Research On Short-term Passenger Flow Combination Prediction Model Of Urban Rail Transit Based On Double-Layer Decomposition

In recent years,the rapid development of urban rail transit has effectively alleviated the problem of urban traffic congestion.However,the network operation and overload operation of rail transit have brought new challenges to passenger flow organization and management.This thesis addresses this key issue and studies the short-time passenger flow prediction problem of non-linear and non-smooth urban rail transit,aiming to accurately grasp the spatial and temporal development trend of rail transit passenger flow and promote the refined management of passenger flow.The main research contents are as follows:Firstly,the data of intelligent automatic fare collection(AFC)system of rail transit were pre-processed to analyze the distribution characteristics of rail transit passenger flow from the time level,and it was found that the overall passenger flow distribution is cyclical,with different trends on different days of the week and different times of the day,while different types of stations show The overall passenger flow distribution is found to be cyclical,with different trends at different times of the day and week.The similarity coefficients and hierarchical clustering methods are also used to analyze the temporal similarity of four types of stations,namely residential,service,comprehensive and general,to provide data support for passenger flow prediction.Secondly,for the non-linear and non-smooth characteristics of rail traffic data,the seasonal-trend decomposition procedure based on LOESS(STL)is used to decompose the inbound passenger flow series in a single layer,and the sample entropy and white noise are used to analyze the decomposed components.The analysis is carried out using sample entropy and white noise.Then,we use the empirical modal decomposition method and its improved algorithm to decompose the residual components containing incomplete decomposition information twice,and obtain the improved adaptive noise-complete empirical modal decomposition algorithm(ICEEMDAN)with the best decomposition performance through alignment entropy and reconstruction error analysis.The STLICEEMDAN two-layer decomposition model is constructed to obtain components with higher smoothness and regularity,which effectively reduces the randomness of the original inbound passenger flow sequence.Finally,the long and short time memory neural network(LSTM)with good adaptability to short-time passenger flow is selected as the base prediction model.The time step,activation function and number of layers of the neural network model are determined experimentally,and the number of neurons,learning rate,batch size and number of iterations of the model are optimally searched using the sparrow search algorithm(SSA)to obtain the best combination of parameters of the neural network.The STL-ICEEMDAN-SSA-LSTM combined prediction model is constructed to predict the components obtained from the two-layer decomposition separately,and the final passenger flow prediction results are obtained by superposition.RMSE,MAE and MAPE are selected as the evaluation indexes,and the prediction effects of the models are compared and analyzed in three aspects: different prediction models,different decomposition algorithms and different times.The results show that: in terms of prediction performance,the combined SSA-LSTM prediction model reduces MAPE by1.84% compared with the single LSTM model;in terms of decomposition performance,the combined STL-ICEEMDAN-SSA-LSTM prediction model combining two-layer decomposition further reduces MAPE by 2.04% to 2.94% on the basis of single-layer decomposition;in terms of different time prediction,the prediction accuracy of the combined STL-ICEEMDAN-SSA-LSTM prediction model is higher for peak hours than for flat hours,and for weekdays than for non-working days.In this thesis,STL decomposition and improved adaptive noise-complete empirical modal decomposition algorithm are introduced to decompose the inbound passenger flow sequence,which provides an effective data processing method for rail passenger flow prediction.Meanwhile,the sparrow search algorithm is used to optimize the neural network hyperparameters,which improves the accuracy of the prediction model and provides a certain theoretical method for the operation management of urban rail transit.