Research Of Short-time Rail Traffic Flow Forecasting Based On Recurrent Neural Network

During the peak hours,the large amount of people gathering makes the congestion of rail transit become normal.Through the analysis of the existing data,the long-term or short-term passenger flow forecast can improve the train operation organization plan.With the continuous updating and improvement of the data acquisition system of rail transit,the passenger flow prediction based on passenger flow data mining is gradually favored by many researchers,but at present,the analysis and prediction of passenger flow characteristics are mostly aimed at onedimensional time series.The passenger flow prediction of one-dimensional time series is easily affected by unexpected events,weather and other external factors,which make the accuracy of passenger flow prediction not ideal.Most of the existing passenger flow forecasts directly adopt the time series modeling method which takes the day as the statistical unit and ignores the variation characteristics of passenger flow at a certain moment.In this paper,first of all,the passenger flow data is divided into vertical and horizontal two dimensions of time series of passenger flow,and based on this research the chongqing common rail transit site and passenger flow change characteristics of transfering hub station,found either transverse or longitudinal time series,time series passenger flow change trend in a particular time period is always similar,and have certain rules to follow,The trend is particularly pronounced during peak hours.Considering the variation characteristics of passenger flow and the factors affecting passenger flow characteristics,the structure of cyclic neural network and Elman neural network were used to describe the data structure of two-dimensional time series,and the short-term passenger flow was modeled and predicted by using the historical passenger flow data of Chongqing rail transit stations from October to December in 2017.Through the prediction of passenger flow of Chongqing railway station during holidays and non-holidays,the prediction results show that the prediction effect of cyclic neural network is the best.It is also found that the prediction effect of inbound passenger flow is better than that of outbound passenger flow in either transfer hub station or ordinary railway station.This research has a certain practical value,through the analysis of the results can optimize the train operation organization to adapt to the change of passenger flow law.This study is not only helpful for making driving plans scientifically,but also an important basis for adjusting operation plans in real time and the change preparation basis for ticket cards and ticket vending machines.Meanwhile,it also provides data support for dealing with emergencies,so as to improve the overall service level of rail transit.