Research On Bus Bunching Prediction Based On Improved Support Vector Regression

Bus bunching is a common phenomenon in bus operation,which would lead to an increase in the average waiting time of passengers,uneven passenger loading capacity and reduce the reliability of bus services,which is a problem that cannot be ignored.Short-term bus bunching prediction is the basis for bus scheduling optimization and the accuracy of the bus bunching prediction directly affects the improvement of bus bunching phenomenon.However,the volatility in the running state of bus vehicles makes it more difficult to accurately predict the bus bunching.Based on the study of bus bunching characteristics,this paper establishes a bus bunching prediction model based on improved support vector regression,aiming at improving the accuracy of bus bunching prediction.This paper analyzes the process of bus arrival and departure at station,clarifies the process of bus bunching formation,and qualitatively analyzes the factors affecting the bus bunching.Based on the actual bus arrival and departure data,this paper analyzes the bus operation status and reflects the characteristics of the bus vehicles by bus operation status diagram.Also,the spatio-temporal variation characteristics of bus bunching are analyzed and summarized.Based on the analysis of the spatio-temporal characteristics of bus bunching,a bus bunching prediction model based on improved Support Vector Regression is proposed which divides the prediction of bus bunching into two stages.The first stage of the bus bunching prediction is based on the single Support Vector Regression Model(SVR).The genetic algorithm is used to optimize the model parameters,and the historical data of bus arrival and departure is used to predict the bus headway.In the second stage,the real time bus arrival and departure data is considered and the result of bus bunching prediction in the first stage is dynamically adjusted by the Kalman Filter.The improved Support Vector Regression Model uses the single Support Vector Regression Model to fit the non-linear factors in the process of bus bunching and dynamically adjust the result of bus bunching prediction by the Kalman Filter.The improved Support Vector Regression Model combines the single Support Vector Regression Model with the Kalman Filter to make up shortcomings of the single Support Vector Regression Model with poor real time performance and the single Kalman Filter which is easily affected by random factors so that the result of bus bunching prediction not only reflects the historical rule of the bus headway,but also enhance the real-time and anti-interference bus bunching prediction ability.This paper selects several bus lines to predict and analyze the bus headway and bus bunching rate by using several evaluation indicators such as mean absolute error,mean absolute percentage error and root mean square error.This paper compares theprediction of improved Support Vector Regression Model and single Support Vector Regression Model and ARIMA model from different time and space dimensions(multiple routes,single station,multiple stations,different time types).The results show that the bus bunching prediction model based on improved Support Vector Regression Model is better than the single Support Vector Regression Model and ARIMA model.Improved Support Vector Regression Model performs better during the period when the bus operation state fluctuates greatly such as the morning and evening peak hours of the working day.The comparison proves that the improved Support Vector Regression bunching prediction model has good adaptability to the bus bunching prediction and can effectively improve the accuracy of the bus bunching prediction.