A Combined Neural Network Model For Forecasting The Subway Passenger Flow In Nanchang

With the continuous expansion of urban rail transit network and the rapid increase of passenger flow,the temporal and spatial characteristics of rail transit passenger flow are becoming increasingly complex,the imbalance between supply and demand at different time periods and different stations is very prominent,the efficient and safe operation of the subway is challenged.Therefore,how to accurately predict the passenger flow in the future period of time greatly affects the quality of subway operation and service.Based on the analysis of the research status and related results at home and abroad,the advantages and disadvantages of various methods in the field of short-term passenger flow forecasting are summarized,and it is found that the neural network model of deep learning is superior to the traditional time series forecasting model and machine learning model in short-term passenger flow forecasting.Therefore,this study proposes a combination of clustering algorithm,graph convolutional neural network(GCN)and long short-term memory neural network(LSTM),and then the combined model predicts the short-term passenger flow of the metro.The main research contents of this paper are as follows:(1)The research status and related results of domestic and foreign passenger flow forecasting are systematically sorted out,the problems in short-term passenger flow forecasting are summarized,and the research direction of this thesis is determined.Through literature research,the advantages and disadvantages of the commonly used models for passenger flow forecasting are found,that is,although the traditional time series model is simple and efficient in calculation,it cannot fit the nonlinear problem of passenger flow.Although LSTM and GCN can capture the temporal and spatial characteristics of passenger flow respectively,only the combination of the two can capture the temporal and spatial characteristics of passenger flow at the same time.Therefore,the research direction is determined as neural network combination model.(2)Preprocessed the raw data and potential passenger flow influencing factors.Firstly,external factor data near the site needs to be collected.Then,according to the time granularity and subway station points,the original data of Nanchang Metro is counted,and the data is integrated with weather data and bus passenger flow data in chronological order,which provides data support for the subsequent analysis of spatiotemporal characteristics of passenger flow.Finally,based on the passenger flow data,the smoothness of the passenger flow data of each station under different time granularities and the correlation between the passenger flow data and the potential influencing factors are analyzed.(3)The spatiotemporal characteristics and descriptive analysis of the passenger flow data of Nanchang subway station points were carried out.The analysis of time characteristics takes the inbound passenger flow of each station on weekdays(Monday)of Nanchang Metro Line 1as an example,mainly analyzes the inbound passenger flow trend and passenger flow scale of the station,and divides the stations into single-peak station and double peak station according to the passenger flow trend.The spatial characteristics analysis takes the functions of the surrounding areas where the three subway line stations are located as an example,mainly analyzes the population density of permanent residents and the surrounding land use types around the stations,and divides the stations into residential area stations,commercial area stations and mixed area stations according to the surrounding land use types.(4)The two-step clustering method to cluster the passenger flow and passenger flow trend of the site.Firstly,the passenger flow index was used to cluster the passenger flow trend of 72 stations in Nanchang Metro in the first step,and a total of 5 categories were obtained.Then,the second step clustering of the 5 parent classes was carried out by using the passenger flow index,and a total of 9 sub-categories were obtained.Finally,the differences in the characteristics of passenger flow in the time dimension and spatial dimension of different sub-stations are analyzed,so as to provide data support for subsequent model training.(5)The LSTM short-term passenger flow prediction model and the GCN-LSTM short-term passenger flow prediction model based on clustering.Firstly,the model uses the same subclass of data for model training to provide more data for model training.Then,GCN and LSTM are combined,and the spatiotemporal features of the passenger flow data are extracted by the combined model.Finally,the passenger flow data of 72 stations of Nanchang Metro is used as an example for predictive analysis.The results show that the two-step clustering method can classify the stations according to the spatiotemporal characteristics of the passenger flow data,the external factors can improve the prediction accuracy of the model,and the GCN-LSTM combination model considering the topology between subway stations can effectively extract the spatiotemporal features of short-term passenger flow,so as to further improve the short-term passenger flow prediction accuracy of metro.