Research On Forecast Model Of Freeway Traffic Volume

With the rapid development of Chinese intelligent freeway,an increasing demand for predictable traffic state in the future of freeway.The accurate prediction of traffic volume as the basic data plays an important role in the intelligent transportation system of expressways.Based on the toll collection data of expressway in a certain province of northeast China,this paper qualitatively and quantitatively analyzes the impact of freeway traffic volume and time characteristics and the weather factors on the traffic volume.Based on the above analysis,the expressway traffic volume forecasting model is studied.The main innovations are as follows:1.A short-term traffic flow forecasting model of freeway based on Data Fusion is proposed.Firstly,based on the analysis of the spatio-temporal characteristics of freeway traffic volume,the traffic time series,periodic similarity and space sequence state vectors are defined,and the input dimensions of the three characteristics are respectively studied by BP neural network prediction model.Then,based on the BP neural network forecasting model,combined with the adaptive weighted data fusion algorithm to fuse the value of space and time characteristics,the model of traffic flow forecast is constructed.The experimental results demonstrated that the average absolute error percentage of the traffic forecast on the working day and rest day is within 5%,and it has better prediction accuracy and practicability than the univariate forecasting model,which provide data support for intelligent traffic management systems.2.A short-term traffic flow forecasting model of freeway based on RF-IABC-MKLSSVM is proposed.Firstly,based on analysis of traffic data characteristics of freeway,a dynamic feature variable selection model of random forest algorithm is established to filter out the important variables influencing the traffic flow forecast of car and trucks.And select the least square support vector machine as the basic prediction model,select the Poly kernel function,Sigmoid kernel function,RBF kernel function to establish multi-kernel function.Then,the parameters of MKLSSVM are optimized by using the improved artificial bee colony algorithm which combines the differential evolution,the optimal location selection and the adaptive adjustment idea,the model of traffic flow forecast is constructed.The experimental results show that the forecast accuracy of traffic flow of car and trucks at peak and no-peak period is high,and the total traffic forecast error is controlled in[-30,20].