Research On Short-term Traffic Flow Forecasting For Highway Tollgates

The problem of traffic flow at high-speed intersection toll stations has always been a well-known bottleneck in traffic networks,and it is a problem of short-term traffic flow prediction in traffic systems.Short-term traffic flow is affected by many external factors.Traditional forecasting methods are not suitable for large data,and the prediction algorithm based on machine learning can make full use of the known information,find the laws from historical data,and more accurately predict the short-term traffic flow.in the future.This paper builds models based on the Support Vector Machine(SVM)algorithm,the Random Forest(RF)algorithm and the Gradient Boosting Regression Tree(GBRT)algorithm respectively,and studies the prediction effect of each model.The main work contents and research results are as follows:(1)We propose statistical analysis for the traffic flow data sets in the experiment and deal with outliers and missing values in the data set.Exploring the factors influencing the changes in traffic flow,we propose a feature design method based on route,weather,and time to construct a training set.(2)A support vector regression model was established to predict the short-term traffic flow,an optimization algorithm based on the linear regression was proposed to solve the problem that the support vector machine did not consider the correlation between the input variables and the prediction targets in the regression prediction,which reduces the computational complexity of the model and improves the efficiency and accuracy of the model.(3)A random forest regression model was established to predict the short-term traffic flow,and compare it with the optimized support vector regression model.The results show that in the short-term traffic prediction problem,support vector model is suitable for processing small data sets,random forest model is more suitable for processing large data sets.(4)A gradient boosting regression tree model was established to predict the short-term traffic flow;and the results were compared with the RF model,showed that in short-term traffic prediction based on boosting ensemble learning method is an effective and feasible method.In addition,the importance of features is evaluated based on gradient boosting tree,a prediction method based on the dynamic selection feature of GBT model was proposed,and a traffic flow prediction experiment was performed after selecting features dynamically by the gradient boosting tree model.The experiment show that compared with the RF model,theprediction of the GBRT model after reduced features is more accurate,and the prediction effect is the best in this paper.