Rail Transit Passenger Flow Prediction Based On Multi Source Data Mining

Rail transit is an important part of public transport,which plays a leading role in urban transportation.In order to avoid the waste of public transport capacity resources,it is necessary to study the passenger’s switch behavior of bus lines along the rail transit lines after the opening of the new rail transit.After the formal operation of rail transit,the increasing passenger flow puts forward higher requirements for the operation of rail transit.In order to better cope with the changes of passenger flow,it is necessary to effectively analyze and forecast the short-term passenger flow of rail transit.This paper uses multi-source data mining to study the above problems:At present,there are some problems such as unreliable data and low prediction accuracy in the research of passengers switch of bus lines along rail transit after the opening of the new rail transit.Therefore,this paper proposes a CART decision tree-based switch prediction model for passengers of bus lines along the newly opened urban rail transit.Firstly,this paper uses the massive passenger swiping card data before and after the opening of the rail transit to construct the switch data set of bus passengers.Secondly,based on the bus passenger switch data set,this paper extracts 12 factors that may affect passenger switch.Finally,this paper uses the extracted influence features to establish a bus passenger switch prediction model based on CART(classification and regression tree)decision tree to predict passenger switch behavior.The data reliability of this method is stronger,and it has a better interpretability,which is conducive to better depicting the switch behavior of passengers.The experimental results on the dataset of Xiamen rail line 1 show that the proposed method obtains a good prediction effect which performs other methods(e.g.Logit,NB,SVM and ANN).According to the generated tree model,this paper analyzes the sensitivity of each influencing factor(the three most important factors are ‘Transfer times needed after switch’,‘Travel distance’ and ‘Number of metro stations to take after switch’),and mine the decision rules of people’s switch decision-making behavior.To solve the problems of current short-term forecasting methods for metro passenger flow,such as unclear influencing factors,and low accuracy,a method for rail station passenger flow short-term prediction based on multi module LSTM network is proposed.Firstly,using historical data as the training set to transform the problem into a data-driven multi-input single-output regression prediction problem,the problem of the short-term prediction of rail station passenger flow is formalized and the difficulties of the problem are identified.Secondly,we extract the candidate temporal and spatial features that may affect passenger flow at a rail station from multi-source passenger travel data.Thirdly,we use a maximal information coefficient(MIC)feature selection algorithm to select the significant impact features as the input.Finally,a short-term forecasting model for rail station passenger flow based on multi module LSTM network is established.The data sources of this method are more diverse,the extracted influencing factors are more comprehensive,and it can better learn the time series characteristics of passenger flow,which is conducive to improving the prediction performance of the model.The experimental results on the dataset of Lianban rail station in Xiamen city show that the proposed method obtains higher prediction accuracy than SARIMA,SVR,and BP network.Because the formation mechanism of rail transit passenger flow is different in peak and off-peak periods,it is not accurate to use the same passenger flow prediction model to predict rail transit passenger flow in peak and off-peak periods.Therefore,in view of the different characteristics of passenger flow in different periods,this paper introduces transfer learning to model and predict separately the rail passenger flow in the morning peak,evening peak and off-peak.Using the similarity and continuity of passenger flow,this paper puts forward a rail station passenger flow prediction model based on transfer learning and multi model.This algorithm can optimize the initial weight value of the network model and effectively improve the training efficiency of the model.In addition,under the condition that the number of available samples is small,the algorithm can avoid over-fitting of the model and improve the generalization ability of the model.The experimental results on the dataset of Lianban rail station in Xiamen city show that the passenger flow prediction model based on transfer learning and multi model is better than the whole period passenger flow prediction model in the morning peak,evening peak and off-peak periods.