Research On Pattern Recognition Of Metro-Bus Transfer Based On Structural Topic Model

With the continuous advancement of the construction of the Chengdu-Chongqing Economic Circle,Chengdu adheres to the concept of public transport-oriented urban development and construction,rapidly developing its public transportation system by improving the “integration of three networks” – subway network,bus network and slow traffic network.This effectively alleviates the problem of urban congestion and enhances the travel experience of residents.Metro and bus lines are gradually networked,greatly improving the urban public transportation capacity levels.In this circumstance,whether the complex transfer patterns of subway-to-bus system that residents choose to travel can be identified accurately,has become the key to next guidance for the planning and construction of the public transportation system,and to continue improving the service of the public transportation system.Based on the subway-bus card-swiping data and the surrounding construction environment data of subway stations,this paper applies data analysis technology to identify the subway-to-bus transfer modes from multiple perspectives.The main work includes the following four aspects:Firstly,the time thresholds of the two transfer modes of B-M and M-B are determined by analyzing the IC card swiping data of bus and the AFC data of subway.On this basis,a multistage travel chain of passengers is built and various indicators of passengers’ transfer are analyzed,which lays a foundation for the recognition of transfer patterns.Secondly,a classification criteria of metro stations based on the land use mix index(LUM),commercial and financial land index(CI),transportation facility accessibility index(MBA),and transportation facility supply index(NBL/NBS)is proposed.Those indexes are regarded as the main reference basis.Furthermore,156 metro stations are classified by using PAM clustering method and K-means clustering algorithm.Moreover,the data cube technology is introduced,combining with the multidimensional characteristics of passengers’ card-swiping data to establish the multidimensional traffic dataset(traffic cube).Besides,the transfer data is analyzed from three dimensions – weekly time,daily time and different subway stations.These different perspectives explore the manifestation of the subway-to-bus transfer mode and its reasons.Last but not the least,on the basis of the topic model,the establishment mechanism of the structural topic model(STM)is redesigned.By converting passengers’ single trip data into a label,it will change the card-swiping data from a total transfer record into a corpus.And finally,the STM model is used to complete the identification and analysis of potential subwayto-bus transfer modes.The results of the pattern recognition of the subway-to-bus transfer in this paper indicate that there are significant differences between transfers for commuting activities and transfers for non-commuting activities.This confirms that there is a certain level of imbalance between land for working and land for housing in Chengdu.The research results provide useful enlightenment for policy makers to build a more balanced relationship of work and life,by adjusting the urban structure.In addition,the commuting transfer mode which is identified in this paper can be viewed as an important reference for the design and optimization of the public transportation networks.It is helpful to formulate and demonstrate policies,such as reducing residents’ dependence on cars for travelling,shortening excessive commuting distance,alleviating traffic congestion.