Research On The Estimation Method Of Bus Passenger Alighting Station Based On Data Augmentation

Bus passenger alighting station inference is the basis of some smart city applications,and its role is to recover the broken travel chain of passengers and provide important guidance and help for urban transportation planning and passenger flow spatial-temporal analysis.However,the natural attributes of bus systems lead to the absence of many passenger trips.Specifically,most domestic city buses use a ridership strategy where "passengers tap their cards when they get on the bus and really do not need to tap their cards when they get off.Currently,most inference methods are based on a single source of data such as passenger transit card records,stripping transit trips to multi-modal combinations of trips,and then calculating and inferring passenger drop-off points.Although this single-source data-based approach can solve part of the application needs in some scenarios,there are still major limitations in the overall prediction accuracy and timeliness.Considering that more and more cities have opened rail transit systems,there are a large number of interchange interactions between bus and rail transit systems.There is a potential opportunity to use metro ticket card records to improve bus passenger disembarkation stop inference.In this study,we take a bus system in a first-tier city as the research object.We try to use metro ticket card records to improve the prediction accuracy of bus passengers and use a big data platform to improve the timeliness of prediction.Specifically,our main work contains the following aspects: First,we classify passenger travel links quantitatively,and divide the travel links into two travel states and five travel scenarios.The three scenarios are: "round-trip travel by bus on the same route","travel by bus on different routes",and "travel by subway".The two scenarios with discontinuous travel status are "high frequency bus trips" and "bus stop attraction trips".Second,the spatial juxtaposition of the subway network and the bus network is explored to explore the interchange relationship between bus and subway stations,and to infer bus passenger drop-off stations using conditional probability.Third,the method in second is deployed on the streaming computing engine to achieve online deployment and real-time prediction of the model to support a common passenger flow prediction application in the bus scenario.The experimental results show that Compared with the baseline method,the method based on data enhancement adds the "bus transfer subway travel" scene in continuous travel,which corrects 25 % of the results that are difficult to calculate the exact drop-off site record,and the accuracy is high to 95%.By mining frequent binomial sets of stations in discontinuous travel,the accuracy of thfe ’ random travel ’scenario is improved by 22.75 %.And the real-time computing scenario can process more than 60 prediction requests per minute.Research on the Estimation Method of Bus Passenger Alighting Station Based on Data Augmentation shows good scalability and can meet the real-time prediction needs of single route buses in big cities during peak hours.