Research On Passenger Flow Forecasting Method Based On Bus Card Data

Urban public transportation has always been an important part of urban development and construction,and has a major impact on the economic development of the city and the daily life of the residents.However,with the increasing number of private cars,more and more cities are facing severe traffic congestion and traffic pollution.Improving the utilization rate of public transportation resources is an effective means to alleviate traffic congestion and traffic pollution.Bus passenger flow forecasting is a hot research direction in the field of transportation.Research on this direction can provide important information for the operation and scheduling of urban public transportation,and help the relevant departments of urban public transportation to develop reasonable operational plans and improve public transportation.The utilization of resources,thus effectively alleviating urban traffic congestion and traffic pollution.Therefore,the study of bus passenger flow forecasting is of great significance to the construction of urban public transport.This paper takes the bus card data of the card as the main information,combined with the weather data,firstly pre-processes the two parts of the data,and then comprehensively analyzes the distribution and changes of the bus passenger flow from the time dimension,the crowd dimension and the weather dimension.From the analysis,the passenger flow has a cyclical variation law in weeks.In addition,the time dimension has a greater impact on passenger flow,and the weather factor also has a certain impact on passenger flow.From the results of visual analysis,this paper constructs the characteristics of time type,the characteristics of weather type and a series of combined features from the two aspects of time type and weather type,and selects the optimal feature by backward search method.This paper starts the experiment of model prediction from two ideas.The first idea is to separate the population types by card type and then superimpose the prediction.The second way is to not distinguish between population types and overall modeling prediction.Through experiments,it is found that the overall modeling effect is better than the separate modeling.In this paper,multiple linear regression,random forest and LightGBM models are used to predict passenger flow.LightGBM is a relatively new integrated learning algorithm and is rarely used in bus passenger flow forecasting.Through experimental comparison,the prediction effect of the LightGBM model is optimal.In the previous study of bus passenger flow forecasting,most scholars used single model for prediction,and rarely used model fusion.At the end of the paper,according to the existence of certain differences between the multiple linear regression,random forest and LightGBM models,the three single models are merged.The experimental results show that the prediction effect after the model fusion is further improved than the LightGBM model.