Research On Passenger Flow Characteristic Analysis And Short-term Prediction Of Urban Rail Transit Based On AFC Data

With the popularization and development of urban rail transit,the subway has become the travel choice for more and more passengers due to its advantages of low carbon,environmental protection,convenience and speed.However,the increasing passenger flow has led to many problems during the operation of the subway,which requires operators to have real-time control of the subway passenger flow to avoid the occurrence of various hidden risks to the greatest extent.At the same time,with the development of big data technology,a large amount of passenger travel data is stored in subway operation,which provides us with historical passenger flow rules and experience.Under the dual requirements of actual demand and technical means,the short-term prediction of subway passenger flow based on AFC data has high importance and necessity.This study summarizes and analyzes the characteristics and laws of passenger flow,and extracts the key factors that affect the change of passenger flow.Combined with characteristics,scientific prediction of short-term passenger flow in subway is carried out and applied in practice.First,describe and preprocess the multi-source data used in the research to lay a data foundation for subsequent research.Next,describe the distribution characteristics of subway passenger flow from three dimensions: space,time,and other factors.From the spatial dimension,the K-Means method is used to cluster subway stations into four categories,and POI is used to verify the clustering results to analyze the spatial characteristics of passenger flow distribution of different categories of stations.In the dimension of time,the periodic repetition and temporal proximity characteristics of passenger flow are analyzed respectively.From the perspective of other factors,the impact of emergencies and weather factors on passenger flow is analyzed.Then,taking the cluster center as the prediction object,using the Pearson correlation to determine the time granularity,and using the above various features as the input,the short-term passenger flow prediction is carried out.This paper proposes a K-Attention-LSTM deep learning model,which uses LSTM to extract the temporal correlation of passenger flows,and labels the sites based on the k clustering results,so as to mine the spatial correlation between passenger flows of the sites,and then combines external multi-source data,introducing an attention mechanism for prediction.The prediction results of the model are compared and analyzed with statistical methods—ARIMA,basic neural network methods—improved LMBP and other deep learning methods.The results show that when using each prediction method,when the prediction objects are Chegongzhuang and Caishikou stations with relatively few passenger flows,relatively regular passenger flow changes,and relatively small passenger flow fluctuations,the forecasting effects of various methods are all good.However,when the prediction object is the passenger flow of stations such as Beijing Railway Station and Tiantongyuan North Railway Station,which have large passenger flow,significant fluctuations and weak regularity,the prediction accuracy decreases.Among them,the prediction effect of the K-Attention-LSTM model proposed by the research is better than other deep learning models such as traditional LSTM and GRU,the LM-BP basic neural network method and the ARIMA statistical method,especially in the processing of detail fluctuations.It also proves the necessity of introducing external data and the superiority of this method,which can adapt to more variable scenarios.