Passenger Flow Forecasting Approaches For An Urban Rail Transit System Based On Deep Learning

With the construction of urban rail transit and the continuous expansion of the network scale,rail transit has become the backbone of urban public transport.It has played an important role in improving the quality and efficiency of urban public transportation supply,alleviating urban congestion and improving urban environment.Timely and accurate ridership prediction of urban rail transit is the premise to solve traffic congestion and optimize the line network.It is more convenient to prepare the passenger flow organization in advance by using the passenger flow prediction results and implement measures such as ridership limiting and dredging in advance.With the wide application of automatic fare collection(AFC)system,the management department has obtained a large amount of historical travel data and established a passenger flow database of urban rail transit.The rapid development of big data and deep learning technology provides a good theoretical method for solving the problem of urban rail transit passenger flow prediction.In this paper,the inbound and outbound ridership of the station is taken as the research object.Combined with the temporal-spatial characteristics of passenger flow,this paper proposes a deep learning-based urban rail transit ridership prediction method,which improves the long short-term memory(LSTM)network.This paper constructs SP-LSTM model for short-term outbound passenger flow prediction and Wave-LSTM model for inbound passenger flow prediction.On the premise of the same data set,the prediction performance of different models is compared,which validates the effectiveness of the proposed models based on deep learning.The main work and achievements of this paper are as follows.(1)SP-LSTM,the short-term outbound passenger flow prediction model for urban rail transit hub station has been established.This paper analyzes the ridership of urban rail transit from the two dimensions of time and spatial.It theoretically demonstrates that the outbound ridership of a certain station has a great relationship with the historical ridership of its related stations.Therefore,this paper proposes to use the multi-dimensional historical passenger flow of multiple stations as the input.Two impact indicators are proposed to determine the impact relationship between each station.This section improves the LSTM model that can only input one-dimensional series into an SP-LSTM,which can input multi-dimensional series.The input of SP-LSTM can be changed according to the actual situation and the model has good ductility.(2)A Wave-LSTM model for short-term inbound passenger flow prediction of urban rail transit has been built.Aiming at the issue of short-term inbound passenger flow prediction,this paper combines wavelet analysis and LSTM to build a combined model Wave-LSTM.It integrates the advantages of wavelet analysis and LSTM.Wavelet decomposition and reconstruction can effectively deal with the data volatility and achieve the purpose of extracting features,LSTM can learn from the long-term dependence of ridership data.The real non-linear and non-stationary passenger flow data is processed by wavelet transform,and the time series with more stable variance is obtained.Then LSTM is used for deep learning and prediction,thereby greatly improving the prediction accuracy.(3)Taking Beijing urban rail transit as the research object,the case study is carried out.Based on the actual AFC data of Beijing urban rail transit in 2017,the inbound and outbound passenger flow of several stations in the Airport Line are predicted.The autoregressive integrated moving average(ARIMA),nonlinear autoregressive model(NAR)and general LSTM model were used for the comparative experiments.The experimental results show that the SP-LSTM and Wave-LSTM,which are constructed in this paper,have better prediction effects than the ordinary LSTM,ARIMA and NAR models.There are 43 pictures,18 tables and 84 references.