Data Driven Prediction Of Bus Bunching And The Control Strategies

To alleviate urban traffic congestion,it is necessary to vigorously develop public transport and optimize the allocation of road traffic resources.While urban ground bus plays a decisive role in public transport system,bus bunching is a common phenomenon during the operation.Due to its backward propagation and continuous influence,bus bunching greatly limits the improvement of operation efficiency and service quality which are becoming more and more critical criterions in people’s daily travel mode choice.Therefore,raising awareness of bus bunching and adopting reasonable approaches to effectively alleviate this phenomenon can significantly improve the efficiency of bus operations and the level of bus services,enhance general satisfaction with bus travel modes,thus increasing the its attractiveness,alleviating urban traffic congestion to a certain extent,and adding vitality to the development of urban transportation.This paper takes Beijing ground bus operation system as the research object.Starting from the typical process of bunching formation,it explores the causes of bus bunching in Beijing based on historical data,designs a bus bunching prediction model to scientifically predict the location and time of its occurrence,and then constructs a reasonable bus bunching control strategy to regulate the running process in a balanced manner,thus providing a systematic solution to bus bunching from the excavation of causes to the construction of control strategies.Firstly,this paper performs data cleaning operations to bus operation data in Beijing(including the correction of bus trajectories and passenger origin-destination information,coordinate conversion,and map matching,etc.)and proposes a bus bunching identification method.Then,it analyzes three root causes and six basic processes of bus bunching,and illustrates specific performances of these three root causes in Beijing bus system via data and schematic diagram.On this basis,this paper extracts basic features that affect the headway of buses,makes them concrete by combing with bus bunching data in Beijing,and constructs bus bunching prediction models based on support vector machine model,XGBoost model and random forest model,from the perspectives of linear model,nonlinear model and integrated learning,thus realizing the prediction of bus bunching location for different routes,stations and vehicles.All these three models are able to well predict the headway of buses.Among them,XGBoost model has the highest accuracy while support vector machine model has the lowest.Finally,based on the above prediction results,considering the effects of uneven passenger flow within station,the scheduled departure frequency and other factors on bus bunching,combined with minimum fleet scale principle of inverse difference function,this paper establishes a bus bunching model with the minimum total travel time of the bus system as the optimization objective and with passenger allocation,bus fleet scale,bus lag time as constraints,uses bus bunching prediction model to find bus lag control point,then solves bus bunching lag time by genetic algorithm.This paper also performs case study on the above model by Beijing bus operation data(e.g.,bus line 94,etc.).Its result shows that inverse-difference-function-based bus holding model can effectively alleviate the urban bus bunching problem.