Study On Characteristic Analysis And Prediction Of Upstream And Downstream Traffic Volume Of Freeway Merging Areas

Freeway merging areas are the "bottleneck" of the highway sections. Vehiclesentering a merging area mainly are upstream of the main line and connected theon-ramp traffic vehicles. In merge area of freeway, the behavior of the confluence ofvehicles easy cause the mainline traffic flow disorder. The large volume of traffic flowand the unsmoothness of the confluence lead to long line of traffic to the main and eventhe mainline jam. As a consequence, the traffic behavior and characteristics of vehiclesin these situations easily lead to security issues and reduce traffic efficiency. Due toflow with the mainline and ramp flow confluence area has a certain relevance, throughthe analysis of the flow characteristic of confluence area upstream and downstream, tomaster the flow rate change rule of confluence area, judge the trend of its development,to ensure the smooth running of vehicle, improve the efficiency of the confluence areatraffic and traffic safety is of great importance.To this end, the paper combines the analysis of spatial and temporal’s correlationand support vector regression machine to implement on-ramp traffic flow short-termforecasting under the influence of more ramps.It based on Highway confluence areainterchange ramp traffic flow time series data analysis. It is researched and establishedthe regression forecast model which is suitable to Highway confluence area interchangeramp short-term traffic prediction.The main research contents include:①Traffic flow characteristic analysis of Highway and interchange ramp flow area.Firstly, the ramp traffic data time series is divided into horizontal time sequence andvertical time sequence.Then, it is introducied by the similarity measure function tomeasure the time correlation and spatial correlation.In the end, similarity measure of"time-sharing"method is uesed tocompare and analyse the changable traffic flow featureof freeway on-ramp.②The prediction model is established based on combined correlation analysis oftime and space with Support Vector Regression. Firstly, the shortcoming of traditionprediction model based on Support Vector Regression (SVR) is analyzed, the input ofwhich is the adjacent n times traffic flow data; secondly, the tradition Support VectorRegression is improved by using correlation analysis results of time and space. And theprediction model is established based on combined correlation analysis of time andspace with Support Vector Regression; finally, the parameters of SVR are given through Grid Search Algorithm, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO)Algorithm. Meanwhile, the effect of prediction is analyzed by making use of the measureddata.③The weighted least squares support vector regression is established to forecastthe traffic peak samples in time series. The traffic peak samples in time series are poorlyfitted in the least squares support vector regression model. Therefore, a new model isgiven based on weighted least squares method and predecessors’ research results andused to predict peak samples in time series. In this model, the weighted correctioncoefficient of fitting error is designed to increase the weight of peak samples fittingerror. Meanwhile, the effect of fitting is analyzed by making use of the measured data.