Based On Neural Network Model Short Term Passenger Flow Forecast Of Urban Rail Transit

With the accelerating process of urbanization in China,a large number of people have poured into the city,which has caused great pressure on the local traffic.The emergence of urban rail transit has alleviated the urban traffic congestion.Because of its large carrying capacity and fast speed,it has gradually become an important part of the urban transportation system.In order to improve the operation efficiency of urban rail transit and ensure traffic safety,rail transit passenger flow prediction is particularly important.Accurate passenger flow prediction of rail transit can help the operation department improve the dispatching work,formulate train operation schedule,and train operation plan,improve the service level of passengers,improve the travel experience of passengers,and consolidate the position of rail transit in urban public transport.Different from long-term prediction,the main purpose of short-term passenger flow prediction of rail transit is to realize rapid and accurate prediction of passenger flow within the specified time granularity.Due to the large difference of passenger volume in different periods of a day,the short-term passenger flow prediction is more difficult,and the influence of non-linearity and randomness is greater.This paper studies and explores the passenger flow of urban rail transit by using the relevant theories and methods of short-term passenger flow.Firstly,after comprehensively analyzing the process and research status of passenger flow prediction at home and abroad,this paper summarizes the advantages and disadvantages of different prediction methods.Because BP neural network model and DBN model have strong data fitting ability and high prediction accuracy,they are selected as the basic passenger flow prediction model.Secondly,it expounds the relevant theories of rail transit and short-term passenger flow,which paves the way for the later data processing and passenger flow prediction.Then,starting from the time and space dimensions and based on the historical data of Dalian metro,this paper analyzes the passenger flow distribution characteristics of urban rail transit,and filters and remove noise the passenger flow data of metro station,so as to screen out the noise data and retain the effective data,taking one week as the cycle,PPMCC(Pearson product moment correlation coefficient)Pearson product moment correlation coefficient and K-means cluster analysis method are used to analyze the correlation of passenger flow data in a cycle,so as to lay a foundation for the short-term passenger flow prediction of urban rail transit below.Finally,the structure and characteristics of DBN model and BP neural network model are described respectively.In view of the shortcomings of BP neural network model,the model is improved,and the input neurons are classified and processed as several independent modules.Taking the Friendship Square Station of Dalian Metro Line 2 as a case study,using the processed data as the training set,a three-layer DBN network is selected through experiments,and the number of hidden layer neuron nodes is 21;The number of input neuron nodes,hidden layer nodes and output nodes of BP neural network are determined through experiments,which are 9 nodes,14 nodes and 1 node respectively;The number of input neuron nodes,hidden layer nodes and output nodes of the improved BP neural network model are changed to 3 nodes,10 nodes and 1 node;Finally,the prediction results of the three models are compared with the actual values.It is concluded that compared with the prediction accuracy of the traditional BP network prediction model,the improved BP neural network prediction model is increased by 11.91% and the DBN neural network prediction model is reduced by 1.79%.