Simulation-based passenger evacuation optimization in underground railway stations with high network-level priority

Abstract

The underground railway system (URS) confronts various risks, significantly influencing its functionality and passenger safety. Unlike other indoor or open areas, underground railway stations are enclosed and confined spaces, which presents unique challenges in emergency response for URSs. Therefore, developing emergency management strategies to investigate the operational status of the URS and ensure passenger safety is of paramount importance. This involves tracking and forecasting passenger flow, identifying high-risk stations, and designing emergency evacuation solutions. Previously, several passenger counting and prediction methods have been proposed, but there is still a lack of effective methods that can provide accurate predictive passenger information for URSs and employ the information to enhance the emergency management of URSs. Besides, the events that may occur within the URS are variable, but existing research mostly concentrated on simulating the evacuation process or optimizing the evacuation performance for specific scenarios. Thus, there is a gap in developing adaptive and proactive evacuation strategies for different scenarios to mitigate the risk of emergencies.

In response, this study aims to develop a synthesized approach to improve the emergency management in URSs. It encompasses (1) establishing an advanced prediction model based on long short-term memory neural network (LSTM NN) and self-attention mechanism (SAM) to accurately forecast and analyze passenger flow within the entire URS; (2) proposing an evaluation method using complex network analysis and multi-attribute decision making (MADM) methods for identifying stations with high risks from a network perspective; and (3) developing a hybrid approach that combines agent-based modeling (ABM), machine learning (ML), and multi-objective optimization (MOO) to reveal how different factors affecting station evacuation efficiency and safety, and devise targeted and adaptable emergency evacuation strategies for stations with high network-level priority. We use the Hong Kong Mass Transit Railway (MTR) system to demonstrate the effectiveness of the proposed model and methods and select Wan Chai station as a case to conduct the evacuation optimization. The MTR passenger flow data, along with geometric data and configuration information of the selected metro station, are employed in this research.

The key results demonstrate that (1) compared with baseline models, the integration of LSTM and SAM can provide more accurate passenger flow prediction results for most stations and most scenarios; (2) the proposed station importance evaluation method can identify critical stations more comprehensively, and can reflect and predict the variations in station importance over time; (3) the integration of ABM, ML, and MOO can deliver optimal evacuation solutions for different scenarios to reduce the evacuation time and additional evacuation risk within a reasonable budget. This research offers a holistic perspective on the dynamics and complexities of emergency situations in URSs and contributes to enhancing the preparedness and responsiveness of URS during emergencies. It explores the interactivity among different existing methods and indicates that the integration of different approaches can compensate for the shortcomings in the management of URSs. Moreover, this research can provide actionable guidelines and recommendations for railway operators, station managers, and relevant authorities of URSs to promote their emergency management capabilities.