Investigation Of Short-term Traffic Flow Prediction Based On GJR-GARCH Model With An Ensemble Learning Method

Today,it is the trend to construct and develop smart cities.The construction and improvement of intelligent transportation systems is an important part of building smart cities.Traffic is as if the blood of the city.The intelligent and systematic of traffic control and management can effectively alleviate the problem of urban road traffic congestion and improve the traffic utilization rate of urban roads.Traffic flow forecasting is one of the key elements of information processing,maneuver control and intelligent management in intelligent transportation systems.Accurate and real-time forecasting of traffic flow not only help to effectively control the traffic flow of roads,improve road traffic capacity and solve traffic congestion road problems,but also can reduce environmental pollution,improve the level of transportation services,and enhance the sense of travel experience of the citizens.Therefore,the real-time prediction research of short-term traffic flow is of great significance to the construction and improvement of intelligent transportation system and the construction of smart city.This paper takes the sensor data of the I80 corridor near Davis,California,as the research object.Carrying out data cleaning and characteristic analysis on the collected traffic flow data,and analyzes the Gaussian mixture distribution characteristics of the traffic data and the inevitability of traffic flow can be regarded as the chaotic time series.The GJR-GARCH model is used for prediction,and the preliminary prediction based on the time series model is obtained.By investigating the Gaussian mixture distribution characteristics of traffic flow data,the noise is predicted and optimized by adding the Gaussian mixture distribution feature to the prediction data.On the other hand,we calculate the basic parameters of phase space reconstruction: delay time and embedding dimension.Then,phase space reconstruction data is further predicted by ensemble support vector regression model.Finally,a simple fusion method is used to combine the prediction results of time series model and the prediction results of ensemble support vector regression model based on phase space reconstruction.Combined with the advantage of time series prediction of peak period and the dynamic prediction under machine learning model.Finally,the prediction results of the fusion model are compared with the time series model and the SVR model prediction results.Experimental results prove that the fusion model has better predictive performance than the unoptimized and unintegrated model,which can predict the short-term traffic flow more effectively.