Abstract
Fully automated operation (FAO) concentrates the critical risks of urban rail transit within the operations control center (OCC), posing unprecedented demands on the emergency response capabilities of dispatching personnel. Existing research, predominantly focused on traditional operational models, has largely overlooked the quantitative assessment of human factors, such as dispatcher cognition and decision making in FAO systems. To address this research gap, this study establishes a human factors–based evaluation framework for FAO dispatcher emergency capabilities, encompassing five dimensions: resource planning, situational awareness, organizational management, information processing, and workload processing capability. An empirical analysis was conducted using the “7.20” extreme rainstorm disaster on the Zhengzhou Metro as a case study. The results indicated that the overall emergency response capability of dispatchers was at a moderate- to high risk level. Specifically, resource planning capability was identified as the most critical vulnerability, exhibiting the highest risk intensity and most significant systematic deficiencies. Furthermore, notable fluctuations and shortcomings were identified in situational awareness, organizational management, and information-processing capabilities. Workload processing capability, although exhibiting a moderate to high risk level, displayed the lowest entropy among all dimensions, indicating that workload-related vulnerabilities were uniformly and systematically experienced across roles and shifts. This study precisely identifies key human-factor vulnerabilities in FAO emergency response, providing a data-driven foundation for optimizing dispatcher training programs and enhancing the safety and resilience of urban rail transit systems.
Keywords
Introduction
The global adoption of fully automated operation (FAO) in urban rail transit has seen rapid growth recently ( 1 ). By replacing manual on-site tasks with automation, FAO significantly boosts operational efficiency. This shift, however, concentrates critical risks within the operations control center (OCC), placing unprecedented demands on the remote command and emergency management skills of dispatchers ( 2 ). Meanwhile, the growing frequency of global emergencies like natural disasters and public health crises underscores the critical need for robust emergency management of high-tech urban infrastructure. Consequently, research into the emergency response of dispatchers in FAO systems has become a pivotal topic for securing urban safety and resilience.
A rich body of research exists on emergency management for rail transit. Initial studies concentrated on the technical and engineering domains, such as improving signaling system reliability and station security standards ( 3 ). The focus later broadened to organizational processes, with numerous studies examining the evaluation of emergency plans and interagency command structures ( 4 , 5 ). Yet, this research is predominantly rooted in traditional operational models, which rely on on-site staff as a key component of emergency response. It largely neglects a critical vulnerability in the FAO model: when remote dispatchers become the central decision makers, their cognitive limitations and human factors can compromise the entire system. This reveals a clear research gap—the absence of a framework to accurately model and quantitatively assess the cognitive and decision-making performance of FAO dispatchers in realistic, high-stress situations.
China’s urban rail transit has expanded with exceptional speed, exemplified by the city of Zhengzhou in the country’s heartland. By December 2025, Zhengzhou’s metro system comprised 13 lines and 450 km of operational track, placing it 10th in the nation. Crucially, the network features three GoA4 fully automated lines, offering a perfect window into the challenges of FAO. Within these systems, dispatchers act as the “brain” of operations, where their response capability dictates the ultimate safety threshold ( 6 ). On July 20, 2021, a severe rainstorm hit Zhengzhou, causing catastrophic flooding in Metro Line 5. Unlike historical incidents, which primarily involved minor station-level waterlogging, this event was characterized by unprecedented rainfall intensity that far exceeded the city’s flood defense standards. The rarity and severity of this “black swan” event posed distinct challenges to the emergency response system, rendering routine experience insufficient. The tragic deaths of 14 passengers on Zhengzhou Metro Line 5 exposed glaring deficiencies in dispatcher performance, particularly in information collation, risk assessment, and decision-making efficiency—shortcomings that ultimately led to the catastrophic outcome ( 7 ). Current research indicates that human performance in such crises is not isolated but heavily constrained by system architecture. Specifically, the “7.20” incident highlighted how structural vulnerabilities can amplify the cognitive load and decision-making challenges faced by dispatchers. Consequently, developing a robust evaluation system to improve dispatcher emergency competence has emerged as a critical and urgent practical issue.
Although the importance of assessing emergency capabilities in FAO systems is recognized ( 8 ), and various models using methods like analytic hierarchy process (AHP) and fuzzy comprehensive evaluation have been proposed ( 9 ), these conventional tools suffer from two critical defects in the context of rail transit emergencies. First, their heavy reliance on subjective expert weighting undermines the objectivity and repeatability of results ( 10 ). Second, they struggle to manage the “fuzziness” and “randomness” inherent in qualitative assessments, failing to capture the ambiguity of performance that often lies between clear-cut categories. Consequently, existing frameworks are too rigid to model the dynamic human factors at play during real-world emergencies, a limitation especially pronounced in FAO’s human–machine interface (HMI).
To break through these limitations, this study leverages the cloud model as its central analytical tool. The cloud model offers decisive advantages. It replaces subjective weighting with a data-driven process by algorithmically generating digital characteristics (Ex, En, He), which greatly improves objectivity. Furthermore, it masterfully integrates fuzziness with randomness, using the distribution of “cloud drops” to visually and quantitatively represent performance states and their uncertainty across a continuous spectrum. This allows for a much more realistic measurement of a dispatcher’s nuanced and dynamic decision-making under stress.
This study’s primary contributions are threefold. Theoretically, it establishes the first human factors–oriented indicator system for evaluating FAO dispatcher emergency response, shifting the focus from traditional process- and facility-centric frameworks. Methodologically, it pioneers the use of the cloud model in rail transit emergency assessment, solving the key problems of subjectivity and uncertainty that limit conventional methods. Practically, this empirical study of the Zhengzhou “7.20” disaster ( 11 ) not only validates the model but also identifies specific weaknesses in dispatcher performance, offering rail operators concrete, data-driven insights for training and management enhancement.
To achieve the aforementioned objectives, this study was structured as follows. First, we conducted a comprehensive literature review to establish the theoretical foundation, analyzing the challenges posed by FAO to OCC dispatchers and deconstructing the human-factor errors evident in historical incidents like the Zhengzhou disaster ( 12 ). This critical review allowed us to identify the deficiencies in existing evaluation frameworks and propose a more targeted, human factor–oriented indicator system. Second, to address the inherent subjectivity and uncertainty in traditional evaluation methods, we introduced and developed a cloud model–based assessment methodology. This section details the rationale for choosing the cloud model and elaborates on its implementation for quantitatively and visually representing dispatcher capabilities. Finally, we validated the proposed model and indicator system through an empirical case study of the Zhengzhou Metro Line 5 incident. The findings not only demonstrate the model’s effectiveness but also provide actionable, data-driven insights for developing targeted training programs and incentive mechanisms, ultimately contributing to the enhancement of safety and resilience in FAO systems.
Literature Review
Operations Control Center in Fully Automated Operation Metro Lines
FAO represents a crucial direction in the modernization of urban rail transit systems ( 12 ). Its core features include driverless operation, a high degree of intelligence, and automation ( 2 ), as shown in (Figure 1). In FAO metro systems, the OCC serves as the neural center of the entire network, responsible for overseeing and managing its smooth operation. The OCC’s primary responsibilities encompass real-time train tracking, automated adjustments for any schedule deviations, and comprehensive monitoring of critical infrastructure such as signaling systems and power supply. The OCC’s ability to promptly detect and address potential issues is essential for maintaining the reliability and safety of FAO systems ( 13 ).
Automated metro terminology.
With the advancement of FAO technology, the operational efficiency of urban rail transit systems has significantly improved, reducing the need for human intervention and lowering operational costs. However, potential safety hazards, such as equipment malfunctions, network outages, and false alarms, necessitate a further in-depth examination of the evolving role of remote OCC dispatchers in emergency response. Identifying vulnerabilities within the rail transit system and formulating targeted emergency response plans and training programs can enhance the system’s ability to effectively manage various emergencies, such as natural disasters and terrorist attacks ( 14 ). Any system anomalies may trigger cascading effects ( 15 ), requiring the OCC to possess rapid identification and response capabilities for both minor technical glitches and major disasters.
FAO systems operate largely autonomously, with train operations and station management relying entirely on the OCC for real-time monitoring and remote control, guided by predefined scenario libraries and rule-based engines ( 16 ). Furthermore, the OCC facilitates rapid emergency response execution through automated commands and coordination with multiple departments. This coordination primarily includes measures such as train control and line adjustments, as well as passenger evacuation and information dissemination ( 17 ), ensuring the timely availability of emergency resources.
Dispatching Personnel’s Challenges in FAO Metro Lines
In FAO metro systems, dispatchers face unique challenges stemming from high system automation and the absence of onboard personnel. FAO systems rely on complex automation technologies, including automatic train protection, automatic train operation, and automatic train supervision (ATS) ( 18 ). Although these technologies form the system’s core, they demand that dispatchers possess both in-depth understanding of technical principles and proficiency in interface operations to promptly identify and resolve system anomalies ( 19 ). The rapid evolution of technology underscores the critical importance of adopting advanced training methods, such as simulation-based exercises, to enhance emergency preparedness and response capabilities among dispatchers ( 20 ). Inadequate practical experience or insufficient training in emergency scenarios can significantly impair response effectiveness ( 21 ).
Emergency response capability assessment should encompass multiple dimensions, including organizational coordination, contingency planning, resource-allocation, and drill implementation, supported by comprehensive evaluation metrics ( 22 ). This becomes particularly crucial during abnormal operations such as signal failures or train derailment warnings, when timely human intervention is essential. Dispatchers must not only diagnose issues rapidly but also balance operational safety, recovery efficiency, and passenger needs, requiring exceptional decision-making speed and accuracy ( 23 ). The high-stress environment during emergencies may induce cognitive biases or judgment errors, potentially leading to overreliance on system recommendations or delayed decision making ( 24 ). Therefore, developing psychological resilience is a critical competency for dispatchers.
Operational complexity escalates during emergencies due to the necessity for precise coordination among dispatching personnel, maintenance teams, station staff, and external emergency services ( 25 ). Furthermore, the human–machine collaboration paradigm in FAO systems introduces ambiguous accountability boundaries, as dispatchers transition from active operators to system supervisors ( 26 ). Establishing clear safety protocols and responsibility frameworks remains a persistent challenge in accident scenarios ( 27 ). Finally, dispatchers require advanced competencies in real-time data interpretation and decisive action, coupled with technical mastery of complex software systems and communication tools to effectively manage automated operations.
Identification of Dispatching Personnel’s Emergency Response Competence Factors
Emergency response capability is defined as the capacity of an individual or department to rapidly and effectively identify problems, make decisions, and act in the face of unexpected events ( 28 ). The assessment of dispatchers in FAO metro systems requires a multidimensional evaluation framework encompassing performance metrics, safety compliance, and emergency response efficacy. This framework has evolved in recent studies to incorporate physio-psycho-machine-environment-management dimensions, enabling data-driven identification of key competencies through advanced modeling techniques ( 29 ). Such approaches emphasize psychological and physiological factors as primary influencers in emergency handling, enhancing the objectivity of hierarchical evaluations.
Effective emergency resource planning is critical for crisis management, significantly enhancing the efficiency and quality of emergency responses and mitigating disaster-induced consequences ( 30 ). Policy guidance provides a framework and principles to guide resource operations during emergencies ( 31 ). Demand assessment facilitates the evaluation of potential needs during emergencies, providing data support for plan formulation ( 32 ). The resource operation plan is vital for ensuring the effective allocation, utilization, and monitoring of resources ( 11 ). Proactive risk identification capability, enhanced by data-driven Bayesian networks that account for complex accident scenes, mitigates the impact of emergencies and enhances overall preparedness in rail transit systems ( 33 ). However, insufficient global awareness can hinder the effectiveness of emergency resource planning. Understanding global trends and best practices can provide valuable insights and help improve local resource planning efforts ( 34 ). Recent advancements highlight the integration of global trends via systematic reviews, improving local planning by incorporating automation’s impact on resource mobilization ( 35 ).
Building on Endsley’s three-level model of situational awareness—perception, comprehension, and projection ( 36 )—OCC personnel in urban rail systems face constrained situational awareness owing to the enclosed, often underground, environment, which can affect emergency judgment. The ability to perceive accurately within the specific context of rail transit is essential ( 37 , 38 ). This includes challenges such as the misinterpretation of external signals, difficulties in multisource data integration, issues with pattern recognition and situational judgment during the comprehension phase, and potential consequence anticipation and delays in anomaly detection during the projection phase. Furthermore, the characteristic risks inherent in rail transit, such as high passenger volumes, dynamic shock waves in underground environments, and the complexity of signaling and equipment systems, must be considered, as real-time conflict models reveal their impact on safety ( 39 ). Prolonged exposure to HMIs can lead to attentional fatigue or cognitive overload among OCC personnel ( 40 ). Additionally, there are limitations in assessing nonobservable system states, such as crowd panic levels, which can lead to underestimating anxiety propagation rates, particularly during high-density emergency situations. Finally, shift rotation patterns and interdepartmental knowledge disparities can create a risk of misaligned mental models owing to informational asymmetries.
Organizational management capability involves the capacity of managers to effectively structure objectives, tasks, and decision-making processes. Some studies have integrated resources, leveraging multilayer complex networks to model risk progression from historical violations to accidents, to design an evaluation framework for emergency logistics capabilities in rail transit systems ( 41 ). Within this framework, emergency organizational management capacity encompasses five key indicators: scientific decision-making proficiency, overall coordination capability, command and dispatch capability, rapid-response capability, and social mobilization capacity, all essential to ensure the effective execution of decisions. This model assesses the capability of relief organizations to supply these resources, emphasizing the crucial role of decision-making authorities and operational entities in regulating relief material provision and facilitating rapid coordination.
With regard to information-processing capability, information technology plays a significant role: it is a critical element of disaster and emergency management. In the context of the Fourth Industrial Revolution, the integration of information technology in relief efforts, coupled with advanced technical monitoring and early warning systems, has been identified as a crucial success factor ( 42 ). Timely information-processing skills and effective information-sharing capabilities enable dispatch personnel to rapidly acquire and disseminate critical information during emergencies, thereby supporting operational decision making in rail transit. Xu and Gong highlighted that information management capability is a key component of emergency logistics support capacity, encompassing emergency monitoring and forecasting, information collection and analysis, and the development of comprehensive databases ( 43 ). This underscores the significance of information processing in emergency response. Du et al. further posited that information-processing capability involves multiple aspects, such as information acquisition, processing, and transmission, which directly affect the effectiveness of equipment support ( 44 ). Among these, early warning technologies are a proactive component of emergency response, and the effectiveness of early warning systems directly influences the timeliness and efficiency of the response ( 45 ). The timeliness of information acquisition is fundamental to information-processing capability; the speed and accuracy of acquisition directly affect subsequent processing and decision making ( 46 ). Information-sharing capability is an indispensable element of emergency response ( 47 ). Research indicates that establishing effective information-sharing mechanisms can significantly enhance the efficiency of interdepartmental collaboration. For instance, creating unified data platforms and establishing information-sharing protocols can ensure seamless information flow across various departments. Finally, monitoring, tracking, and providing feedback are vital for ensuring the efficient operation of emergency response efforts ( 48 ). Regular monitoring of task progress and timely feedback on issues enable the effective adjustment of resource allocation, ensuring timely task completion.
The capacity to manage work-related stress exerts a significant influence on the emergency response capabilities of dispatch personnel. Under stringent time constraints, dispatchers must rapidly formulate response strategies and issue verbal directives, placing considerable demands on their psychological resilience and decision-making abilities. Furthermore, multitasking capability is essential, requiring dispatchers to effectively manage concurrent task information to address complex emergency scenarios. Within communication-related indicator frameworks for dispatchers, stress management capability emerges as a crucial factor ( 49 ). This encompasses several risk factors, including a lack of job awareness risk, time pressure risk, delayed knowledge and information update risk, parallel task processing risk, and overtime workload risk ( 50 ). (See Table 1.) In summary, recent literature synthesizes these factors through multidimensional and computational approaches, revealing that psychological resilience, automation trust, and real-time data integration are pivotal for updating traditional frameworks, thereby fostering more adaptive emergency competencies in FAO systems.
Competence Factors Identification
Note: ATS = automatic train supervision; CCTV = closed circuit television; OCC = operations control center.
Methodology
The methodological framework of this study integrates AHP with the cloud model. The fundamental theory of the cloud model is based on Li et al.’s study ( 51 ), and AHP originates from Saaty and Vargas ( 52 ). We adapted the AHP–cloud model integration approach from Zhang et al.’s research for our context of urban rail transit resilience ( 53 ). This framework was empirically applied to a case study of Zhengzhou Metro Line 5, for which data were collected through field surveys. This process included the development of an evaluation index system and the selection of interviewees, as detailed in the subsequent sections.
Theoretical Basis of Cloud Model
The cloud model is based on the probability theory of fuzzy mathematics and random mathematics, which can realize the transformation of fuzzy concepts and numerical representations ( 54 ). The cloud model is the uncertainty transformation model between the qualitative concept expressed in natural language value and its quantitative representation. Integration of the cloud model with multicriteria decision-making methods, such as AHP, significantly enhances the accuracy of risk assessment ( 55 ). This method not only considers the complexity and multidimensional nature of the emergency response process but also effectively integrates quantitative data and qualitative assessments, overcoming the limitations of traditional assessment methods in handling such information.
In a cloud model, a set of precise numbers of values,
The eigenvalues of the cloud model are usually composed of three digits
Normal cloud.
The forward cloud generator obtains a series of cloud drops by inputting three digits of the eigenvalues, and the specific steps for calculating the three digits as follows:
Step 1. According to the expected value of
Step 2. The normal random number,
Step 3. The determination degree,
and then the coordinates (
Step 4. Repeat Steps 1 to 3 in turn until a cloud drop is produced.
Then, by calculating
Through the given cloud drop,
Entropy,
Process of Cloud Evaluation Model
In a cloud model, a standard cloud represents the risk ranks of five risk levels ( 53 ), from the lowest to the highest, as a reference object. Furthermore, after determining an indicator’s weight, the eigenvalues of the cloud model can be calculated to draw an indicator-specific cloud, which can be called a “current cloud.” The current cloud represents the eigenvalues of the first-level indicator that is the target of the risk evaluation. Ultimately, locating each of the indicator-specific clouds in the standard cloud shows the risk-level assessment of the five first-level indicators ( 58 ). Here, Python was chosen to achieve such an evaluation based on the cloud model (see steps in Figure 3).
Flow chart of emergency response assessment using a cloud model.
Evaluation Index System of the Dispatching Personnel
The dispatching personnel are responsible for the overall work of the team, maintain effective communication, and ensure the safety of passengers and operations during emergencies. This study, through data collection and analysis, developed an indicator system to evaluate the emergency response competence of dispatching personnel working in the subway station.
Based on the competence factors and explanations presented in Table 1, the five first-level indicators—resource planning (
Evaluation Indices in Dispatching Personnel Competences
Index Evaluation Criteria Based on Cloud Model
To systematically evaluate the emergency response capability of dispatching personnel, this study used a framework consisting of a target layer and a criterion layer, which are dimensions of the evaluation. A core principle in emergency management is the “principle of impact-capability matching,” which posits that response capabilities should correspond to the potential impact level of the risks ( 8 ). Adhering to this principle, we developed a grading scale to quantify the experts’ linguistic assessments of emergency response capability. The capability levels are defined as follows:
The normal cloud model is used to describe the five types of impact level ( 53 ). Because the evaluation criteria have both upper and lower bounds (Cmin, Cmax), the following formula is used to transform the evaluation into a normal cloud matrix model.
Based on previous research (
53
), let
Standard feature cloud image.
Determination of Index Weight
The AHP method (put forward by U.S. operations research scientists Saaty and Vargas [ 52 ]) can divide complex and multiple problems into progressive hierarchical groups according to the dominant relation and the overall goal. The relative weight of each indicator within the group is calculated via pairwise comparison. Then, the overall ranking of the scheme’s relative importance is determined according to the subjective judgment of the decision maker. This method integrates various factors including subjective, objective, quantitative, and qualitative factors through relative scale, which is a systematic analysis method combining qualitative and quantitative factors. This study used AHP to transform the interviewees’ qualitative expression of evaluation indicators into quantitative values, and determined the importance of indicators based on the expression of language sets.
First, the judgment matrix,
where
Then, calculate the weight according to the following formulas:
where
Meaning and Assignment of Scaling Method
Finally, undertake a consistency test of the judgment matrix,
In AHP, the consistency of pairwise comparisons is crucial. Where
Create a Cloud Model
After ratings from experts,
where
To compute the eigenvalues of first-level indicators, it is a matter of integrating the second-level indicators. It is a conceptual upgrade that combines two or more clouds into a generalized one. Here, we adopted the virtual cloud comprehensive algorithm ( 59 ) to figure out cloud eigenvalues for the whole project risk evaluation, as shown in Equation 15
Case Study
Case Background
Located in the geographical center of China, Zhengzhou is not only a key city in the central region but also a national comprehensive transportation hub and the core city of the Central Plains Economic Zone. According to data from the Ministry of Transport of China, on February 10, 2023, Zhengzhou Metro Group recorded over 1.5 million passenger trips in a single day, reaching a “million-level” daily ridership. As of December 2025, Zhengzhou Metro had opened 14 lines, connecting the airport, and one tourist city, Xuchang, with a total operational length of 458.53 km. The annual new operational mileage exceeded 100 km, ranking first in China for two consecutive years.
On July 20, 2021, Zhengzhou was hit by an extremely rare and severe rainstorm, with rainfall reaching an extreme level in a short period and causing severe waterlogging across the urban area. At around 6 p.m., the retaining wall (i.e., the flood barrier separating the ground level from the parking lot) of the access line to the Wulongkou Metro Depot of Metro Line 5 was breached by the floodwater, which rushed into the tunnel ( 31 ). This led to a train carrying over 500 passengers being stranded between Haitansi Station and Shakou Station, which are marked with red flags in Figure 5. Ultimately, the disaster claimed 14 lives, plunging many families into sorrow and attracting widespread societal attention. Before the “7.20” disaster, Zhengzhou Metro had only experienced minor local water accumulation, with no historical record of stranded trains or casualties. This event was an unprecedented “super-standard” disaster, far exceeding the design flood protection levels of both the city and the metro system. With regard to preparedness, although the operator conducted annual flood-control drills, these were primarily routine exercises focusing on basic responses such as installing flood baffles at station entrances. Crucially, no high-intensity simulations had been conducted for extreme scenarios, such as floodwater backflow into tunnels or train entrapment. This lack of simulation for black-swan events created a cognitive gap between routine training and the actual catastrophic conditions. Furthermore, the structural failure of the retaining wall was a direct physical cause of the water ingress, resulting from quality issues and the lack of real-time monitoring.
“7.20” flood-related stations and OCCs on Zhengzhou Metro lines.
This severe disaster exposed issues such as inadequate emergency response protocols and deficiencies in operational command and dispatching during the operation of Zhengzhou Metro. As the staff who directly interact with passengers, dispatchers’ behavior, work attitude, and service quality directly affect the daily operations of the stations and the efficiency of emergency response operations, posing ongoing risks to metro operations. Figure 5 identifies Line 5 within the red dashed rectangle, with the specific incident location marked by a red flag; the OCCs are marked as blue circles.
Data Collection and Participant Selection
To construct the cloud models for evaluating emergency response capabilities, this study adopted an expert evaluation methodology. The selection of experts followed a purposive sampling strategy, targeting individuals from the Zhengzhou Metro system with extensive practical experience and in-depth institutional knowledge.
The Zhengzhou Metro system operates through a network of four OCCs. Three of them are currently active, and the backup one remains unfinished This architecture delegates overall network coordination to the Zhengzhou Rail Transit OCC, whereas the Port Area North OCC and Tielu OCC oversee independent line management. This study focuses on the two most representative nodes: the central hub, selected for its networkwide command role, and the Port Area North OCC, chosen for its unique position as a regional center managing the Zhengzhou–Xuchang intercity connection. This selection strategy ensures a comprehensive analysis of both core network and regional intercity command structures.
A total of 27 qualified experts were invited to participate. The selection criteria were stringent, requiring participants to meet at least one of the following qualifications: (1) holding a senior operational title such as Senior Dispatcher or Duty Director; (2) possessing over 5 years of direct experience in emergency management; or (3) having participated in the command structure for at least one major real-world emergency event.
The distribution of these experts was strategic, for example, 12 experts were from the selected OCCs: 8 from the central Zhengzhou Rail Transit OCC and 4 from the regional Port Area North OCC; 15 experts were from key operational lines: 5 each from Line 8 (a new, fully automated line), Line 10 (a mature urban line), and the Zhengzhou–Xuchang metro line (an intercity express line), reflecting the diverse challenges managed by these OCCs.
The expert panel consisted of 25 males and 2 females, with ages ranging from 25 to 50. Each expert provided linguistic assessments for the evaluation indicators, which were subsequently transformed into quantitative cloud models for analysis.
Results
Based on the survey data of the dispatching personnel of Zhengzhou Metro Corporation, the expert scoring method was used to compare the mutual importance of indicators, and Equations 7 to 10 were used to calculate the weight of indicators and to test the matrix consistency by Equations 11 to 13. These results are shown in Tables 4 to 9.
Dimension Judgment Matrix and Index Weight of Resource Planning
Note: CI = consistency test index; CR = consistency ratio.
If CR = 0.01240 < 0.1, then the salary judgment matrix passes the consistency test.
Dimension Judgment Matrix and Index Weight of Situational Awareness
Note: CI = consistency test index; CR = consistency ratio.
If CR = 0.067499 < 0.1, then the work itself judgment matrix passes the consistency test.
Dimension Judgment Matrix and Index Weight of Organizational Management Capability
Note: CI = consistency test index; CR = consistency ratio.
If CR = 0.06219 < 0.1, then the judgment matrix of equity protection passes the consistency test.
Dimension Judgment Matrix and Index Weight of Information-Processing Capability
Note: CI = consistency test index; CR = consistency ratio.
If CR = 0.06502861 < 0.1, then the leader colleague judgment matrix passes the consistency test.
Dimension Judgment Matrix and Index Weight of Workload Processing Capability
Note: CI = consistency test index; CR = consistency ratio.
If CR = 0.06219 < 0.1, then the social value judgment matrix passes the consistency test.
Dispatching Personnel Emergency Response Competence Evaluation Index
Using the reverse cloud generator, the evaluation indices were converted into cloud model characteristics (Ex, En, He), and then combined with the AHP weights (ω) via Equations 14 and 15 to form the target-layer cloud parameters. Here, He was set to 0.5. The cloud parameters for both levels of indicators are shown in Table 9.
In combination with the dimension layer, that is, the standard eigenvalue cloud map, the characteristic values of the first-level indicators (Table 9), that is, the results of the target layer, were rated and visualized. Python here was used to draw the evaluation cloud map of each first-level indicator. Figures 6 to 11, respectively, show the evaluation results of five first-level indicators, and the target cloud is represented by the red cloud map.
Cloud map of resource planning.
Cloud map of situational awareness.
Cloud map of organizational management capability.
Cloud map of information-processing capability.
Cloud map of workload processing capability.
Cloud map of total emergency response competence of dispatching personnel.
Discussion
This study evaluated the emergency response framework of Zhengzhou Metro’s OCC, revealing a medium-high overall emergency capability level (Figure 11). This finding provides a valuable benchmark, but its significance lies primarily in the human factors approach adopted, contrasting with previous research that primarily focused on organizational aspects ( 8 ). Although organizational structure undoubtedly plays a role, this research underscores the critical impact of individual dispatcher performance on the effectiveness of emergency responses.
The traditional mean value method ( 60 ) can only provide a single, deterministic evaluation score, a value that fails to reflect the dispersion of expert opinions or the inherent ambiguity of the results. Structural equation modeling, although a powerful confirmatory factor analysis tool with significant advantages for large-scale survey data ( 61 ), would have faced two major challenges in the specific context of this study. It typically requires a large sample size and struggles to effectively handle the randomness and fuzziness inherent in the assessment language itself.
In contrast, the cloud model is specifically designed to address such “small-sample, qualitative concept quantification” problems. It can directly transform linguistic concepts into numerical characteristics—encompassing Ex, En, and He—making it perfectly suited to the data characteristics and objectives of this research.
To further validate the robustness of our model and its conclusions, a sensitivity analysis was conducted. The weights derived from the AHP, which represent a potential source of subjectivity, were identified as the key parameters for this analysis. We selected the indicator with the highest weight within each of the five dimensions and adjusted its weight by ±10%. The weights of the other indicators in the same dimension were then proportionally adjusted to ensure their sum remained 1. The model was rerun with these adjusted weights. As shown in Table 10, the final evaluation weight exhibited only minor fluctuations and, critically, the overall emergency capability rating remained the same. This result demonstrated that our findings were not sensitive to reasonable variations in expert judgment, thus confirming the reliability and stability of the evaluation framework.
Robustness Test Result of the Key Indicators
The five dimensions—resource planning, situational awareness, organizational management capability, information-processing capability, and workload processing capability—represent key determinants of successful crisis management in the context of urban rail transit. For instance, the dispatcher’s ability to maintain situational awareness directly influences their capacity for rapid and accurate decision making during emergencies. Similarly, effective resource planning ensures that the necessary personnel and equipment are available when and where they are needed most.
To achieve a more nuanced and dynamic assessment than traditional methods allow, this study employed a novel methodological approach. AHP provided a structured framework for weighting the relative importance of each dimension, whereas the use of Python-based cloud models offered a powerful way to represent the inherent uncertainty and variability associated with human performance in emergency situations. Traditional evaluation methods often rely on static, deterministic models, which may fail to capture the complexities of real-world crisis events. The combination of floating and comprehensive algorithms further enhanced the analysis, allowing for a more fine-grained and context-sensitive evaluation of dispatcher job satisfaction and, consequently, their emergency response capabilities. The results presented in Table 9 and Figures 6 to 11 clearly illustrate the varying influence of these five key dimensions. Below are more details of each dimension:
Resource planning capability assessment. The resource planning dimension yielded an expectation value of Ex = 91.18 with a low entropy of En = 1.76, placing U1 firmly within the high-impact cloud and indicating that the deficiency was systemic and tightly clustered across experts rather than dispersed (Figure 6). At the secondary-indicator level, demand assessment mistake (U12: Ex = 93.8, En = 1.30) and policy guidance misleading (U11: Ex = 91.2, En = 1.81) carried the highest risk, suggesting that the OCC consistently misread upstream policy signals and underestimated demand surges such as large passenger-flow events, line interruptions, and adverse weather. Operation plan mistake risk (U13: Ex = 94.2, En = 2.71) was similarly high with concentrated entropy, implying a pervasive, structural deviation in resource-deployment logic rather than an occasional error. Lack of global awareness (U15: Ex = 91.2, En = 2.31) showed comparable severity but with a noticeably higher entropy, reflecting frequent priority conflicts in cross-departmental coordination. By contrast, lack of proactive risk identification capability (U14: Ex = 63.4, En = 2.41) had a clearly lower expectation but high entropy, meaning that proactive identification was not uniformly weak but uneven across roles and scenarios.
Situational awareness capability assessment. Situational awareness sat in the medium-impact cloud at Ex = 60.29, with a notably high entropy value of En = 2.73, indicating that the dispatchers’ awareness was acceptable, on average, but highly volatile across scenarios. Dynamic signal misjudgment risk (U22: Ex = 82.6, En = 3.11) was the most acute weakness, exposing the brittleness of current signal-monitoring and anomaly-detection logic under nonroutine events. Multisource information fusion failure risk (U21: Ex = 74.0, En = 4.01) and latent-state inference failure risk (U25: Ex = 72.0, En = 3.51) both exhibited moderate expectations but the largest entropies in this dimension, meaning that the OCC sometimes integrated heterogeneous data streams like signaling, video, passenger-flow, and weather successfully and sometimes catastrophically—precisely the unstable behavior that black-swan events such as the “7.20” rainstorm punish the hardest. Cognitive overload on the HMI risk (U24: Ex = 72.0, En = 2.01) and spatiotemporal model deviation risk (U23: Ex = 71.6, En = 1.91) were moderately high, but with more concentrated entropy, suggesting that interface load and spatiotemporal model errors were persistent baseline issues. Team mental model misalignment risk (U26: Ex = 65.8, En = 2.21) was the strongest subitem, indicating that tacit shared cognition across departments was preserved. Recommended interventions include hierarchical alarm prioritization to relieve U24, AI-driven predictive analytics for latent-state inference (U25), and digital-twin replay of spatiotemporal evolution to compress U22/U23 uncertainty.
To address these issues, it is recommended to enhance latent-state inference capabilities through AI-driven predictive analytics and employ digital-twin platforms to simulate spatiotemporal event propagation. Furthermore, integrating the findings of this study into existing, regularly conducted cross-functional collaborative scenario drills could reduce the risk of team cognitive dissonance.
3. Organizational management risk assessment results analysis. Organizational management returned the lowest first-level expectation, Ex = 53.84 with En = 1.84, locating it near the low- to medium-impact boundary with relatively concentrated entropy. The narrow envelope around a moderate expectation was itself a finding: the OCC’s organizational weaknesses were stable and broadly recognized, not transient. Command and coordination capability risk (U31: Ex = 65.8, En = 1.81) was the most pronounced subitem, pointing to chronic flaws in cross-departmental resource scheduling and task allocation under emergency rules. Fast response capability (U32: Ex = 64.2, En = 4.21) carried the second-highest expectation but by far the highest entropy in the dimension, meaning rapid-response performance collapsed unpredictably in dynamic scenarios and was highly contingent on individual judgment rather than institutional reflex. Lack of interdepartmental communication efficiency (U33: Ex = 62.0, En = 2.01) and lack of team collaboration efficacy (U34: Ex = 61.8, En = 2.21) formed a tightly clustered mid-band with elevated entropy, reflecting information silos and feedback delays—particularly the communication gap between dispatching and engineering departments. Lack of command comprehension and dissemination (U35: Ex = 62.0, En = 1.50) had the lowest entropy in the dimension, suggesting that the loss of command fidelity during downward transmission was a stable, structural defect. Mitigation should employ social-network analysis to locate communication bottleneck nodes, an adaptive contingency-plan engine to stabilize U32, and structured cross-functional drills to repair U33/U34/U35 in tandem.
4. Information-processing capability assessment results analysis. Information processing (Ex = 78.87 with En = 3.37) was in the medium-high-impact cloud but with the largest entropy among all five dimensions. Its risk was severe, on average, yet highly heterogeneous across subitems, and this dispersion was the dimension’s key signal. Task monitoring and feedback mechanism (U44: Ex = 81.4, En = 2.61) was the most pronounced subitem, indicating that closed-loop monitoring of dispatched tasks and timely feedback were systematically lacking. Alert response failure (U41: Ex = 79.6, En = 2.31) was the second-highest, plausibly driven by chronic “alarm fatigue”: a large proportion of the system alerts was false, misdirected, or noncritical, forcing dispatchers to rely on personal experience to triage and verify. Data acquisition latency (U42: Ex = 76.2, En = 1.30) was moderately high with the lowest entropy in the dimension, signaling that acquisition lag was a pervasive, baseline bottleneck—most likely traceable to legacy field equipment that had not been upgraded. Information-sharing impediments (U43: Ex = 74.2, En = 6.32) displayed the highest entropy of all indicators across the entire model, which we interpreted as an organizational–cultural artifact: interdepartmental data-sovereignty disputes cause some incidents to share information seamlessly while others stall completely. Concretely, the OCC should (a) replace nuisance-alarm streams with an AI-filtered alert layer to attack U41, (b) deploy closed-loop task-tracking dashboards for U44, (c) upgrade acquisition infrastructure under bounded resource constraints for U42, and (d) adopt a federated-learning/anonymized-sharing scheme together with cross-organizational drills to compress U43.
5. Workload processing capability assessment results analysis. Workload processing (Ex = 78.88, En = 0.83) was a medium-high expectation paired with the lowest entropy of any first-level indicator. The interpretation was striking: workload-related risk was not only high but uniformly experienced across roles, scenarios, and shifts, making it the most systemic of the five dimensions. Multitasking overload (U54: Ex = 82.6, En = 2.91) was the most acute subitem; in peak hours, when train headways compressed to 2 to 3 min, concurrent task execution dispersed cognitive resources and inflated error rates ( 62 ). The relatively high entropy further signals that the manifestation of multitasking overload varied with role and task complexity, requiring dynamic resource-allocation strategies rather than a single fix. Excessive work-hour exposure (U55: Ex = 80.2, En = 1.30) had the second-highest expectation but a markedly low entropy, indicating that the corrosive effect of chronic overtime on decision making was pervasive and consistent, as documented in occupational health research linking chronic overtime to degraded decision making and fatigue accumulation ( 63 ). This risk was controllable only through institutional limits on working hours rather than individual coping. Task awareness deficiency (U51: Ex = 78.6, En = 2.11) carried a moderate expectation but elevated entropy, reflecting a multilevel mismatch in what “the task” means: frontline staff attributed deficiency to unclear responsibility boundaries whereas management framed it as insufficient skills training. Outdated knowledge/information (U53: Ex = 78.6, En = 0.60) and time constraint overload (U52: Ex = 77.0, En = 0.50) presented low entropy and moderate-high risks. The former was partially buffered by standardized procedures, whereas the latter was mitigated by intelligent scheduling tools. This finding contrasts with earlier studies that considered time pressure a core stressor ( 64 ). Their persistence nonetheless warrants flexible task-allocation and continuous knowledge-update mechanisms. Together, U5 demands a coordinated package: cognitive-load-aware task dispatching for U54, statutory caps and recovery protocols for U55, role-clarification manuals plus contextualized training for U51, and continuous knowledge-refresh pipelines for U52/U53.
6. Policy and managerial implementations for a resilient OCC. The findings of this study culminated in a set of urgent policy and managerial implications that address the systemic challenges crippling the OCC’s emergency response. Our analysis revealed a system caught in a dangerous equilibrium: on the one hand, dispatchers are overwhelmed by cognitive burdens, evidenced by “alarm fatigue” from false system alerts (U4) and difficulties in interpreting multisource data (U2). On the other hand, this overload is paradoxically alleviated by a rigid, rule-based culture that, while offering a psychological “safe harbor” from blame, paralyzes effective response to unforeseen events. This paralysis is exacerbated when safety protocols are designed by top-level management who are disconnected from on-site realities, forcing frontline dispatchers to navigate the conflict between rigid rules and dynamic disaster evolution independently. This creates a vicious cycle where a lack of predictive accuracy (U1) and fragmented cross-departmental communication (U3) are not merely technical flaws, but symptoms of a deep-seated, risk-averse institutional culture. To break this cycle and forge a truly resilient OCC, a multipronged reform is essential. At the institutional level, regulatory bodies must establish a “safe harbor” or “duty of care” legal framework, shifting the focus from blame avoidance to resilient performance and thereby fostering the psychological safety needed for proactive decision making. Concurrently, metro operating companies must spearhead a coordinated internal transformation. This begins with immediate technological empowerment by deploying an AI-driven decision support system integrated with a digital-twin platform, which will filter “nuisance alarms” and enhance predictive capabilities. This must be coupled with managerial authorization through a clear “dynamic authorization matrix” to empower dispatchers with scalable authority. In the long term, this cultural shift must be cemented by reforming performance appraisal systems to reward proactive crisis management and sound judgment over mere procedural compliance. The evaluation model developed in this study can serve as a foundational tool for these reforms, providing an objective basis to assess and cultivate the very capabilities this comprehensive approach demands, ultimately transforming the OCC from a passive rule-follower into an agile, decisive, and intelligent command center.
Conclusion
This study applied the cloud model to the assessment of emergency response capability gaps among fully automated metro dispatchers, thereby revealing the core characteristics of Zhengzhou Metro’s operational resilience. The empirical results indicated that resource planning capability exhibited the highest risk intensity and relatively low entropy, reflecting highly consistent systemic deficiencies in demand anticipation, policy transmission, and resource operation planning within the OCC. Situational awareness capability sat at a moderate level but with considerable fluctuation, exposing the inadequacy of static monitoring systems in meeting the demands of dynamic adaptation; notably, dynamic signal misjudgment and multisource information fusion performed extremely unstably under nonroutine scenarios. Organizational management capability recorded the lowest expectation value among the five dimensions, with rigid command coordination and rapid response being heavily contingent on individual on-site judgment in dynamic environments, forming prominent weaknesses. Information-processing capability, although overall at a medium-high level, exhibited the highest entropy among the five dimensions, indicating strong heterogeneity across subitems. Alert response failure and task monitoring and feedback mechanisms represented persistent deficiencies, whereas information-sharing impediments displayed the highest entropy value in the entire model, rooted in organizational cultural factors such as interdepartmental data-sovereignty disputes. Furthermore, workload processing capability, despite a moderate expectation value, registered the lowest entropy among all five dimensions, indicating that workload-related risks were uniformly experienced across roles, scenarios, and shifts. Among its subitems, multitasking overload and excessive work-hour exposure were particularly pronounced, constituting the most systemic ergonomic weakness across the five dimensions. The aforementioned deficiencies in organizational coordination and information processing also corroborated the real-world problems highlighted by the 2021 Zhengzhou rainstorm disaster, most notably cross-departmental coordination breakdowns and difficulties in graded-warning classification and disposal.
Drawing on these research findings, lessons and targeted optimization measures can be formulated across two dimensions: strategic management and operational execution. At the strategic management level, a systematic emergency response plan tailored to black-swan extreme events should be established, breaking away from the conventional mindset of static flood-control planning and routine operational management. Low-probability, high-consequence disaster scenarios such as extreme flooding should be incorporated into routine risk assessment frameworks. Communication channels between top-level strategic deployment and frontline execution should be unblocked, cross-departmental collaborative linkage mechanisms strengthened, emergency resource-allocation procedures streamlined, and hierarchical management disconnects eliminated. At the operational execution level, specialized capability training should be provided for OCC dispatchers, with sustained attention paid to their daily physical and mental well-being. Training should focus on strengthening proactive risk assessment and real-time situational awareness, incorporating AI-assisted simulation tools for emergency tabletop exercises. These measures will comprehensively enhance emergency resource allocation, multisource information integration, and graded-warning classification and disposal capabilities under extreme disaster conditions, thereby fundamentally reducing the probability of casualty incidents.
Beyond the immediate research context, the emergency response capability assessment framework developed in this study could be further extended to major cities nationwide and comparable metro systems worldwide. Multiregional horizontal comparative assessments could be conducted to continuously validate the applicability and practical value of this evaluation model, thereby synthesizing universal improvement strategies and differentiated optimization plans for metro emergency capability development across diverse operational environments.
Footnotes
Acknowledgements
The authors extend their sincere thanks to the Zhengzhou Railway Vocational & Technical College and Henan Engineering Research Center of Rail Transit Operation Safety and Transport Efficiency Technology and Application lab for providing the research platform, and to the Humanities and Social Sciences Research Team of Henan Province for their generous project funding, which made this study possible. Our gratitude also goes to all the mentors who offered their help during the course of this research.
Author contributions
The authors confirm contribution to the paper as follows: study conception and design: F. Yuan, Y. Zhang, L. Wang; data collection: P. Wang; analysis and interpretation of results: H. Zhu; draft manuscript preparation: Y. Zhang and M. Liang. All authors reviewed the results and approved the final version of the manuscript.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by Henan Soft Science Research Project (262400410403), Henan Province Educational Science Planning Project (2023YB0481), Humanities and Social Sciences Project of Education of Henan Province (2026-ZDJH-110), Scientific Research Project of Zhengzhou Railway Vocational and Technical College (23KJCXTD04), Henan University Philosophy and Social Science Innovation Team Funding Project (2020-CXTD-12; 2024-CXTD-10), and Philosophy and Social Sciences Planning Project in Henan Province (2024CZH022).
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.