-term forecasting of highway traffic flow,a short-term forecasting model of highway traffic flow based on multi-features GBDT is proposed in this paper.This paper applies the Gradient Boosting Decision Tree(GBDT)model to short-term traffic flow forecasting problems for the first time and mined several effective new features such as weather and time through analyzing the influencing factors of the short-term traffic flow on the highway.Based on this,a short-term traffic flow forecasting model based on multi-features GBDT was constructed.Finally,the validity of multi-features GBDT model was verified by the real data of Guangzhou Airport Highway.(3)In order to improve accuracy and stability of the short-term forecasting of highway travel time,a short-term forecasting model of highway travel time based on TS-SVR is proposed in this paper.According to the theory of traffic flow,travel time is closely related to traffic flow.Therefore,this paper combines short-term travel time forecasting with short-term traffic flow forecasting,constructs a short-term forecasting model for highway travel time based on the TSSVR(Two Step-Support Vector Regression,TS-SVR): firstly,predict the traffic flow in the next period,and then add the traffic flow in the next period as a feature to the input vector of the short-term travel time prediction model based on SVR(Support Vector Regression).Finally,the effectiveness of the proposed model is verified by the actual data of Guangzhou Airport Highway.