Research On Subway Short-term Passenger Flow Forecasting Method Based On Deep Learning

The construction of rail transit system is one of the effective ways to alleviate the traffic pressure of urban traffic.Prediction of short-term passenger flow in Metro is of great significance for solving urban congestion,optimization of line network,safety protection of public transport and building a smart city.At present,the prediction of short-term passenger flow in the metro is mainly based on the analysis of passenger flow characteristics combined with traffic theory or based traditional time series methods.In this paper,the prediction of short-term passenger flow is mainly realized by the neural network based on deep learning.The main work is as follows:(1)Based on the smart card transaction data,this paper analyzed the spatial and temporal characteristics of metro passenger flow,and proposed an algorithm for calculating passenger flow of each line.At the same time,this paper proposed a mobile terminal feature data processing algorithm and a data fusion algorithm based on smart card transaction data and passenger mobile terminal feature data collected by metro station.The above algorithm can be used to calculate the passenger flow in the station area.(2)In this paper,author using the clustering algorithm to classify the date and time segments of the passenger flow,and modeling each type of passenger flow separately to improve the accuracy of passenger flow prediction.By constructing a recurrent neural network combined with a short-term passenger flow prediction model of long short-term memory networks,the author realized the short-term forecast of passenger flow in line,in-and-out station and in-station area.Finally,through continuous experiment adjustment,it is determined that a set of optimal super-parameter combinations,which makes the passenger flow prediction more accurate.(3)Finally,this paper introduced other common short-term passenger flow forecasting methods,and compared the prediction performance of the Moving Average model,SVR model,PROPHET model and Neural Network model under using the same data set.The results show that the short-term passenger flow forecasting model based on in-depth learning is more effective.Due to the non-linear and time-varying characteristics of metro short-term passenger flow,this paper chooses neural network based on deep learning as the prediction model by analyzing the characteristics of passenger flow.In order to predict the passenger flow in metro line and in-station area,this paper propose an algorithm for calculating passenger flow separately on line and a data fusion algorithm for calculating the passenger flow in-station area.Finally,a short-term passenger flow prediction model based on cyclic neural network and long-term and short-term memory network is constructed.