Research On The Method Of Short-term Passenger Flow Prediction And Passenger Density Estimation In Metro

Under the situation of large passenger flow development in subway station,how to use effective technical means to accurately grasp the key indicators of passenger flow in subway station,provide decision data for early warning management,and ensure the safe operation of the station,has become the core problem of subway operation.In order to better track the trend of passenger flow,master the passenger flow density in the station.Taking a single subway station as the research object,this paper makes an in-depth study on how to obtain two important indexes of passenger flow: the passenger flow rate and the passenger flow density in the station.A short-term passenger flow prediction algorithm based on AFC data and a passenger flow density estimation algorithm based on video surveillance are proposed.The prediction results of the two algorithms can provide relevant decision data for the early warning of passenger flow in subway stations.The main contents of this paper are as follows:1.Prediction algorithm of subway short-term passenger flow.Based on AFC data,this paper analyzes the spatial and temporal characteristics of subway passenger flow,proposes the short-term passenger flow prediction model GA＿LSTM,and realizes the 10-minute passenger flow prediction.Firstly,this paper uses LSTM network,which is very effective for time series data processing,to build a short-term passenger flow prediction model.Then,aiming at the problem that it is difficult to calibrate the input time step,number of neurons,number of batches of samples and number of iterations in LSTM model super parameters,a genetic algorithm is designed to optimize the super parameters.Finally,an experiment is carried out to predict the passenger flow of hangzhou east railway station on weekdays and non-weekdays.At the same time,it is compared with other prediction models.The results show that the GA＿LSTM model proposed in this paper is more effective than other models in predicting the short-term passenger flow of the station,and the genetic algorithm designed in this paper further improves the prediction accuracy of the LSTM model.2.Estimation algorithm of passenger flow density in metro area.In this paper,based on the metro video monitoring data and taking the whole regional passenger flow as the research object,an estimation algorithm of passenger density and number of passenger flow based on convolutional neural network is proposed.Aiming at the two problems of learning multi-scale features of images and improving the resolution of generated density maps,this paper designed CMSNet,a crowd multi-scale feature aggregation network based on convolutional neural network,and optimized the model by multi-task learning mechanism.In addition,a new scale adaptive truth-density map generation algorithm is proposed.Experimental results show that the proposed algorithm can estimate the number of regional passenger flows and output a high resolution density map on two public population data sets and the S＿ST data set for subway scene.3.Early warning analysis of subway passenger flow.Combined with the sample site,this paper discusses the passenger flow threshold and regional passenger density threshold of the site.Based on the warning threshold system,the forecast results of passenger flow and the estimation results of regional passenger density are analyzed.It provides reference for the early warning work of the metro operation department,so as to carry out the overall safety early warning for the station.