Scenario inference model of urban metro system cascading failure under extreme rainfall conditions

Highlights

• Timely shutdown of subways effectively reduces casualties.

• Persistent rain will increase the economic losses.

• Government guidance of public opinion effectively alleviates social panic.

• The emergency measures can greatly reduce the probability of the outcome events.

• In the case study, the ROC-based CIA-ISM model is more realistic than the BN model.

Abstract

Metro systems have become high-vulnerability entities due to the increasing frequency and severity of urban flooding. Flood events may cause cascading failure to metro systems; therefore, exploring the cascading failure risk of the metro system is a prerequisite for urban flooding prevention and risk management. This study presented a Rank-Order Centroid (ROC) based CIA-ISM (Cross-Impact Analysis, and Interpretive Structural Modeling) method to accurately assess the reliability of emergency management in metro systems under extreme rainfall conditions. We applied this approach to a metro flooding case in Zhengzhou on July 20, 2021. The reliability results show that efficient rescue and timely shutdown notification are the most critical causal events in the cascading failure scenarios. The events of system vulnerability that have the most significant impact on casualties, property losses, and social panic are, respectively, timely notification of the shutdown, humanitarian aid, and public opinion guidance. In forecast scenarios with emergency management measures in effect, the probability of casualties, property losses, and social panic decrease by 96.3%, 58.58%, and 64.28%, respectively. Moreover, a comparison with Bayesian Network (BN) model verified the effectiveness of the ROC-based CIA-ISM approach. Based on the study, we suggest the metro companies release a timely notification of the shutdown. This study can provide scientific data for decision-makers to reasonably develop emergency strategies, significantly reducing flood losses and promoting cities’ sustainable development.

Introduction

Recently, with China's urbanization, many metro lines have been constructed to reduce congestion [1], [2], [3]. Metro systems have been the lifelines of megacities, and hazards or incidents involving metro systems will cause severe problems in cities’ working and living conditions [4]. The metro systems are considered to be the most significant infrastructures of megacities. However, climate change has profound implications for the effectiveness and viability of critical infrastructures (especially metro systems), making the issue increasingly topical [5]. In 2021 alone, many metropolitan areas in China suffered severe metro flooding. For example, from 16:00 to 17:00, on July 20, the hourly rainfall was 201.9 mm and the passenger flow was 967. Floodwater caused by torrential rain encroached on the tunnel from Shakeoulu and Haitansi stations on Zhengzhou metro line 5. The flood in the metro reached people's necks, causing deaths and five injuries among the more than 500 trapped people [6]. The increasing damage due to extreme rainfall events has forced authorities in flood-prone cities to reevaluate their policy regarding the reliability of metro systems to future heavy rainfall events.

Cascading failure [7] is a kind of failure in a system comprising interconnected parts, in which a part's failure can trigger successive parts' failure. Due to the coupling relationships in complex networks, which are interconnected and continuously generate other nodes. Variations in any one node will affect other nodes potentially in the network; in particular, it will affect unexpected nodes that may amplify the damage of the accident. A well-known phenomenon of cascading is The Butterfly Effect. Cascading failure can be predicted by scenario-based methods, otherwise serious side effects can occur if unpredictable. To address the above issue, numerous scholars have studied the cascading failure of various critical infrastructures using the risk assessment methods, such as Bayesian networks (BNs) [8], [9], [10], Directed acyclic graph (DAG) [11], [12], [13], Petri Net (PN) [14], [15], [16], Markov process (MP) [17], [18], [19], etc. Fan et al. [20] proposed a method based on BNs, MP and deep reinforcement learning (DRL) to improve the reliability of supply in natural gas pipeline networks. For a similar problem of pipeline networks [21], the reliability of pipeline networks is optimized by a novel Manifold-based Conditional Bayesian network model. Liu et al. [22] explored the integration of triangular membership and regional center method to SPN for the modeling and evaluation of gas leakage emergency rescue process in gas transmission station. As the traditional critical infrastructure, the hospital is a crucial social component, especially during the spread of the Coronavirus (COVID-19) pandemic. Liu et al. [23] evaluated the seismic resilience of hospital buildings using BNs, considering the cascading relationships on nonstructural components. Silva et al. [24] proposed reliability and availability models adopting stochastic Petri net (SPN) to quantify the impact of energy resources and rejuvenate medical sensor networks' dependability. Similarly, the cascading failure mode of metro systems has been studied widely. Ye et al. [25] proposed a novel grey- Markov prediction model to describe different contribution degrees of historical information of future change trends of system characteristics. Li and Wu [26] proposed a deep reinforcement learning (RL)-based decision support system for stakeholders to optimally manage the critical components of transportation networks to minimize the network-level losses induced by hurricanes. Ghoushchi et al. [27] proposed a novel approach to select the optimal landfill for medical waste using Multi-Criteria Decision-Making (MCDM) methods. Wang et al. [28] applied the CIA-ISM approach to analyze emergency scenarios. Chen et al. [29] discussed the complex interrelationships among the barriers to building-integrated photovoltaics in Singapore. With the rapid development of technology, the emerging infrastructures, such as the Internet of Things (IoT) [30], cloud data centers [31], electric-cyber infrastructure [32], and computerization in supply chains [33], have aroused wild attention.

In this paper, the metro system is identified as a dynamic one impacted by extreme climate. The analysis of cascading process and the decoupling are complex in this case. Scenario-based methods can clearly describe the cascading failure and coupling mode, which can improve the decoupling efficiency and accuracy. In addition, system elements are difficult to quantify. Therefore, they are simplified into events. The system components involved in the cascading process are in the form of events, which reduce the computational dimension.

This study aims to address the research gaps inherent in the previous researches, which are summarized as follows:

1. Cascading failure analysis in the metro system has been conducted based on research with interaction coupling relationships and decoupling among system elements. With the development of extreme climate, the analysis of cascading failure mode becomes difficult in complex systems. Hence, how to properly assess the cross-impact relationships between essential events places an emerging challenge for the research into cascading failure.

2. Emergency measures are critical factors in reducing losses under extreme rainfall conditions. However, when investigating metro flooding risk, few studies have attempted to assess the efficiency of emergency management measures. Therefore, how to precisely analyze the priority and efficiency of emergency measures is the problem to be resolved in this study, which we believe is essential to the cascading failure research on metro systems.

3. In most studies, the CIA-ISM method's event probabilities are determined in two ways: assessed by historical case data and the Delphi method. The former CIA-ISM has accurate results but it is always hard to collect data, while the latter method is efficient in assessment but usually has violent subjectivity. Therefore, it is necessary to develop a scenario-based method that makes a trade-off between veracity and objectivity.

In this study, we focus on the metro systems under extreme rainfall conditions as the unit of analysis and regard excessive rainfall as the external perturbation of the system. Then based on the conventional CIA-ISM model, the ROC-based CIA-ISM model is proposed to infer the cascading failure scenario process in metro systems under extreme rainfall conditions, especially the cross-impact relationships between critical events. Finally, several representative scenario graphs are built to predict the influence of the specific events on the outcome events¹. Furthermore, we adopt the BN model to demonstrate the efficiency of the ROC-based CIA-ISM. The findings can provide suggestions for metro management and emergency repair strategies for the metro system department under extreme rainfall conditions.

Our main contributions to this paper are:

• We adopt the ROC method that was originally developed in multicriteria decision analysis (MCDA) for the elicitation of criteria weight, which turns ordinal judgments into ratio-scale information [[34], [35], [36]]. This type of elicitation only requires experts to provide a ranking of events according to their likelihood, providing a fast and nonnumerical elicitation process. Probabilities are subsequently approximated from the ranking by an algorithm based on the principle of maximum entropy. By using the ROC-based CIA-ISM method, we could fast obtain the initial probability of events. In addition, we could accurately assess the scenario inference process of cascading failure in complex systems and improve their evaluation efficiency.

• The adoption of the ROC-based CIA-ISM approach that we proposed in this article was able to assess the reliability of critical emergency measures in metro systems under extreme rainfall conditions. We had predictably formalized its use for the ranking of emergency measures. In addition, we extended the prediction scale, so it could also estimate the efficiency of emergency management measures in complex systems comprehensively. Furthermore, we provided several special emergency measures, indicating their significance.

• Using a simulation study, we tested the priority of emergency events under three outcome conditions. We compared the method to BN model and find that the proposed method is more accurate than the scenario-based method that we test. We took the ROC-based CIA-ISM approach in practice and adopted it to the “7.20” accident to get the cascading failure process, and gave some advices.

The rest of the paper is organized as follows: Section 2 describes the methodology we developed to perform the cascading failure analysis; Section 3 describes the test case used to demonstrate the applicability of the developed method; The analysis and results are presented in Section 4; Section 5 presents the conclusions.