Research On Bus Arrival Time Prediction Method Based On SVM And Kalman Filtering

Nowadays,with the rapid development of society and economy,the population density of most cities in our country has also increased rapidly,which bring a lot of pressure to road traffic.Therefore,the development of intelligent public transportation system has become the most important issue today.Improving the reliability and accuracy of the bus arrival time prediction model can help residents to arrange travel time rationally and reduce the waiting time at the platform,which can helps public transport operators to rationally dispatch public transport vehicles and promote the development of urban public transport.However,due to the complexity of the urban road network and the various factors that influencing the running of the bus,it is difficult to accurately predict the bus arrival time.This paper analyzes the traffic conditions of specific cities and establishes an appropriate bus arrival time prediction model,which is of great practical significance.Comparing and analyzing the research status of bus arrival time prediction at home and abroad,this paper focuses on the related theories and technologies,offers the overall structure of the forecast model.Analyzing the error of bus GPS data from many aspects,thereover offers the corresponding solutions based on the origin of the error.The existing bus historical data is used to analyze the factors that affect bus travel.By using the sample data and combining with poly Class analysis algorithm to analyze the driving laws of buses in the whole daytime period;analyzing the rules between historical sample data and real-time sample data,discussing the problem of feature input selection,drawing some methods and strategies of feature selection.After analyzing the influencing factors and driving laws of bus,the input conditions and the reference data of the forecasting model are determined.According to the state of vehicle driving,the total driving time is divided into two parts:the driving time between stations and the stopping time of the stations.The driving time between stations is predicted by SVM model based on improved particle swarm optimization and multiple linear regression model.Site docking times are predicted by using frequency-weighted data fusion techniques,so as to obtain a predicted reference time,and then real-time data based on Kalman filter algorithm to dynamically modify the reference time and then get the predicted arrival time.Finally,the GPS raw data of 10 bus lines in Dalian are selected to test the prediction model.The results show that comparing with the typical algorithms,the prediction accuracy of this model is obviously improved.