Prediction Of Abnormal Passenger Flow In Urban Rail Transit Based On Multi-source Data Fusion

The abnormal passenger flow of urban rail transit refers to the situation that the passenger flow of urban rail or the changing passenger flow deviates from the normal scope due to the impact of abnormal conditions such as large-scale events,holidays,extreme weather,and epidemics.The abnormal passenger flow often have a destructive impact on the rail transit system,so we should pay particular attention.Especially in the context of today’s extreme climate,normalization of the epidemic,and frequent flow of people,the psychology of fear the rain and the epidemic has seriously affected people’s travel choices,and the frequency of abnormal passenger flows has far exceeded the past.However,the current urban rail passenger flow prediction models mostly use a single data source to make predictions for conventional conditions,and forecasts under abnormal conditions have inherent defects.Under the premise that only feature learning from historical passenger flow data can no longer meet the demand for predicting abnormal passenger flow in urban rail transit,this paper adopts the method of multi-source data fusion,trying to use the mutual support,supplement,and correction between multiple data sources to predict the target.To comprehensively solve the existing problems of abnormal passenger flow in urban rail and provide a theoretical basis for the construction of the prediction model of abnormal passenger flow in urban rail,this paper conducts the following four aspects of research:(1)Urban rail abnormal passenger flow prediction technology based on multi-source data fusion.This paper measures the difficulty of passenger flow prediction based on a single data source through an abnormal difficulty measurement method.The results prove that it is difficult for a large number of stations to obtain accurate prediction results from a single data source under a large number of abnormal conditions,which proves the limitations of single data source prediction and multi-source data and the necessity of fusion forecasting.This paper combines urban rail operation management expertise,passenger flow feature analysis,data fusion theory,and a deep learning model to propose an urban rail abnormal passenger flow prediction framework based on multi-source data fusion.The framework guides the construction of three different passenger flow prediction models and provides sufficient flexibility and scalability for new data sources or prediction tasks.(2)Research on abnormal passenger flow prediction of the whole network integrating interest point data.Aiming at the holidays and abnormal weather conditions that may cause abnormal passenger flow in urban rail,this paper proposes a networkwide prediction model.In terms of data,we develop the point of interest data as a new data source.Using the three-stage method proposed in this paper,the point of interest data can be converted into site function type indicators for model input.In terms of model construction,we use three modules to construct abnormal disturbance dependencies feature,temporal dependencies feature,and spatial dependencies feature.In the abnormal dependencies feature extraction module,we use CNN to extract the spatio-temporal impact of the disturbance;in the temporal dependencies feature extraction module,we use three GRU networks to fit the temporal closeness,periodicity,and trend patterns respectively;in spatial dependencies feature extraction module We use GCN to extract the complex spatial topological relationship of the urban rail network.Finally,the fusion algorithm is used to fuse the features of the three parts and get the prediction result.The experimental results on the real dataset of Beijing urban rail transit show that the model proposed in this paper is superior to other baseline models.Comparative experiments between different variants show that all three modules are of great significance to the improvement of model prediction accuracy.(3)Research on the prediction of abnormal urban rail passenger flow based on website data.Aiming at large-scale events that may cause abnormal urban rail passenger flow,this paper proposes a critical station prediction model,which can predict future multi-step passenger flow.We use website data as a new data source to extract information related to large-scale events from website data and perform feature representation.In terms of models,we also constructed three sub-modules similar to the previous chapter.We extracted the features of the three modules based on the LSTM network and the fully connected network.We have adopted some feature engineering methods in specific feature selection,such as using dynamic space-time distortion or average travel time to select critical features.Finally,our multiple attention LSTM networks extract the prediction results from the fused features at each step.The experimental results on the real dataset of Beijing urban rail transit show that our method is better than other baseline models.The multi-step prediction results are more accurate and feasible.The appropriate amount of relevant station information can improve the prediction effect of the model.(4)Research on the prediction of abnormal passenger flow in urban rail-based on search engine data.In response to the epidemic situation that may cause abnormal passenger flow in urban rail,this paper proposes a city-wide total passenger flow prediction model.In terms of data,we have developed search engine data to capture the changing trend of passenger flow under the epidemic and obtain key passenger flow predictors through the search-query-mining and feature selection process.In terms of models,we use the SAE-DNN network to learn and predict from the features after multisource data fusion.Experiments on the accurate data set of Beijing urban rail transit show that our model can successfully capture urban rail transit changing patterns under the epidemic,and its prediction effect is significantly better than other models.Based on the above main work,the innovations and main contributions of this paper are summarized as follows:(1)This paper proposes an urban rail abnormal passenger flow prediction framework based on multi-source data fusion.The framework uses the branch conversion fusion strategy to divide the urban rail abnormal passenger flow prediction task into three branches: disturbance dependencies feature,time dependencies feature and spatial dependencies feature.(2)In this paper,a network-wide abnormal passenger flow prediction model integrating interest point data is proposed,which can predict the passenger flow on the premise of considering different station types;In the model,the interest point data is applied to the passenger flow prediction of urban rail transit for the first time.(3)This paper proposes an abnormal passenger flow prediction model of key stations integrating website data.The model integrates website data and can predict passenger flow in multiple time steps in the future based on a multi-task learning strategy;This paper presents a set of methods to obtain information related to large-scale activities from the Internet and express their features.(4)This paper proposes a total abnormal passenger flow prediction model integrating search engine data.As far as we know,this is the first research on passenger flow prediction under the epidemic situations,which fills the research gap in related fields;Innovatively apply search engine data to traffic flow forecasting.