Short Term Passenger Flow Prediction Of Rail Transit Based On Deep Learning

With the further acceleration of China’s urbanization and sustained economic development,traffic congestion is becoming increasingly serious.As an environmentally friendly,safe and efficient public transport mode,urban rail transit has been regarded as the preferred mode of transportation by more and more citizens.As the passenger flow of urban rail transit is large and has the characteristics of periodicity,it is of great practical significance to timely and accurately predict the passenger flow in and out of the station,so as to make more reasonable operation plans and improve the service quality of urban rail transit.This paper starts from the method of studying the short-term passenger flow of urban rail transit in recent years,reviews the related literature about short-term prediction at home and abroad,summarizes the related research methods.It is found that although the current research methods have their own advantages,most of them ignore the influence of weather,date and station,and fail to predict the number of people in and out of the station separately.In spite of the prediction method has achieved good results,but there are some deficiencies,on this basis,this paper put forward a more accurate based on deep learning of urban rail transit traffic short-term traffic forecasting method.The main research contents are as follows:(1)Multi-source data acquisition and processingIn this paper,AFC data and weather data of Hangzhou urban rail transit stations are collected.First of all,preprocess the data set,and sort out the passenger flow data of each station of Hangzhou urban rail transit from the original AFC data every 10 minutes.Then the passenger flow data and weather data are integrated according to time,which provides data support for the subsequent passenger flow spatial and temporal feature analysis.(2)Spatial and temporal characteristics of passenger flow in and out of urban rail transit stationsBased on the passenger flow data,firstly,the passenger flow in and out of different stations is counted from the time dimension,and the distribution difference and periodic change rule of the passenger flow in working days and non-working days are analyzed.After that,the paper analyzes the spatial distribution of passenger flow data from the spatial dimension,and studies the spatial distribution differences of passenger flow of different sites.By analyzing the influence of temporal and spatial characteristics on the short-term passenger flow prediction of the station,a foundation is laid for the prediction of the subsequent model.(3)Establish a short-term passenger flow prediction model for combined deep learningBy summarizing the existing short-time passenger flow forecasting methods at home and abroad,the advantages and disadvantages of different forecasting methods are discussed,and then the deep learning algorithm is introduced into the short-time passenger flow forecasting field to clarify the advantages of the method in the shorttime passenger flow forecasting field.The convolution neural network is used to extract data features,and then multi-dimensional vector is transformed into one-dimensional vector,which is transmitted to long-term and short-term memory neural network for time series prediction.The short-term passenger flow prediction of urban rail transit station based on combined deep learning is established,and the model optimization parameters are found by grid search method.(4)Comparative analysis of short-term passenger flow prediction modelTaking Hangzhou Rail Transit Line 1,line 2 and line 4 as the experimental objects,it is divided into working days and non-working days to predict the passenger flow in and out of the station every 10 minutes.By establishing the evaluation index system of the model prediction results and taking the traditional LSTM and XGBoost as the contrast model,the validity,applicability and accuracy of the combined deep learning model established in this paper are verified.