Research On Bus Passenger Flow Prediction And Bus Plan Optimization Model Based On Big Data And Machine Learning

In recent years,with the rise of big data technology,through a large number of user data and machine learning methods,understanding the non-aggregate user traffic demands has been widely concerned by scholars.Compared with the statistical method,machine learning and deep learning method can reveal the internal relationship between the data better;from the perspective of non-aggregation,traffic planners can better grasp the user’s travel rules and characteristics.Therefore,big data processing technology and machine learning method are favored in transportation area.In this paper,big data processing technology,machine learning and deep learning methods,operations research and other methods are used to integrate various types of data,and the relevant research on the travel characteristics of public transport passengers,traffic demand prediction and bus plan optimization are carried out.The details are as follows.(1)Analysis of travel characteristics of public transport passengers.From the perspective of non-aggregation,the smart card data of passengers were reconstructed to form a complete travel chain.DBSCAN algorithm was used to analyze the time and space travel rules of passengers.Based on the characteristics of passenger travel,passengers were divided into regular travel and irregular travel,and their travel preferences and travel rules were analyzed.Based on the importance of nodes,this paper designed a bus stop clustering method,which clustered the bus stops to form a non-intersection area,and analyzed the characteristics of regional passenger flow and the relationship between regional passenger flows.(2)Passenger flow forecast of bus stops.In order to meet the different demands of multi-line and single line bus passenger flow prediction,this dissertation proposes multiline passenger flow prediction model and single line high-precision passenger flow prediction model based on the algorithm of Xgboost and LSTM in machine learning.The multi-line passenger flow prediction model is the Xgboost passenger flow prediction model based on the points of interest(POI)data(XPPM-POI).Multi-line passenger flow forecasting requires the model to balance the accuracy and efficiency,that is,the model should be established in an acceptable time range.Therefore,this paper chooses Xgboost as the core algorithm of multi-line passenger flow prediction to accelerate the training speed of the model.The POI data was considered in the model to improve the data dimension and increase the accuracy of the model.For the single line passenger flow prediction problem,after the analysis of passenger travel characteristics,the dissertation proposes the LSTM passenger flow prediction model considering the passenger travel characteristics and interest points(LPPM-TC).Due to the small amount of input data,the main goal of single line passenger flow prediction is to improve the prediction accuracy.Therefore,this paper chooses the LSTM model which can consider the influence of the previous data on the prediction data as the core algorithm.In order to improve the prediction accuracy,LPPM-TC model not only uses the POI data,but also expands the number of parameters and improves the accuracy of the model by grouping passengers.Therefore,the organization of LPPM-TC model is more complex and suitable for single line passenger flow prediction,while XPPM-POI model is relatively simple and can meet the needs of accuracy and efficiency of multi-line passenger flow prediction.In this paper,by comparing with history average model,support vector machine(SVM)model and other models,the excellent performance of the proposed model is verified.The advantages and disadvantages of different models in passenger flow data prediction are analyzed.In the process of XPPM-POI model training,the best query radius of POIs around the bus stop is calibrated and the influence of different types of interest points on XPPM-POI is calculated.(3)Design of bus operation plan.In actual production and life,public transportation companies usually design different operation plans.With the help of big data and machine learning algorithm,it is possible to design and change bus plans in real time.Therefore,this dissertation applies big data technology and machine learning algorithm to study the transfer coordination problems and the shuttle bus problems.Based on the passenger flow prediction model proposed in this paper,RTTBO model and MPSBO model are established respectively.RTTBO model takes XPPM-POI model as data input,which can predict the number of passengers in multi-lines in real time,update the transfer passenger number in real time by a novel passenger travel probability network,and update the realtime transfer coordination plan.RTTBO model includes four parts: passenger flow prediction,passenger substitution probability prediction,travel time calculation and optimization model.The passenger flow prediction part is based on XPPM-POI model,and the passenger transfer probability prediction part establishes a neural network(ANTPP)model to calculate the passenger transfer probability,so as to obtain the number of passengers in real time.The MPSBO model aims to design shuttle bus plan for different time periods,which can generate different shuttle bus plans for different periods.The LPPM-TC model is used as the input of the model,which can output the shuttle bus plan including monthly plan,weekly plan and hourly plan.Each plan includes the passenger flow prediction part based on LPPM-TC model,the travel time calculation part based on historical average model,and the departure scheme design part based on optimization.In order to solve different these two models,this dissertation designs the fast solution of the optimization model considering the legal departure interval and the genetic algorithm considering the weight model of different optimization objectives to find the optimal solutions of the optimization problems.The advantages of RTTBO model and MPSBO model are verified by comparing with existing plans and research methods.