Study On The Mechanism Explanation And Prediction Method Of Bus Bunching

Bus bunching is a very common phenomenon in bus operation.The bunching phenomenon can lead to a series of adverse consequences,like the increase of average waiting time for passengers,the unreasonable waiting time for some passengers and uneven loading among different buses.This paper manages to formulate the mechanism of bus bunching by analyzing buses' dwell time at bus stations,delay at intersections,and travel time at intervals.By mining the archived data of bus operation and using the basic concepts of particle filter theory,this paper generates models to predict bus bunching,which can be helpful for operators to predict bunching before happening,and provide basis for real-time bus scheduling.This paper starts with the analysis of bus data to observe the phenomenon of bus bunching in actual operation.For the problem that some GPS point is deviate from the route,an orthogonal projection method is adopted,and the vehicle trajectory is corrected in accordance with the principle of proximity and time series.Then this article marks the location of GPS,intersections and stations,to extract the arriving time to the node,draw the space-time map of the vehicle and the distribution of the headway to intuitive analysis of variety characteristics of headway.Before the establishment of the dynamic equation of bus bunching,the paper first summarizes the four causes of bus bunching,including the delay of the vehicle in the interval,the early arrival of the vehicle,the multiple passengers at the station,and the interference of the signalized intersection.After setting the conditions for the determination of bus bunching,a bus bunching model can be established that can explain the fluctuations in the vehicle headway caused by differences in dwell time,delays at the signalized intersections,and fluctuations in the running time.Among them,the dwell time model established in this paper describes the proportional relationship between the dwell time and the headway;in the intersection time model,based on the use of the vehicle trajectory to calculate the signal timing of the intersection,the actual passing intersection time of the vehicle is calculated.This model also shows the adjustment mechanism of the headway at the signalized intersection.For the running time of the bus,this paper gives five candidate distribution forms,and introduces the method of parameter estimation and k-s test.Using the actual information of the bus data to obtain the key information,this article calibrates the parameters in the bus bunching model.In the dwell time model,the per capita swiping time can be calibrated by a unique value,and the passenger's real-time arrival rate can be fitted by a quadratic function.In the derivation of intersection signal parameters,when the iteration period coincides with the actual period,the distribution of the track points at the intersection zero point is obviously different from the error iteration period.For the fitting of interval running time,the Log-logistic distribution has the best fitting effect.After setting boundaries of the full day data,the Log-logistic distribution passes all K-S test.Finally,based on the idea of particle filter,this paper establishes a prediction model of bus arrival time,and predicts the space-time location of bus bunching in the future.Because each particle's predicted trajectory is independent from each other,the method of averaging all particle's weights to get the predicted value in traditional particle filter is no longer applicable.This article selects the median of the effective particle's prediction result as the prediction value of the arrival time.The prediction results of actual data show that the prediction model has good effect on the prediction of bus bunching events with fewer stations.With the extension of the interval,the prediction error increases gradually.And the forecasting effect of considering the signalized intersection is better than without considering the signal crossing,as well as the accuracy of Pingfeng period prediction is higher than that of the peak period.