Clustering Analysis Of Stations Based On AFC Data And Intelligent Short-Term Passenger Flow Forecast

In recent years,the subway has become the mainstream of transportation in all cities,and increasingly become the main means of passage for urban residents to travel.Short-time passenger flow prediction is an important step in the construction of urban rail transit big data platform.Accurate prediction of daily urban rail transit passenger flow in and out of stations in a short period of time can help improve the scheduling organization and operation efficiency of metro management,thus increasing the coping ability of large passenger flow due to unexpected events.Based on the Automatic Fare Collection System(AFC)data,the thesis analyzes the time series feature dimensions of passenger flow from the perspective of station attribute analysis and establishes a set of adaptive prediction features for passenger flow at metro stations across the network.The main research contents are as follows:(1)A data pre-processing method is designed to address the shortcomings of the current metro short-time passenger flow prediction only at a single station or a single time node,and a time series prediction scheme is designed based on K-means clustering combined with an adaptive long and short term memory neural network model to classify station clusters based on the characteristic dimensions of stations,so that the developed model has the ability to predict panning in similar clusters of data.(2)An adaptive sinusoidal disturbance strategy sparrow search algorithm(ASDSSA)and its detailed mathematical model are proposed to address the problems of slow convergence,low accuracy of finding the optimal solution and the tendency of the optimal solution to fall into local optimality of the current sparrow search algorithm.Firstly,the initial population quality of the algorithm is improved by fusing cubic chaos mapping and perturbation compensation factor methods;secondly,the sinusoidal disturbance strategy is introduced to update the existing mathematical model for computing the discoverer position.The simulation proves that the sinusoidal disturbance strategy improves the information exchange ability of the population and the full domain search performance of the newly proposed algorithm.Finally,the ability of the algorithm to jump out of the local optimal solution is improved by the adaptive Cauchy variation strategy.The simulation results verify the effectiveness of the adaptive sinusoidal disturbance strategy sparrow search algorithm by experimenting with eight benchmark test functions,CEC2017 test functions,and Wilcoxon rank sum test.And based on the proposed adaptive sinusoidal disturbance strategy sparrow search algorithm,the proposed ASDSSA-LSTM passenger flow prediction model is applied to the selection of the number of hidden layer neurons,the learning rate and the number of iterations hyperparameters of the Long Short-Term Memory(LSTM)neural network for passenger flow prediction model.(3)Passenger flow prediction example analysis and verification.Taking a university site in Fuzhou as an example,the simulation results demonstrate that the ASDSSA-LSTM prediction model has the smallest errors in statistical indicators MAPE,RMSE,MAE,and R2,which reduces the impact of the volatility and non-periodicity of classified station data on the model,and further verifies the effectiveness and feasibility of the improved algorithm in practical engineering on the subway passenger flow weekday,non-workday,and holiday data sets.