Study On Passenger Flow Statistics And Prediction Of Transfer Path In Urban Rail Transit Station

As the economy continues to develop and the urbanization process advances,the urban population continues to expand,and resident trips increase accordingly.Urban rail transit has become an important choice for residents with its outstanding advantages.With the increasingly perfection of urban rail transit network,the number of transfer stations in the network has increased year by year.The passenger flow in the transfer routes within the stations is large and the transfer routes are complicated.How to use pedestrian detection to realize the real-time statistics of the passenger flow of the transfer path in the station,grasp the characteristics of the passenger flow of the transfer path and make a short-term prediction,provide the basis for the passenger flow control according to the predicted data of the path passenger flow;these three aspects are of great significance.Firstly,by analyzing the transfer mode and the composition of passenger flow in the transfer path,this paper determines the conditions that the research scene of this paper needs to meet;it focuses on which feature extraction algorithm is selected,and based on the characteristics of urban rail passenger flow and research results of pedestrian detection at home and abroad,using the gradient direction histogram(HOG)and support vector machine(SVM)to build a pedestrian detection model combined with the continuous adaptive mean shift algorithm(Cam Shift)to design and implement the transfer path passenger flow statistical method.Then,based on the obtained passenger flow time series data,wavelet analysis is introduced to perform multi-scale analysis on the passenger flow time series data,and it is found that the transfer path passenger flow time series in the research scene exhibits different characteristics in different frequency domains.For prediction accuracy,selecting ensemble empirical mode decomposition(EEMD)to decompose time-series data into components in different frequency domains,and using component importance and permutation entropy to determine the time granularity of the prediction and reorganize the components,respectively.Then the maximum Lyapunov exponent is used to prove that each component is chaotic,and reconstructing each component according to the delay time and embedding dimension,so as to mine the characteristics of time series data in higher dimensions;comparing the recurrent neural network(RNN)and long-term short-term memory(LSTM),two common time series prediction models;With consideration of the research on short-term prediction at home and abroad and the chaos of data,this paper builds a long-term and short-term memory prediction model based on phase space reconstruction,and gives means of data set construction,LSTM network construction,model evaluation and adjustment.Finally,using prediction data,an optimization model with the minimum transfer waiting time as the goal is established for the passenger flow of the transfer route.Finally,this paper collects labeled samples in examples,calculates HOG features,trains SVM,uses Python-Opencv to implement Cam Shift tracking,conducts passenger flow statistics in the research scene to obtain passenger flow time series data with a time granularity of 1s;it preprocesses the data and according to the data characteristics,analyzes and constructs different network structures to make predictions,and compares them with a variety of prediction models,showing that the model has higher prediction accuracy,lower complexity of the model structure,and less over-fitting risks.Finally,MATLAB is used to calculate optimization model of passenger flow control.