Research On Urban Rail Transit Passenger Flow Prediction Based On Deep Learning

With the advent and development of the 21 st century,people’s demands for quality of life and material spirit are getting higher and higher,and at the same time,the requirements for travel are getting higher and higher.The travel mode of residents is gradually moving towards the field of public transportation,and rail transit,as the most important mode of transportation in the city,is developing rapidly.With the construction of urban rail transit,its passenger flow and network structure are becoming more and more complex,which brings great risks to operation management.Effectively predicting the long-term and short-term passenger flow of rail transit can remind people of travel conditions,and can also help relevant departments operate rationally to improve the efficiency of urban construction,which is of great significance.Therefore,from the perspective of rail transit passenger flow,this paper has read domestic and foreign literatures related to it in recent years,and summarized some research methods and specific contents.On the basis of thinking and hands-on experiments,New York City and Beijing were selected as the research objects to study the changing laws of their passenger flow,and build different long-term and short-term prediction models for different cities.Due to the particularity of the data,the New York City rail transit passenger flow forecast focuses on the long-term passenger flow forecast,and the Beijing rail transit focuses on the short-term passenger flow forecast.Design and build the Rf-1DC-L model based on random forest(Rf),one-dimensional convolutional neural network(1DCNN),long short-term memory network(LSTM)and long short-term memory network,graph attention network(GAT)The L-Ga model.The specific research contents include:(1)Use the characteristics of the two cities to study and process the data respectively.The New York subway builds a time series with a granularity of 4 hours,and Beijing builds a time series with a granularity of 5 minutes.(2)The Rf-1DC-L model mainly incorporates additional factors such as wind,temperature,and humidity for modeling,uses random forests to capture useful features,and one-dimensional convolutional neural networks to extract spatial features,and then uses LSTM for model prediction.(3)In the L-Ga model,the main focus is on the capture of spatial relationships,and the GAT graph attention network is added to the LSTM model unit for spatial connection,thereby improving the model accuracy.