Research On Recognition And Prediction Of Conventional Bus Operation State

Ideally,bus vehicle will arrive at the bus station at the scheduled time according to the established schedule."Bus Bunching" refers to the phenomenon that on the same route,buses depart according to a predetermined schedule,but as time goes by,two or more bus vehicles bunched together or arrive at the station simultaneously.This abnormal variation of bus operation state will not only significantly affect the efficiency and quality of bus service,but also increase the additional waiting time for passengers.In order to better manage and control the bus system,this paper studies the identification and prediction of the bus operation state.This article focuses on the following aspects:First,based on the analysis of the formation process of bus bunching,the microscopic bus operation state is divided into five different operating states which are bus bunching,bus bunching transition,normal operation,large interval transition and large interval respectively;by pre-processing the GPS data,effective arrival time interval samples obtained finally.Using K-means clustering algorithm to cluster the bus arrival time interval samples and analyze the stability of the clustering results.Based on the identification of the micro-bus operation state and the threshold,and the macro-bus operation state is identified,which are Cluster state,PreTransition State,Post-Transition State and Normal State.Then,the probability distribution fitting is performed on the arrival time interval samples of the bus lines at different stops and different time periods,and a more uniform probability distribution fitting form for each bus line arrival time interval sample is obtained.Based on the state judgment threshold,the occurrence of bunching and normal operation at different stations and different time periods is calculated respectively,and analyzes the temporal and spatial distribution of the bus operation status.Next,based on the Markov property of state transition in the bus operation process,the evolution law of bus operation state is analyzed by using Markov correlation theory.The study found that: the probability that the bus vehicle maintains the initial bus operating state decreases with the increase of the running distance,and the probability of transferring into another bus operating state increases with the increase of the running distance.Finally,the bus operation state prediction models based on two different prediction ideas were established,which were the Multiple Logistic bus operation state prediction models and the Support Vector Machine bus operation state prediction models;using the bus line arrival time interval samples,the unknown parameters in the prediction model are calibrated,and the prediction effects of the two prediction methods on the running state of the bus were compared.The study found that,due to the consideration of the evolution law of bus operation status,the prediction performance of Multinomial Logistic prediction model is better than the Support Vector Machine prediction model for the prediction performance of bus operation state.