Risk Analysis Of Mass Passenger Flow Events In Urban Rail Transit Considering Temporal-spatial Heterogeneity

Mass passenger flow events are the most common event in urban rail transit,threatening the normal operation of urban rail transit and the safety of passengers.This study aims to assess the risk of mass passenger flow events by quantitatively analyzing their probability and severity.Based on the mass passenger flow event log,passenger flow,and urban weather information,two data environments were constructed,including a continuous data environment and an under-sampling data environment.Mass passenger flow event delay was denoted as the indicator of mass passenger flow event risk,then the Hurdle model was employed to analyze the risk.To account for the temporal-spatial heterogeneity of mass passenger flow events,the Crossed Random Effects Hurdle model was introduced.Based on the binary parts of models,the probability of mass passenger flow events was analyzed and predicted.Based on the count parts of models,the severity of mass passenger flow events,which was quantified by the mass passenger flow event delay,was analyzed and predicted.To identify the most important factor influencing the risk of mass passenger flow events,the Gradient Boosting Machine was employed to calculate the importance of factors.This study will not only provide technical support for the control of mass passenger flow event risk,but improve the management,decisionmaking,and the safety of urban rail transit.The main work of this study is as follows:(1)Cause and consequence analysis of mass passenger flow events.Based on the statistical analysis of historical data,causes of mass passenger flow events were analyzed in terms of urban planning,tidal passenger flow and et al.Possible consequences of mass passenger flow events were analyzed in terms of passengers,facilities,and management.(2)Construction of two data environments.The continuous data environment was constructed by integrating three datasets,which included mass passenger flow event records,passenger flow,and meteorological data.According to epidemiological theories,case and control observations were extracted from the continuous data environment to construct the under-sampling data environment.(3)Analysis of mass passenger flow event risk.The Hurdle model was employed to analyze mass passenger flow event delay.To account for the unobserved temporal-spatial heterogeneity of mass passenger flow events,station-and-time related random effects were added to the Hurdle model to construct the Crossed Random Effects Hurdle model.Results indicate that the Crossed Random Effects Hurdle model outperforms the Hurdle model in both fitting and prediction.In terms of the overall and probabilistic prediction of mass passenger flow event risk,models in the continuous data environment outperform models in the under-sampling data environment.In the prediction of mass passenger flow event severity,models in the under-sampling data environment perform better than models in the continuous data environment.(4)Factor analysis of mass passenger flow events.The Gradient Boosting Machine was employed to compute the importance of influencing factors.Results suggest that inflow,upstream flow,upstream average load,and the standard deviation of upstream average load are the top four important factors.Factors influencing the probability and severity of mass passenger flow events were quantitatively analyzed through the Hurdle model and the Crossed Random Effects Hurdle model.(5)Prevention and control for mass passenger flow events.Based on the result of risk analysis,prevention and control for mass passenger flow events were proposed from four aspects: management,facilities,planning,and education.