Research On Urban Bus Bunching Prediction Based On Machine Learning

Giving priority to the development of public transport can effectively alleviate traffic congestion and improve the utilization rate of road resources,as an important part of the public transport system,public transport should be focused on development.However,public transport is extremely disturbed by external factors and deviates from the original plan in the process of operation,which will directly lead to the phenomenon of bus bunching,thus,increasing the uncertainty of passenger waiting time,forming the phenomenon of empty bus,causing the waste of public transport resources,and reducing the willingness of passengers to choose public transport.Under this research background,this thesis intends to analyze the time-space characteristics of bus bunching,deeply understand the formation process of bus bunching,excavate the factors affecting the bus bunching from various aspects,and build a bus bunching prediction model to predict the bunching state of bus,so as to provide theoretical basis and decision support for public transit operation managers to conduct public transit scheduling in advance.This thesis takes Xi’an bus line as the research object,and specifically carries out the following research:Firstly,it clarifies the definition of bus bunching,defines the degree of bus bunching as the judgment standard of bus bunching,and based on the bus operation data in Xi’an,through the preprocessing of missing and abnormal data,calculates and sorts out the data of bus headway and bus bunching degree,and analyzes the space-time change law of bus bunching;Secondly,the generation process of the bus bunching is clarified from three aspects:station,operation interval,and departure interval.Referring to previous research results and specific analysis of actual situations,14 factors affecting bus bunching are preliminarily selected from the above three aspects,weather and time,and through factor analysis method comparison,Lasso regression model is selected for factor analysis,and 7 factors that have an important impact on bus bunching,among them,the number of passengers on the bus,the departure interval and the bus lane have a significant positive impact on the bus bunching degree,while the number of passengers on the previous bus,the speed difference between stations,the time period and the date have a significant negative impact on the the bus bunching degree;Finally,based on the seven factors that have been screened out and have significant impact,Light GBM bus bunching prediction models is constructed,and an improved Light GBM model is constructed by integrating Bayesian optimization algorithm into the Light GBM model,and the prediction performance of the improved Light GBM model is compared with the basic Light GBM,RF,and XGBoost models,the results show that the Light GBM model optimized by Bayes has the best fitting effect on the predicted value and the real value of the bus bunching,the evaluation index R~2 reaches 0.882,and the mean absolute error and root mean square error are only 0.123 and 0.147 respectively,compared with other models,it has the highest prediction accuracy,by dividing the prediction results of this model according to the criteria for determining bus bunching,84.52%of bus bunching states can be accurately identified,verifying the better applicability of the model in predicting bus bunching problems.