Determination And Application Of Alighting Stops For Bus Passengers Using IC Card Based On Multi-source Data Mining

Conventional public transport is an important part of urban public transport.The missing bus GPS data caused by rainy weather and other factors will affect the determination of boarding and alighting stops for IC card passengers,so it is necessary to repair bus missing arrival data.Only boarding information of IC card passengers can be found in most cities,so alighting stops need to be identified.Moreover,the selection of staying adjustment bus lines along rail transit before its operation need to be determined.In order to solve the above problems,the main work of this paper can be summarized as follow based on multi-source data mining.(1)A method based on multi-source data and DBSCAN algorithm is proposed to estimate missing bus arrival data.The proposed method uses association analysis with multi-source data to complete the missing arrival name,longitude and latitude,and uses data from similar classes clustered based on DBSCAN algorithm to determine the missing arrival time.The proposed method can improve the shortcomings of low accuracy and poor universality of existing methods by theoretical analysis.The case of Xiamen shows that the presented method can repair all missing station name,longitude and latitude correctly,and it also has higer accuracy and universality than the clustering method based on GPS data and the maximum probability estimation method when completing the missing arrival time.(2)A method based on two-layer Stacking framework is proposed to determine alighting stops of unlinked trips for IC card Passengers.In the first layer of two-layer Stacking framework,five methods are used at the same time as the individual high frequency stop method,the method based on site heat,the method based on transferring convenience probability,the method based on land use attraction probability and the group historical set method.In the second layer of two-layer Stacking framework,the logistic regression model is used to determine the appropriate weight of each method in the former layer for the final results in different data sets,which can help to improve the generalization?ability.The proposed method can improve the shortcomings of low identification rate and low accuracy of existing methods by theoretical analysis.The case of Xiamen shows that the method based on two-layer Stacking framework has higher identification rate than the method combining the high frequency sites and site heat,and it also has better accuracy than the method based on KNN,the method based on decision tree,the method based on random forest and the method combining the high frequency sites and site heat.(3)A method based on IC card data is proposed to select bus lines staying adjustment along rail transit.The proposed method can be used to construct indexes considering influences of conventional bus lines and rail transit prior to the operation of rail transit.Then,constructing and solving the super-efficiency DEA model to get efficiency scores of bus lines whose staying adjustment orders need to be determined.The proposed method can improve the lack of combining the impacts from bus lines and the opening rail transit,and the inability to determine the orders of bus lines staying adjustment by theoretical analysis,which can help to get the higher accuracy than existing methods.The case of Xiamen shows that the proposed method can select bus lines staying adjustment with the high accuracy when the number of lines need to be adjusted is known,and the proposed method is more targeted in studying different lines and better in determining orders of lines to be adjusted than the two compared methods: the method based on collinear lengths of bus lines,and the method based on the generalized travel time costs and travel time savings proportion.