Research On OD Estimation Method Of Urban Rail Transit Based On Machine Learning

With the continuous development of urban rail transit,residents are more and more inclined to choose subway travel,coordinating millions of passenger flows every day has become the most difficult and urgent problem needed to be solved in rail transit operation management.Traffic OD(Origin-Destination,OD)can intuitively reflect the traffic demand between the departure and arrival points of all passengers in the urban rail transit network.With the rapid expansion of the scale of urban rail transit lines and the rapid increase of station sites,the explosive growth of OD passenger flow matrix and passenger flow data has brought about.It is not difficult to apply the traditional OD passenger flow matrix prediction method research to the current field of domestic urban rail transportation.The emergence of machine learning solves the problem of data explosion caused by the explosive growth in many industries.Based on this,a OD passenger flow estimation model for urban rail transit based on machine learning is put forward.Firstly,space character and time character of rail transit passenger flow distribution are summarized in this paper,then,by analyzing the data of passenger.It is found that the OD passenger flow matrix group also corresponds to the spatial distribution characteristics and time advancement features.On this basis,the composition of the fully convolutional neural network and found that the fully convolutional neural network is capable of grasping the whole and detail characteristics of the space is analyzed,but it cannot handle the time sequence features;Besides,based on the analysis of the principle of memory neural network,long-short-term memory neural network is found that it is suitable for processing time sequence information.Thus,a combined model based on fully convolutional neural network and long-short-term memory neural network is proposed,which used to estimate the OD passenger flow matrix.The model firstly convolves the OD passenger flow matrix data by the convolutional layer of the fully convolutional neural network to extract the spatial characteristics of the OD passenger flow matrix;secondly,the long-term and short-term memory neural network performs sequential sequence processing to learn the periodicity of the OD passenger flow matrix;Finally,deconvolution is performed on the feature data after the dual processing of space and time to complete the prediction.Finally,take the actual OD passenger flow data of Nanjing rail transit as research object and simulated to verify the model.The original data is divided into six data sets according to three characteristic days(regular passenger flow,working day passenger flow,holiday passenger flow)and two kinds of passenger flow granularity(15min interval,30 min interval),and model parameters of each data set has been adjusted according to periodic features.Two error indicators are used to compare and analyze the overall forecast results of the model with different characteristics and the actual OD passenger flow.Under all conditions,the forecast error of the model is kept within 15%.The data of the prediction results show that the prediction effect of this model is relatively great,so it can meet the operational management needs of urban rail transit.