RSRI-Based Modeling of Coal Mine Gas Explosion Accident Causation Networks

Abstract

Coal mine gas explosions remain a major occupational hazard, driven by the interaction of multiple risk factors. In this study, a systematic framework was developed for accident causation analysis and prevention by integrating root–state risk identification (RSRI) theory with complex network modeling. An analysis of 102 accident reports identified 112 primary risk factors, which were incorporated into a causation network. Nodes were prioritized through entropy-weighted TOPSIS, and edge vulnerability analysis was applied to reveal dominant evolutionary pathways. The results indicate that gas accumulation in the heading face constitutes the most critical direct cause, while insufficient safety supervision is the principal indirect driver. The most hazardous pathway involves inadequate ventilation inspection, reduced air supply, gas accumulation, weak supervision, limited safety training, and unsafe blasting practices. These findings underscore the pivotal role of organizational and behavioral deficiencies in risk propagation. The proposed framework advances current approaches to risk assessment by systematically identifying key factors and critical paths, thereby providing actionable insights for enhancing supervision, strengthening preventive strategies, and reducing catastrophic accidents in coal mines.

1. Introduction

As a globally important source of energy, coal plays an irreplaceable role in the economic development of many countries [1]. However, underground coal mining is usually carried out under high-risk and complex environmental conditions and involves many major risks, such as fire, explosion, toxic gas leakage, mine collapse, and water seepage [2]. Although the frequency of coal mine gas explosions is relatively low, they are particularly destructive, often leading to mass deaths and serious property losses [3]. Therefore, gas-related accidents have always been the focus of attention in the field of mine safety. From a global perspective, the risk of gas explosion is a prominent safety hazard in the coal industry. In European countries that still maintain large-scale coal mining activities (such as Poland), research shows that gas and coal dust disasters occur due to many key factors, such as mining, transportation, and coal preparation, and have strong systematic characteristics [4]. Similarly, in mine accidents in Pakistan, mine collapse and gas explosion accounted for about 80% of all deaths [5], indicating that the risk level is extremely high. Coal has remained the primary source of energy in China for decades, as the country leads the world in both coal production and consumption. According to 2019 statistics, China’s coal consumption accounts for 56.6% of the country’s total fossil energy consumption [6], indicating that the coal-dominated energy structure will continue for a long time to come [7]. However, the coal industry has the typical characteristics of high labor intensity, complex working environments, and high accident rates. Coal mine safety has always been the core problem restricting the high-quality development of the industry [8]. Coal mine accidents have occurred frequently over the years; official data from the Ministry of Emergency Management of the People’s Republic of China show that gas explosions account for nearly 70% of coal mine accidents involving more than 10 deaths, establishing them as the dominant cause of serious incidents. Gas explosion is the biggest killer that threatens the safety of coal mining. It not only causes serious casualties and huge economic losses but also leads to a series of negative social effects [9]. As shown in Figure 1, from 2000 to 2024, a total of 3542 gas explosion accidents occurred in China, accounting for 20.7% of the total number of accidents. However, the number of deaths accounted for 39.4% of the total, ranking first. It is therefore urgently necessary to explore the causes of accidents and propose effective risk mitigation methods. In recent years, the introduction of intelligent technology into mining has achieved remarkable results in reducing labor intensity, improving operation efficiency, and strengthening safety guarantees. The coal mine accident rate and death toll have decreased significantly [10]. In 2019, the death rate per million tons of coal mines in China (DRPMT) fell below 0.1 for the first time [11] (the specific data are shown in Figure 2). Although the application of intelligent technologies has significantly improved the safety level of coal mine operations, statistics indicate that three major coal mine gas explosion accidents occurred in China in 2023, resulting in 47 fatalities [12]. This reality highlights the fact that accident prevention remains a core challenge in ensuring coal mine safety.

Figure 1. Statistics of accidents in coal mines in China from 2000 to 2024.

Figure 2. Statistics of death toll and DRPMT in coal mine accidents from 2000 to 2024.

A systematic summary and analysis of the causes of historical accidents can provide valuable insights and effective support for developing preventive strategies. The researchers in [13] statistically examined coal mine accidents in China between 2018 and 2022, analyzing their patterns, causal factors, and relative significance. The results show that the incidence of general accidents in coal mines is the highest, and human factors are the main causes of accidents. In [14], GIS spatial analysis and rescaled range analysis were used to comprehensively reveal the spatial and temporal distribution characteristics and evolution rules of coal mine accidents based on 2269 survey reports in China from 2005 to 2022. These macro-level statistical analyses reveal the general rules of coal mine accidents. However, they often fail to provide an in-depth understanding of the mechanisms and evolution of accidents, which limits their effectiveness in accident prevention. Therefore, some researchers began to seek quantitative methods for use in specific situations, aiming to make up for the shortcomings of macro analysis in revealing the mechanism of accidents. The authors of [15] proposed a risk assessment method based on management supervision, risk tree (MORT), and energy release theory. This method provides a quantitative multi-dimensional risk analysis framework for coal mine water inrush accidents. In [16], the relationships between risk factors were analyzed using risk coupling theory from macro, meso, and micro levels. The researchers constructed a multi-factor coupling risk assessment model for gas explosions based on the C-OWA operator and interaction matrix. These studies have promoted the development of a quantitative risk analysis framework. Nevertheless, they are biased towards post hoc analysis and focus primarily on accident results. Risk management should not be implemented only after injury or damage occurs; instead, it should focus on training, processes, and systematization to address potential future risks. Effective risk management begins with accurate identification, which forms its foundational step. In the mining sector, both managers and workers must recognize potential hazards in their operational settings to minimize the likelihood of accidents.

Therefore, a systematic approach is essential to identify and analyze the underlying mechanisms of accidents, as well as to trace their root cause. Accident causal models such as the 2–4 model [17], HFACS [18], and STAMP [19] are widely used. In [20], the 2–4 model was applied to a coal mine gas explosion accident, revealing the evolution process of gas explosions and the interactions among multiple contributing factors. The researchers in [21] employed an improved HFACS model to classify 29 human factors across the supervisory, managerial, and operational levels, and further validated the clustering of these factors using complex network analysis. A STAMP-Game model was proposed in [22] for accident analysis in oil and gas storage and transportation systems, aiming to enhance the safety of existing infrastructures. Although traditional accident causation models have substantially advanced the identification of contributing factors, root cause analysis, and the understanding of causal mechanisms, they still exhibit several limitations. First, large-scale studies on coal mine gas explosions have mainly relied on subjective approaches, such as surveys and the Delphi method, to determine contributing factors. These approaches are often influenced by the researcher’s personal experience and expertise, which can lead to incomplete analytical frameworks or biased information, thereby affecting the reliability of factor identification. Second, most existing models concentrate on human factors, resulting in factor systems that are not sufficiently comprehensive and may overlook critical triggers of accidents. Third, while causal models elucidate the mechanisms of accidents from the perspective of causation, they only provide limited insight into the complex nonlinear interactions among factors and the dynamic evolution of accidents.

Specifically, a more systematic and comprehensive method is needed to identify accident risk factors, determine key incentives, and analyze risk evolution paths, so as to formulate a more efficient prevention strategy for coal mine gas explosion accidents. Based on this, in this paper, an accident cause analysis framework is proposed that combines root–state risk identification (RSRI) theory and complex network theory. Firstly, the RSRI theory is used to systematically analyze 102 accident reports, and 112 risk factors are extracted. The risks are divided into “root risk” and “state risk”. The internal relationship between them is analyzed in depth, and a systematic risk factor framework is constructed according to the four dimensions of unsafe human behavior, unsafe equipment, unsafe environmental conditions, and unsafe management measures. Secondly, based on complex network theory, a directed weighted causal network model of coal mine gas explosion accidents is constructed, and its topological structure is analyzed to reveal the complex interaction between risk factors. Thirdly, to evaluate the overall significance of each node and pinpoint major risk elements, the entropy weight TOPSIS (EW-TOPSIS) approach integrates four network metrics: degree centrality, betweenness centrality, clustering coefficient, and PageRank. Edge vulnerability is determined by the three indices of edge betweenness, average path length, and connectivity to determine the key evolution path. Finally, the causal network simulation reveals the significant impact of key risks and key risk interactions on network complexity. Based on this, a targeted risk mitigation strategy is proposed. The research results provide a more systematic quantitative analysis tool for coal mine safety management to support more scientific risk management and control decisions and reduce the probability of accidents.

The remainder of this paper is arranged as follows: Section 2 covers the materials and methods; Section 3 describes the results; Section 4 is the discussion; and Section 5 presents the conclusion.

2. Materials and Methods

Figure 3 outlines the four steps of this study: data collection and coding, network modeling, network analysis, and proposing risk mitigation strategies. During data preparation and encoding, 102 accident reports were examined through the application of root–state risk identification (RSRI) theory. The extracted risk factors are designated as nodes in the network, and the directed edges represent the determined causal relationship between these factors. In the follow-up stage of the study, a network model of coal mine gas explosion accident causation was established. Through the analysis of the network, the key factors and critical paths were determined. Finally, the network changes were verified after eliminating key factors and paths, and then targeted risk mitigation strategies were proposed.

Figure 3. The analytical framework of this study.

2.1. Root–State Risk Identification (RSRI) Theory

The theory of root–state risk identification (RSRI) originated from the theory of Root–Hazard Identification (RHRI) [23]. Both have strong continuity in terms of theoretical basis and analytical thinking, and their core differences are mainly reflected in their identification objects: RHRI pays more attention to identifying “hazard sources” that may cause accidents, while RSRI further expands this concept into “risk factors”, extending the identification content from narrow hazard sources to broader systematic risk components. Although both of these models aim to reveal the cause of accidents in essence, RSRI emphasizes the interaction of internal and external factors in the conceptual framework, and divides the identification results into “root risks” and “state risks”, thus building a more structured and hierarchical risk identification system on the basis of the original RHRI. Through this extension, RSRI can make a more comprehensive and forward-looking analysis of potential risks in complex systems.

Root–state risk identification (RSRI) theory divides the risks identified from accident reports into root risk and state risk. The root cause risk is the root cause of the accident. These risks are intrinsic to underground coal mining and production processes, regardless of whether they are recognized. They can be categorized into four types: human, equipment, environmental, and managerial risks. Specifically, the fundamental risk caused by people refers to the jobs of coal mining enterprises, such as mine managers, gas miners, and blasting workers. The root risks caused by machines include all mechanical equipment, such as shearers, hydraulic supports, and ventilators. The environment includes the working environment and natural geological conditions, including methane, coal, and groundwater. Fundamental organizational hazards encompass elements such as organizational structure, safety culture, and operational regulations [24,25]. State risk refers to unsafe conditions or behaviors caused by potential hazards that are the direct cause of accidents [26]. For example, human state risk mainly refers to unsafe behaviors, such as the improper installation of ventilators by ventilation workers or non-compliance with gas extraction processes by gas discharge workers. Equipment state risk mainly refers to the unsafe state of equipment, such as machine leakage, mine lamp explosion, or unqualified detonators. Environmental state risk mainly refers to the unsafe state of coal spontaneous combustion, insufficient air volume in the working face, and high gas concentrations in the heading face. Institutional state risks mainly refer to problems within institutions, such as an unreasonable organizational structure or insufficient safety supervision. As a new risk identification method, RSRI has not been widely used in prior research, but it can be used to effectively analyze the objective information of accident reports and classify them reasonably. The researchers in [27] used the root–state identification method to screen and evaluate the risk factors of cold chain logistics, which expanded the scope of the RSRI application.

When the RSRI theory is applied to coal mine gas explosion analysis, a tree structure diagram is constructed from the root risks to illustrate human, equipment, environmental, and managerial factors. Then, each root risk is subdivided to find the corresponding state risk according to the root risk, and the risk system is finally determined. The identification process is shown in Figure 4.

Figure 4. RSRI-based identification of root risk and state risks.

2.2. Complex Network Construction

Complex networks are widely used to describe diverse systems in nature and society [28]. They reveal the evolutionary processes of interactions among risk factors, explain the collective behavior of complex systems, and clarify the influence of individual elements on the system. Consequently, complex network analysis has been increasingly applied to accident prevention and risk assessment. It is represented as a graph $G(V,E,W)$, where $V=\left\{v_1,v_2\cdots ,v_n\right\}$ denotes the set of nodes in the network, $E=\left\{e_{11},e_{12},\cdots ,e_{1n}\right\}$ denotes the set of edges in the network [29], and $W$ represents the weight of the edge.

According to the descriptions provided in accident investigation reports, the causes of accidents are not independent of each other; rather, there is a causal relationship [30]. A complex network is a topological description of complex systems or objects and can be used as a modeling and analysis method for identifying the correlations between various factors in the system [31]. Complex networks include directed networks and undirected networks. In a directed network, the direction of an edge indicates causality between factors, whereas its weight represents the intensity of this relationship [32]. Different accidents may involve similar causal factors, leading to duplicated edges in the network. The frequency of such repetitions defines the corresponding edge weight. The construction of the directed network is illustrated in Figure 5.

Figure 5. Directed network construction from accident chains.

An adjacency matrix is established according to the content in Figure 5, as shown in Table 1.

Table 1. The adjacency matrix of the example.

	V1	V2	V3	V4	A1
V1	0	0	1	1	0
V2	0	0	1	0	0
V3	0	0	0	0	2
V4	0	0	0	0	1
A1	0	0	0	0	0

Accordingly, the $N\times N$ adjacency matrix $A$ forms the basis for constructing a directed weighted network. $N$ stands for the quantity of nodes involved in the network, and $E$ defines the edge set that connects them:

$$A=\left\{\begin{array}{l}{w}_{ij},i\to j\in E\\ 0,i\to j\notin E\end{array}\right.$$

(1)

$w_{ij}$ is the weight of the edge, that is, the frequency of repeated occurrence of causality.

2.3. Causal Network Topology Analysis

In complex network analysis, the characterization of topological features usually depends on a series of statistical indicators, such as degree centrality and betweenness centrality [33]. By applying these indicators, the structural characteristics of the network can be revealed from different dimensions, so as to identify the key risk nodes. Therefore, in this study, four representative network indicators were selected—degree centrality, betweenness centrality, clustering coefficient, and PageRank value—to determine key risk factors. These indicators can comprehensively reflect the local and global characteristics of the network and characterize the connectivity of nodes, the control of information flow, the degree of cohesion of neighborhoods, and the influence of nodes in the overall network.

(1)

Degree centrality

The centrality of a node is a significant indicator that quantifies its direct connectivity with other nodes in the network [34]. In the context of coal mine safety, a node with a higher degree centrality represents a risk factor that directly interacts with multiple other hazards, indicating that it should be prioritized for preventive inspection and control. In directed networks, degree centrality is defined as the combined value of in-degree and out-degree, as presented in Equation (2):

$$C_D(i)={\displaystyle \frac{a_{ij}+a_{ji}}{2(N-1)}}$$

(2)

$a_{ij}$ and $a_{ji}$ are the node input degree and the node output degree, respectively. $N$ is the total number of nodes in the network.

(2)

Betweenness centrality

Betweenness centrality is defined as the frequency with which a node appears on geodesic paths between pairs of nodes within the network [34]. This index can identify important nodes that play a role as a bridge in the network, even if their degree centrality is low. In coal mine accident evolution, a node with high betweenness centrality serves as a bridge or key transmission point in the risk propagation chain and should therefore be treated as a critical node for safety monitoring and emergency intervention, as expressed in Equation (3):

$$C_B(i)={\displaystyle \sum _{\begin{array}{l}s,t\in V\\ s\ne t\ne v\end{array}}\frac{{\sigma }_{\mathrm{st}}(i)}{{\sigma }_{st}}}$$

(3)

$\sigma (i,j)$ is the number of lines passing through node $i$ in the shortest path from node $s$ to node $t$. ${\sigma }_{st}$ is the shortest path number from node $s$ to node $t$.

(3)

Clustering coefficient

The clustering coefficient reflects the local density of connections among a node and its neighbors [35]. As defined in Equation (4), the greater the value of the clustering coefficient, the closer it is to the surrounding nodes. In the context of coal mine safety, a high clustering coefficient indicates locally concentrated risk factors, suggesting that these regions require enhanced inspection and coordinated management to prevent chain accidents.

$$C(i)={\displaystyle \frac{2e_i}{C_D(i)(C_D(i)-1)}}$$

(4)

$e_i$ is the number of edges actually existing between the neighbors of node $i$; $C_D(i)$ represents the out-degree of node $i$.

(4)

PageRank

Different from degree centrality and betweenness centrality, this method determines the criticality of nodes according to the input/output relationship of nodes [36]. In coal mine safety management, a node with a higher PageRank value indicates a risk factor with strong global influence in the accident evolution process, which should be regarded as a strategic target for system-level risk control and resource allocation.

$$PR(i)={\displaystyle \frac{(1-\alpha )}{n}}+\alpha {\displaystyle \sum _{j\in N_i^{in}}\frac{a_{ji}PR(j)}{C_D^{out}(j)}}$$

(5)

$\alpha$ is the attenuation factor, taking 0.85 [37]; $C_D^{out}(j)$ is the out-degree of node $j$; $N_i^{in}$ is the set of all incoming neighbors pointing to node $i$; and $a_{ji}$ represents the degree of node $j$.

2.4. Node Importance Evaluation Method

Since network indicators highlight different aspects of node significance, a comprehensive assessment is required. The existing literature often uses heuristic methods to sort the multiple centrality values of nodes and form multiple lists. A node is considered important if it appears in the top 5% of at least two lists [38]. However, the heuristic method has certain limitations, such as ambiguity in distinguishing the relative importance of nodes. To address this, the EW-TOPSIS approach is employed in this study to integrate degree centrality, PageRank, betweenness centrality, and clustering coefficient, thereby enhancing the accuracy of node identification and enabling the generation of a comprehensive ranking of node importance.

Key node identification in complex networks can be achieved through EW-TOPSIS, which merges entropy weighting and the TOPSIS decision-making framework. To implement this method for the evaluation matrix with $m$ objects and $n$ indicators, this process consists of the following main steps.

2.4.1. Entropy Weight Method

As a quantitative approach derived from information entropy principles, the entropy weight method enables integrated evaluation across multiple criteria. It mainly reflects the uncertainty of information by calculating the entropy value of the index so as to determine the weight of each index [39]. The calculation process can be divided into the following steps:

(1)

Define the initial decision matrix $X$, in which each element $x_{ij}$ represents the score of object $i$ under indicator $j$. The matrix has $m$ objects and $n$ indicators.

$$X=\left(\begin{array}{cccc}{x}_{11}& x_{12}& \cdots & x_{1n}\\ x_{21}& x_{22}& \cdots & x_{2n}\\ ⋮& ⋮& \ddots & ⋮\\ x_{m1}& x_{m2}& \cdots & x_{mn}\end{array}\right)$$

(6)

(2)

Standardized decision matrix.

$$X_{ij}^{\prime }={\displaystyle \frac{x_{ij}-\min (x_j)}{\max (x_j)-\min (x_j)}}$$

(7)

(3)

Calculate the value of the $ith$ object under the $jth$ index as the proportion of the index $P_{ij}$

$$p_{ij=}{\displaystyle \frac{X_{ij}^{\prime }}{{\displaystyle \sum _{i=1}^m{X}_{ij}^{\prime }}}}$$

(8)

(4)

Calculate the $jth$ index entropy $e_j$. When $p_{ij}=0$, $p_{ij}\ln (p_{ij})=0$, $e_j\in \left[0,1\right)$

$$e_j=-{\displaystyle \frac{1}{\ln (m)}}{\displaystyle \sum _{i=1}^m{p}_{ij}}\ln (p_{ij})$$

(9)

(5)

Calculate the weight of each indicator $w_j$, $w_j\in \left[0,1\right)$

$$w_j={\displaystyle \frac{1-e_j}{{\displaystyle \sum _{j=1}^{n}1-e_j}}}$$

(10)

where $1-e_j$ represents the information utility value of the indicator; that is, the higher the certainty of information, the greater the weight.

2.4.2. TOPSIS

TOPSIS, a classical method in multi-criteria decision-making, ranks alternatives according to how close they are to the ideal solution and how far they are from the negative ideal solution [40]. The specific steps are as follows.

(1)

Calculate the weighted standardization matrix. That is, each column of the standardized matrix $X_{ij}^{\prime }$ is multiplied by the corresponding weight.

$$T={(t_{ij})}_{mn}=\left(\begin{array}{cccc}{t}_{11}& t_{12}& \cdots & t_{1n}\\ t_{21}& t_{22}& \cdots & t_{2n}\\ ⋮& ⋮& \ddots & ⋮\\ t_{m1}& t_{m2}& \cdots & t_{mn}\end{array}\right)$$

(11)

$t_{ij}$ represents the weighted score of objects $i$ on indicator $j$.

(2)

Calculate the positive ideal solution $A_j^{+}$ and the negative ideal solution $A_j^{-}$. The positive ideal solution is defined as the maximum value of each column in the weighted normalized matrix, whereas the negative ideal solution is defined as the corresponding minimum value.

$$\left\{\begin{array}{l}{A}_j^{+}=\max (t_{1j},t_{2j},\dots t_{mj})\\ A_j^{-}=\min (t_{1j},t_{2j},\dots t_{mj})\end{array}\right.$$

(12)

(3)

The Euclidean distance is used to calculate the distance between the object and the positive ideal solution and the negative ideal solution.

$$\left\{\begin{array}{l}{D}_i^{+}=\sqrt{{\displaystyle \sum _{j=1}^n{(t_{ij}-A_j^{+})}^2}}\\ D_i^{-}=\sqrt{{\displaystyle \sum _{j=1}^n{(t_{ij}-A_j^{-})}^2}}\end{array}\right.$$

(13)

(3)

Calculate the relative closeness $C_i$ of each decision object.

$$C_i={\displaystyle \frac{D_i^{-}}{D_i^{+}+D_i^{-}}}$$

(14)

2.5. Analysis of Key Evolution Path Based on Edge Vulnerability

From the cause network of coal mine gas explosion accidents, it can be seen that the occurrence of accidents is usually the interaction of multiple risk factors, which evolve and accumulate through different paths, eventually leading to accidents. Therefore, from the perspective of the edge of the causal network, in this study, we comprehensively apply the three indicators of edge betweenness, average path length, and connectivity to determine the vulnerability of the edge, and then identify the key evolution path.

(1)

Edge betweenness

Edge $t$ is characterized by the count of shortest paths traversing it, which reflects the extent of its influence in the coal mine gas explosion accident cause network [41]. If a certain edge $t$ is removed, the larger the value of $B_t$ is, the more significant the contribution of edge $t$ to risk evolution is, and the easier it is to cause accidents.

$$B_t={\displaystyle \sum _{j\ne k\in V}{\sigma }_{jk}^t}$$

(15)

(2)

Average path length

This is used to represent the evolution speed of coal mine gas explosion accidents in the network [42]. The influence of each side on the accident evolution is determined by comparing the change in $D$ after removing some edges in the network. If a certain edge $t$ is removed, the larger the value of $D_t$ is, the more significant the effect of edge $t$ on risk evolution is, and the easier it is to cause accidents.

$$L_{avg}={\displaystyle \frac{1}{N(N-1)}}{\displaystyle \sum _{i\ne j,d(i,j)<\infty }d(i,j)}$$

(16)

$d(i,j)$ denotes the number of edges in the shortest path from node $i$ to node $j$;$d(i,j)<\infty$ means that only those node pairs with the shortest path are considered—that is, node $i$ to node $j$ are reachable.

(3)

Connectedness

Connectedness denotes the ratio of nodes connected to node $i$ relative to the total number of nodes in the network, reflecting the stability of the coal mine gas explosion accident evolution network [43]. The larger the value of $H_t$, the more difficult it is for the accident evolution network to be destroyed. When edge $t$ is removed, a smaller $H_t$ value indicates a greater impact of this edge on risk evolution and a higher likelihood of accidents.

$$H_t={\displaystyle \frac{n_i}{N}}$$

(17)

(4)

Vulnerability of the edge

Edge vulnerability is used to indicate the importance of edge $t$ in the network and the degree of influence on the network. Connectivity $H_t$, average path length $D_t$ and edge betweenness $B_t$ are selected to comprehensively evaluate the vulnerability of edge $t$ in the network.

$$V_t={\displaystyle \frac{B_t{D}_t}{H_t}}$$

(18)

3. Results

3.1. Dataset

The purpose of studying the mechanism of accident occurrence is to explain the mechanism of accident occurrence and provide a reference for risk assessment and accident analysis. Most existing research uses the accident causal model for analysis. The accident causal theory provides theoretical support for this model. It is a fundamental theory in safety science and offers a framework for accident analysis and prevention [44]. Therefore, the use of accident causation theory in accident analysis can lead to more scientific results and preventive measures being obtained. By systematically summarizing the experience and lessons of previous accidents, we can enhance the scientificness and pertinence of preventive measures. This, in turn, can further promote the formulation of relevant laws and regulations. It can also contribute to the construction of a risk control mechanism and the dissemination of safety knowledge [45,46,47]. In view of the limited causes of accidents presented by a single accident, it is necessary to rely on the systematic analysis of a large number of accident cases to form comprehensive and systematic information to support decision-making. When the accident sample accumulates to a certain scale, the common cause law can be extracted, which provides a basis for the formulation of effective prevention strategies and the promotion at the industry level [48]. The core of accident cause analysis in coal mines is to identify the root causes of accidents and eliminate potential risks. This provides theoretical support and practical guidance for accident prevention, safety management improvement, and the future optimization of the working environment.

A total of 102 coal mine gas accident investigation reports published by the official website of the National Mine Safety Supervision Bureau (NMSA) were collected for this study. These reports were publicly released from 2000 to 2024, and the accident levels were major accidents and catastrophic accidents. Only reports that clearly identified the accident type as a gas explosion, contained a complete description of the cause, and were officially approved and archived by the NMSA were included in the dataset. This selection process ensures the comparability and reliability of the data used in this study. See Table 2 for accident classification.

Table 2. Accident grade classification.

Grade of Accidents	Death Toll	Injuries	Direct Economic Loss (Yuan)
General	<3	<10	>106, <107
Significant	≥3, <10	≥10, <50	≥107, <5 × 107
Major	≥10, <30	≥50, <100	≥5 × 107, <108
Catastrophic	≥30	≥100	≥108

3.2. Risk Factor Identification

In this study, risk is defined as a source or state that may cause injury or loss. Risk can be tangible or intangible. In this study, the root–state risk identification method was used to draw the risk structure diagram of coal mine gas explosion accidents layer by layer from the 102 accident reports collected. The fundamental risks and state risks from people, equipment, environment, and management were organized, and a total of 112 factors (except for accident nodes) were obtained, as shown in Figure 6.

Figure 6. RSRI-based framework for identifying gas-explosion risks.

Although all reports are official reports published by the State Administration of Mine Safety, due to differences in the completeness and document standards of reports in different years, some deviations may be introduced in the process of factor extraction. To minimize this impact, only those reports containing complete accident cause reports were included in the dataset. The robustness and reliability of the extracted data were ensured by cross-checking the causality of multiple reports. In addition, we independently reviewed the extracted factors to ensure their completeness, internal consistency, and representativeness in describing the causes of gas explosion accidents. After examination, the identified factor set reached theoretical saturation. The risk factors were divided into four categories: unsafe human behavior, unsafe state of equipment, unsafe environmental factors, and unsafe management measures. The specific risk factors are shown in Table 3, Table 4, Table 5 and Table 6.

Table 3. Unsafe human behavior.

Node Name	Risk Factor	Node Name	Risk Factor
PE01	The resumption of production without permission	PE28	Illegal blasting operation
PE02	Deliberately concealing the production location	PE29	Illegal use of coal slime instead of water cannon mud to block the blasthole
PE03	Illegal organization of production	PE30	No water stemming used to seal the blasthole
PE04	Single-hole mining	PE31	Undetected gas concentration
PE05	Safety monitoring data falsification	PE32	Missing inspection by tile inspectors on empty shifts
PE06	Illegal construction of permanent closures	PE33	Inspector’s windpipe not timely
PE07	Insufficient management of enclosed fire zones	PE34	Illegal discharge of gas
PE08	Management of pyrotechnics not standardized	PE35	No gas drainage measures
PE09	No governance in place during the fault period	PE36	Excavation instead of mining
PE10	Engineering quality supervision and acceptance not strict	PE37	Illegal mining security coal pillar
PE11	Illegal unsealing of closed areas	PE38	No longer sealed according to regulations
PE12	Illegal mining of the super-layer cross-border	PE39	Roadway coal mining
PE13	Multi-faceted organization production	PE40	High fall risk in coal mining
PE14	Mechanical and electrical management not in place	PE41	Outburst prevention measures not in place
PE15	Failure to conduct ventilation inspections as required	PE42	Illegal top coal caving
PE16	Implementation of air leakage prevention measures not in place	PE43	Illegal repair of mine lights
PE17	No effective gas prevention measures	PE44	Illegal smoking ignition
PE18	Improper installation of the local fan	PE45	Illegally carrying non-explosive-proof items into the well
PE19	Fire door not installed	PE46	Workers’ lack of safety awareness
PE20	No methane sensor installed	PE47	Open/Stop fan at will
PE21	Fire extinguishing measures	PE48	Main fan disabled for a long time
PE22	Systems to lead teams down the well not implemented	PE49	Lack of safety technical knowledge among workers
PE23	Reported material intentionally falsified	PE50	No need for safety precautions
PE24	‘One shot, three inspections’ and ‘three-person chain’ systems not implemented	PE51	Illegal operation of electrical equipment
PE25	Serious violation of technical management	PE52	Illegal use of woven bags as local fan ducts
PE26	Illegal live working by electricians	PE53	Illegally sending electricity to send wind
PE27	Illegal gas welding operation

Table 4. Unsafe state of equipment.

Node Name	Risk Factor	Node Name	Risk Factor
EQ01	Roof caving	EQ15	Local fan multi-sided air supply
EQ02	Roof collapse	EQ16	Imperfect ventilation system
EQ03	Friction sparks	EQ17	Deformed fan blade in operation
EQ04	Electric sparks	EQ18	Wind bridge not erected
EQ05	Blasting sparks	EQ19	Local fan draws circulating air
EQ06	Naked fire	EQ20	Local fan stops the wind
EQ07	Underground power leakage	EQ21	Air leakage in the gas extraction borehole
EQ08	Electric cable short circuit	EQ22	Sealing quality is unqualified
EQ09	Explosion of electrical equipment	EQ23	Closed wall air leakage
EQ10	Metal equipment impact	EQ24	Tile inspection instrument is inaccurate
EQ11	Miner’s lamp lost explosion	EQ25	Tile detector fails to automatically power off the alarm
EQ12	Light junction of the signal line caught fire	EQ26	No comprehensive dustproof system
EQ13	Lifting winch signal device explosion	EQ27	Power supply network failure
EQ14	Coal mining machine and rock friction

Table 5. Unsafe factors in the environment.

Node Name	Risk Factor	Node Name	Risk Factor
EN01	Geological conditions and poor structure	EN14	Large areas of empty roof in the goaf
EN02	Underground impact pressure	EN15	Air leakage in the gob
EN03	Coal spontaneous combustion	EN16	Blind roadway gas accumulation
EN04	Gas gushing from fractures in the fault zone	EN17	Gas accumulation in the heading face
EN05	Coal dust involved in the explosion	EN18	Gas accumulation in the roadway
EN06	Working face power failure	EN19	Coal and gas outburst in the heading face
EN07	Full negative pressure ventilation system has not been formed	EN20	Ground instability
EN08	Insufficient air volume in the working face	EN21	Gas accumulation in the upper corner
EN09	Underground airflow short circuit	EN22	Abnormal gas emission
EN10	Working face series circulating wind	EN23	Gas burning
EN11	Connection of the working face is closed	EN24	Coal seam exposed by the return air shaft
EN12	Gas accumulation in the goaf	EN25	Gas accumulation in the return airway
EN13	Air path in the goaf is blocked

Table 6. Unsafe management measures.

Node Name	Risk Factor	Node Name	Risk Factor
MA01	Employment management is not standardized	MA05	Lack of professional and technical personnel and special operations staffing
MA02	Insufficient safety investment	MA06	Insufficient safety supervision
MA03	Special types of workers unlicensed to work	MA07	Safety supervision and inspection efforts are poor
MA04	Safety education training not in place

3.3. Coal Mine Gas Explosion Accident Causation Network Model

3.3.1. Network Model Construction

Each of the 112 risk factors identified using the RSRI theory was regarded as an independent element involved in the causation mechanism of gas explosion accidents. Accordingly, each factor was modeled as a node to represent its potential role as a cause or consequence in the accident evolution process. Directed edges were established based on causal relationships extracted from historical accident reports, where the edge direction indicates the “cause–effect” relationship between two factors. The arrow points to the result, and the tail indicates the reason. The edge weight corresponds to the frequency of repeated occurrence of each causal link across multiple accident cases [32,49]. Factors belonging to the same category were assigned distinct colors (People: purple; Equipment: green; Environment: blue; Management: orange), while the accident node (AC01) was highlighted in dark green. The node size reflects its degree centrality, indicating the number of direct causal connections of each factor.

As shown in Figure 7, a directed weighted network with 113 nodes (including an accident node) and 341 edges was constructed by using Gephi (version 10.1) The network presents a dense core structure centered on nodes EN17, EN08, and MA06, showing the highest degree of centrality, indicating that gas accumulation in the heading face, insufficient air volume in the working face, and inadequate safety management play a key role in accident evolution.

Figure 7. Causal network of risk factors for coal-mine gas explosions.

3.3.2. Node Network Model

The small-world network model was originally proposed in [35]. The small-world characteristics of complex networks have a short average path length (no more than 10) and a high clustering coefficient (no less than 0.1) [50]. In order to further analyze the network properties, according to the definition of small world [51], we used Python’s NetworkX (version 3.4) library to create a random network with the same vertices and edges. As shown in Table 7, through comparison, it can be seen that the causal network has an average path length similar to the random network and a relatively high clustering coefficient, showing obvious small-world network characteristics.

Table 7. Comparison of the causation network and the random network.

Network Model	Average Path Length	Clustering Coefficient
The cause of the network	2.64	0.13
Random network 1	2.94	0.04
Random network 2	2.82	0.06

The average degree of the causal network calculated by Gephi is 3.018, indicating that each node is related to 3–4 risk factors. The results of out-degree, in-degree, and total degree of nodes with degrees greater than 3 are shown in Figure 8a. The results show that, except for accident node AC01, the degrees of PE03 (illegal organization of production), EN08 (insufficient air volume in the working face), EN17 (gas accumulation in the heading face), and MA06 (insufficient safety supervision) are relatively high, indicating that they have a strong direct influence on the network. The intermediary centrality (Figure 8c) and PageRank value (Figure 8d) of the two risk factors, PE28 (illegal blasting operation) and EN17 (gas accumulation in the heading face), are higher in the network, indicating that they have a strong influence on the process of risk evolution and play a key role in the process of accident risk evolution. Most of the accident evolution paths pass through these nodes. Therefore, cutting off the connection between these nodes and other nodes can effectively suppress the spread of accident risk. Figure 8b shows that the clustering coefficient of PE07 (insufficient management of enclosed fire zones) is the highest, reaching 1. This indicates that if closed fire area management is not in place, its impact will quickly spread to the adjacent nodes, forming a locally concentrated risk outbreak. Due to poor ventilation, the closed area is prone to gas accumulation. If an open fire occurs in a closed area, it will quickly detonate the gas in this area, resulting in the rapid onset of an accident. In addition, some nodes are at the forefront of PageRank ranking, especially gas accumulation in the heading face (EN17), which shows that the information transmission efficiency in the network is more prominent. Other nodes give greater weight to the information transmission of EN17, and EN17 has a higher frequency of occurrence in the accident evolution pathway. In short, controlling these key risk factors plays a vital role in blocking the evolution of risk and reducing the frequency of accidents. The betweenness centrality (Figure 8c) and PageRank value (Figure 8d) of two risk factors, PE28 (illegal blasting operation) and EN17 (gas accumulation in the heading face), are high in the network. It can be inferred that these factors exert considerable influence over risk communication and accident formation, thereby warranting heightened attention and proactive control measures.

Figure 8. Statistical distributions of network indicators for the coal mine gas explosion accident causation network: (a) degree centrality; (b) clustering coefficient; (c) betweenness centrality; (d) PageRank value.

Considering the positive correlations between degree centrality, betweenness centrality, clustering coefficient, PageRank value, and node importance, the EW-TOPSIS method was used to calculate the node importance. Table 8 gives the entropy value and weight of the four indicators. Information entropy reflects the amount of information or uncertainty of an index. A higher information entropy value indicates greater dispersion in the index data, leading to increased system uncertainty. As a result, the impact on decision-making decreases, and the weight assigned to the index becomes smaller [39]. The weight of mediating centrality is the highest, and the degree is the lowest, indicating that the degree distribution of factors is more dispersed. Among the four indicators, the intermediary centrality weight accounts for the highest proportion and has the greatest impact on decision-making.

Table 8. Information entropy and the weight of each index obtained using the entropy weight method.

Index	Information Entropy	Entropy Weight Method Weight
Betweenness centrality	0.748	0.385
PageRank	0.839	0.247
Clustering coefficient	0.862	0.212
Degree centrality	0.898	0.156

Referring to the research in [52], Table 9 presents the top 20 factors calculated by EW-TOPSIS after excluding accident node AC01. The comprehensive importance analysis shows that risk nodes such as EN17 (gas accumulation in the heading face), MA06 (insufficient safety supervision), and PE07 (insufficient management of enclosed fire zones) have a greater impact on the entire network. Among them, the closeness degree of “gas accumulation in the heading face” reaches 0.707, which is the most likely factor to induce accidents. According to statistics from the Ministry of Emergency Management of the People’s Republic of China, most accidents in coal mines are located at the heading face. Gas accumulation in the heading face increases the pressure in the mine, which can trigger a gas outburst and lead to a gas explosion [53]. This aligns with our research conclusion. The lack of safety supervision (MA06) is the most influential indirect factor, and the sustainable development of coal mine safety is significantly correlated with government safety supervision [54].

Table 9. Proximity of causative network nodes.

Node Name	Posting Progress	Node Name	Posting Progress
EN17	0.707	PE52	0.177
MA06	0.406	PE53	0.177
PE07	0.306	PE19	0.177
EN08	0.305	PE48	0.177
PE03	0.271	EN18	0.164
EN12	0.232	PE34	0.157
PE05	0.227	PE15	0.149
EQ20	0.223	EQ04	0.145
EQ06	0.215	EN22	0.143
PE28	0.197	EN10	0.142

Table 9 shows that, among the key risk factors, unsafe human behavior accounts for half of all accidents. The researchers in [13] also point out that human-induced accidents account for 71.6% of all accidents. The root cause lies in the fact that, compared to the controllability of mechanical equipment risk, the unsafe behavior of operators—whether intentional or unintentional—is often associated with significant concealment and randomness. These characteristics make it difficult to fully eliminate such behaviors through conventional technical methods. Reducing unsafe human behaviors requires the development of more effective strategies. Moreover, illegal organization of production (PE03), tampering with safety monitoring data (PE05), insufficient management of enclosed fire zones (PE07), and illegal blasting operations (PE28) are among the top ten key causes of accidents. It is worth noting that the inadequate management of closed fire zones (PE07) has the highest weight in the category of “unsafe human behavior” and should be the focus of prevention and control. In addition, although only insufficient safety supervision (MA06) belongs to the category of “unsafe management measures”, its risk amplification effect through an accident chain reaction is particularly significant. The authors of [55] confirmed that the failure of the regulatory system is a common cause of major accidents in coal mines. Specifically, the efficiency of coal mine safety management shows a significant multi-level attenuation characteristic: while the mine chief is the person legally responsible for safety, the safety responsibility of the middle- and grass-roots managers is generally blurred. This systematic regulatory failure directly leads to the long-term existence of high-risk behaviors such as the illegal organization of production (PE03) and safety data fraud (PE05). The safety risks arising from poor organizational behavior have a significant multiplier effect, thereby increasing the risk of coal mine accidents.

3.3.3. Identification of Key Evolution Paths

The approach used in this study differs from previous literature that only used edge betweenness [52], edge weight [56], key factors [57], and other methods to determine the key risk evolution path. This study comprehensively used edge betweenness, average path length, and connectivity to analyze edge vulnerability. This approach reveals the intermediary role of edges in shortest path transmission. It also reflects the impact of deleting edges on the overall information transmission efficiency and connectivity structure of the network. The analysis aims to improve the accuracy and robustness of key edge identification. Additionally, it provides a more comprehensive evaluation of the contribution of edges to the overall stability and function maintenance of the network. The top 20 ranking results of edge vulnerability calculated by Python are shown in Table 10.

Table 10. Top 20 edge ranks for vulnerability.

Initial Nodes	Goal Node	Edge Betweenness ${\mathit{B}}_{\mathit{t}}$	Average Path Length ${\mathit{D}}_{\mathit{t}}$	Connectedness ${\mathit{H}}_{\mathit{t}}$	Vulnerability ${\mathit{V}}_{\mathit{t}}$
EN17	MA06	2195.833	2.899	0.619	10,275.224
EN08	EN17	1556.000	2.694	0.619	6767.281
PE15	EN08	1376.000	2.663	0.619	5914.690
EN22	PE15	980.583	2.668	0.619	4222.286
MA06	PE03	769.500	2.678	0.619	3326.054
MA04	PE46	724.000	2.755	0.619	3219.258
PE28	EN22	723.500	2.740	0.619	3199.984
EQ20	EN17	664.000	2.654	0.619	2844.107
EN06	EQ20	551.000	2.598	0.575	2488.505
PE15	EQ16	376.000	2.676	0.619	1624.174
PE11	PE03	376.500	2.658	0.619	1615.091
MA06	MA04	370.000	2.660	0.619	1588.940
EN12	PE15	365.000	2.671	0.619	1573.654
EQ15	EN08	356.500	2.654	0.619	1527.167
MA06	PE14	351.000	2.671	0.619	1513.125
PE46	PE28	321.000	2.653	0.619	1374.782
EN17	PE17	300.333	2.666	0.619	1292.282
MA06	PE49	288.500	2.642	0.619	1230.142
PE36	PE15	283.083	2.648	0.619	1209.787
EN14	EQ20	282.000	2.646	0.619	1204.293

It is found that the connectivity of key nodes and key edges is excellent, and the network transitivity is strong. The relationship between key nodes is studied, and the evolution path of key risks is obtained. The paths with causal relationships were connected from high to low according to the edge vulnerability value, and the risk evolution path was constructed. Figure 9 shows three representative risk evolution paths. The arrows denote the direction of causal propagation along the evolution path. The node colors correspond to four categories of risk factors—unsafe behavior, unsafe equipment states, unsafe environmental conditions, and unsafe management measures. Among them, the highest vulnerability value of the chain is “excavation instead of mining → failure to conduct ventilation inspections as required → insufficient air volume in the working face → gas accumulation in the heading face → insufficient safety supervision → inadequate safety education and training → insufficient safety awareness of workers → illegal blasting operation → open fire → accident”, which is the most critical path for risk evolution. Ventilation conditions are one of the key factors affecting the severity of gas explosions [58]. If mine workers violate the method of excavation and the ventilation manager fails to perform ventilation inspection and maintenance procedures according to the regulations (PE15), this can result in insufficient air volume in the working face (EN08). When mining work is carried out, gas accumulates in the heading face (EN17). However, due to the failure of the higher authorities to effectively implement safety management (MA06), safety education and training are not in place, and the safety knowledge and awareness levels of blasting workers are relatively low. Therefore, workers are prone to serious habitual violations and unsafe practices. If the gas concentration reaches the critical value, a gas explosion will occur [59]. As an example case, workers did not realize that gas accumulation had occurred and were unaware of the importance of early safety preparation work. During preparation, they did not use coal slime to block the blasthole according to the specifications, and violated the “one shot, three inspections” system. The explosion operation (PE28) was carried out in violation of the rules, resulting in an open fire (EQ06), and the accumulated gas exploded. However, an accident is not necessarily due to the initial stage becoming out of control. If there is accumulated gas or a potential ignition source in the working face, an accident may occur rapidly once any stage is out of control. Therefore, every node manager on the critical pathway should be highly valued.

Figure 9. Key paths of risk evolution in the causal network model.

3.4. Network Simulation with Critical Nodes and Links Removed

The key risks and the role of edges in the network, shown in Table 9 and Table 10, were determined using EW-TOPSIS and edge vulnerability, but their contribution to network complexity is unknown. The researchers in [60] quantified the impact of removing critical nodes and links on risk networks by recalculating density, cohesion, and reachability matrices, demonstrating their strong effect on network complexity. Density and cohesion describe connectivity and complexity at the global level, whereas the reachability matrix reveals node accessibility at the local level. Following [60], simulation was used in this study to analyze the network effects of key risks and the critical evolution path.

Through recalculation, the results show that the new network is simplified to 93 nodes and 139 edges. Because Gephi automatically ignores isolated nodes (the remaining 86 nodes are shown in Figure 10 and compared with Figure 7), the complexity of the network is greatly reduced, and the graph density is reduced from 0.027 to 0.019, a decrease of 29.63%. The accessibility results for the risks are shown in Table 11. Prior to eliminating key risks and interrupting critical evolution paths, 7801 risk-reachable matrices were identified, corresponding to 61.64% of the theoretical maximum (113 × 112 = 12,656). After eliminating key risks, the number reduced to 1979, or 23.13% of the adjusted maximum (93 × 92 = 8556), which is considerably lower than the initial proportion. For the 93 non-key factors, the reachable matrices totaled 4994, which was obtained by subtracting the contributions of key nodes and links from the overall network calculation (6136−75−75−5−75−1−75−75−75−1−5−5−75−75−75−75−75−75−75−75−75 = 4994). The results show that disconnecting the key node factors from other nodes in the network or isolating key nodes hinders the evolution of accident risk to a certain extent [12].

Figure 10. Causation network of coal-mine gas explosions after removing key risks and paths.

Table 11. Statistics of accessible quantity before and after risk control.

	Reachable Quantity Before Risk Control				Reachable Quantity Before Risk Control
Risk Factor	Other 93 Non-Critical Risk Factors	All Risk Factors	Achievable Quantity After Risk Control	Risk Factor	Other 93 Non-Critical Risk Factors	All Risk Factors	Achievable Quantity After Risk Control
AC01	0	0	0	EQ04	5	8	—
PE01	75	95	0	EQ05	2	2	2
PE02	74	94	61	EQ06	5	7	—
PE03	75	94	—	EQ07	74	94	1
PE04	74	94	58	EQ08	74	94	4
PE05	75	94	—	EQ09	74	94	5
PE06	74	94	50	EQ10	6	9	6
PE07	5	7	—	EQ11	3	3	3
PE08	5	8	0	EQ12	5	8	0
PE09	76	96	1	EQ13	5	9	0
PE10	75	95	52	EQ14	6	9	6
PE11	74	94	50	EQ15	74	94	3
PE12	74	94	65	EQ16	74	94	50
PE13	75	95	62	EQ17	6	9	6
PE14	74	94	7	EQ18	74	94	0
PE15	75	94	—	EQ19	74	94	0
PE16	74	94	50	EQ20	75	94	—
PE17	74	94	3	EQ21	74	94	4
PE18	74	94	50	EQ22	74	94	51
PE19	1	1	—	EQ23	74	94	50
PE20	2	2	2	EQ24	2	2	2
PE21	4	7	2	EQ25	1	1	1
PE22	74	94	50	EQ26	2	2	2
PE23	74	94	52	EQ27	75	95	1
PE24	74	94	0	EN01	75	95	0
PE25	74	94	59	EN02	75	95	0
PE26	5	9	0	EN03	5	8	3
PE27	5	9	0	EN04	75	95	0
PE28	75	94	—	EN05	1	1	1
PE29	74	94	1	EN06	74	94	0
PE30	74	94	1	EN07	74	94	4
PE31	74	94	50	EN08	75	94	—
PE32	74	94	6	EN09	74	94	50
PE33	74	94	0	EN10	75	94	—
PE34	75	94	—	EN11	75	95	0
PE35	74	94	1	EN12	75	94	—
PE36	74	94	51	EN13	0	0	0
PE37	74	94	51	EN14	74	94	50
PE38	74	94	50	EN15	74	94	50
PE39	74	94	57	EN16	1	1	1
PE40	74	94	51	EN17	75	94	—
PE41	74	94	0	EN18	75	94	—
PE42	74	94	50	EN19	74	94	1
PE43	5	9	0	EN20	74	94	52
PE44	5	8	0	EN21	74	94	2
PE45	5	8	0	EN22	75	94	—
PE46	74	94	50	EN23	4	7	2
PE47	74	94	51	EN24	76	96	1
PE48	75	94	—	EN25	1	1	1
PE49	74	94	50	MA01	74	94	52
PE50	74	94	1	MA02	74	94	50
PE51	6	10	1	MA03	74	94	51
PE52	75	94	—	MA04	74	94	50
PE53	1	1	—	MA05	74	94	50
EQ01	75	95	6	MA06	75	94	—
EQ02	74	94	54	MA07	74	94	50
EQ03	5	8	5

The simulation results show that removing key nodes or interrupting key evolutionary paths will significantly weaken the structural connectivity of causal networks, indicating that the risk propagation ability of the system is substantially weakened. The decline in structural connectivity means that the number of effective transmission channels of potential risks in the system is significantly reduced, thus inhibiting the chain evolution process caused by multi-factor coupling. Based on this, the simulation results provide quantitative evidence, which shows that intervention for key nodes and key links can obviously enhance the robustness and safety resilience of coal mine accident systems. This discovery also provides an important empirical basis for subsequent safety management strategies.

4. Discussion

4.1. Comparative Analysis with Similar Studies

This study focuses on the risk analysis of coal mine gas explosion accidents, with the aim of proposing a new risk management framework. The comprehensive identification of risk factors is the premise of risk management, which provides the necessary basis for subsequent risk analysis and processing. On this basis, eliminating key risk factors has become the primary task of risk management because these factors are closely related to many potential risks and can easily lead to accidents. Previous studies found that the failure to detect gas or leakages in time [61], illegal live work [62], etc., are important risk factors for gas explosion accidents. However, this study found that due to the particularity of the condition of coal mine gas explosion accidents, the lack of air volume in the working face leads to an increase in gas concentration [63], resulting in gas accumulation in the heading face; this represents an ignition source in the environment, which is the most direct cause of accidents [64]. The lack of closed fire zone management (PE07) is also one of the most important causes of accidents. The reason for this is that, compared with electric sparks and explosion sparks, open fire is more likely to reach the explosion limit because its ignition induction period is shorter, and the flame may flow downwind to other areas, causing secondary explosions. Gas explosions often release a large number of toxic gases. This can lead to employee coma, increased difficulty in search and rescue, and more casualties. In addition, insufficient safety supervision (MA06) obviously increases the likelihood of accidents. Previous research [65] has also proven the importance of organizational management factors. Leaders must enhance safety awareness and increase the intensity and frequency of safety training. Unsafe behaviors exhibited by production personnel are strongly influenced by limited safety awareness, inadequate knowledge and skill sets, and a lack of systematic safety training [66]. Unsafe behavior greatly increases the frequency of accidents. Despite the rapid advancement of intelligent coal mining technologies, underground workers—particularly in small and medium-sized mines—often have limited educational backgrounds. Coupled with insufficient investment in safety training, this leads to a lack of awareness of potential dangers and the failure of workers to comply with safety regulations. For example, illegal blasting operations carried out by blasters (PE28) have been identified as a major contributor to gas explosion incidents [67]. Illegal blasting may produce blasting sparks that can easily cause open flames [68], a finding that is consistent with our research conclusions. However, if the “one shot, three inspections” and “three-person chain” system is implemented before blasting, the accumulation of ignition sources can be greatly reduced. Due to the generally low educational background of coal miners, it will take time to strictly implement this system. However, raising the educational requirements of workers may cause a shortage of labor for a period of time. The most effective means to control this risk is to carry out safety training regularly, conduct assessments as required, implement a reward and punishment system, and improve workers’ safety awareness.

It is worth noting that although previous studies rarely considered security monitoring data falsification (PE05) to be a direct risk factor, this behavior may pose a serious threat to coal mine safety. This is because coal mine regulatory agencies rely on the information and data provided by the head of the mine. However, since the head of the mine has more direct control over the actual operation of the mine, they may deliberately and fraudulently conceal or tamper with safety data to evade supervision [69], meaning that the regulatory agency does not have full details of the real situation of the mine. Information asymmetry may lead to the failure of regulatory agencies to identify potential security risks in a timely manner, thus missing opportunities for prevention and intervention. Such information asymmetry and data fraud often lead to coal mine accidents. The illegal organization of production (PE03) is another risk that is often overlooked. Li et al. [9] mentioned this risk, but it was not regarded as a key factor, and the study focused more on elimination. In fact, the illegal organization of production is a frequent contributor to accidents. It is concerning that 15 out of 25 major accidents in 2019 were associated with illegal mining activities, including over-level mining, the unauthorized resumption of work during shutdown, cross-border mining, and production subcontracting. Illegal production activities are often accompanied by risky behaviors such as cross-boundary mining to evade supervision, making it difficult for coal mine regulatory authorities to conduct routine safety inspections in cross-boundary areas. As a result, miners may remain unaware of existing but undetected hazards during mining operations, which increases the risk of accidents. The unauthorized resumption of work during shutdown is another key factor for such accidents. If a coal mine exceeds production requirements or its equipment is unsuitable, the safety supervision department imposes shutdown requirements, but the mine manager often disregards this instruction. Resuming production without authorization significantly elevates the risk of accidents. In order to curb the emergence of such violations, law enforcement should be strengthened, and the penalties should be increased.

4.2. Risk Amelioration Strategy

The key factors determined by the accident cause network analysis include unsafe human behavior (PE07, PE03, PE05, PE28, PE52, PE53, PE19, PE48, PE34, and PE15), unsafe equipment (EQ20, EQ06, and EQ04), unsafe environmental conditions (EN17, EN12, EN18, EN22, and EN10), and unsafe management measures (MA06). The evolution path with the highest vulnerability is “PE36 → PE15 → EN08 → EN17 → MA06 → MA04 → PE46 → PE28 → EQ06 → AC01”. Obviously, the key risk evolution path is the dynamic evolution process in which the key risk factors are excited by and coupled with each other, eventually leading to the occurrence of gas explosion accidents. Therefore, for accident prevention and process safety management, the control of key risk factors is an important part of preventing accidents. If key risk factors are not promptly controlled, the associated risks may rapidly propagate to adjacent nodes, forming a risk evolution pathway that can ultimately result in accidents. The effective mitigation of these critical risk factors requires the implementation of targeted preventive or corrective measures to address potential problems before they materialize, thereby reducing the likelihood of accidents [70]. Furthermore, risk control strategies should be tailored to the severity of individual risk factors to ensure the safety and stability of the coal mine system while achieving optimal resource allocation [71,72]. For high-risk factors, the most stringent preventive and control measures should be implemented. Moreover, given the specific conditions associated with coal mine gas explosion accidents, organizational safety management should prioritize the prevention of direct triggers, such as gas accumulation and open flames, which may initiate these accidents. In response to high-risk factors such as gas accumulation in the heading face (EN17), gas accumulation in the goaf (EN12), gas accumulation in the roadway (EN18), open fire (EQ06), and electrical sparks (EQ04), several measures can be implemented. These include the formulation and enforcement of strict safety regulations, strengthening the real-time monitoring of gas concentration and the management of alarm devices, and enhancing on-site inspections and safety checks to identify and immediately correct any issues. At the same time, the gas emission approval process is formulated to ensure that the emission records are traceable and that illegal emission behavior is immediately reported. Combining an effective gas drainage system with real-time gas concentration monitoring ensures that the gas concentration remains within a safe range. In case of abnormalities, emergency ventilation and emission measures are initiated promptly to avoid safety risks caused by gas accumulation. Obvious warning signs are set up in the operation area, and live working and illegal blasting operations are strictly prohibited to prevent accidents caused by electrical sparks and open flames. In addition, due attention should be paid to increasing workers’ safety awareness, reducing unsafe behavior, and providing comprehensive training programs focusing on hazard identification, risk assessment, and emergency response. Proactive safety training includes daily safety exercises and scenario-based testing. This improves staff safety awareness and the ability to respond to emergencies. These measures significantly reduce the likelihood of accidents. Due to the particularity of the coal mine industry, in addition to unified safety training for all employees, it is also crucial to establish a strict pre-operation inspection system for special operators. During electrical maintenance operations, electricians are required to cut off the power supply before the operation and implement the “two-level inspection” system to ensure that the power supply is completely disconnected. Blasting workers must strictly implement the system of “one shot, three inspections” before blasting operations. The strict implementation of rules and regulations plays an important role in reducing accidents. In addition to strengthening safety training, improving safety management levels is also an important measure to reduce accidents. The ongoing reinforcement of safety culture, coupled with active employee engagement in safety education, is essential for improving safety awareness and responsibility across the workforce. Additionally, increased investment in safety infrastructure and incentive-based participation in safety management provide additional support. Collectively, these efforts facilitate proactive accident prevention and contribute to the long-term safety and stability of coal mining operations.

In the process of formulating and implementing risk control measures, it is recommended that enterprises adopt the RSRI (people, equipment, environment, management) risk identification framework to comprehensively and dynamically identify and control the root risk and state risk in coal mines. With the proper allocation of personnel and resources, a comprehensive safety supervision framework can be established. Supporting procedures then ensure effective risk identification, emergency management, and related operations. For improving the overall effectiveness of coal mine safety management, the adoption of a three-tier defense-based risk management mechanism is proposed [73]. This will enhance the overall risk prevention and control ability, as shown in Figure 11. The first line of defense is the formulation of control standards, which are the boundaries of risk control. They specify the boundaries within which risks can be effectively managed, with provisions formulated according to relevant national, industry, and manufacturer standards. The formulation of control standards covers the technical requirements of equipment, personnel, and environment, especially the safety standards in coal mine operational environments, operation processes, and equipment maintenance. For example, the safety code of conduct for operators, design and operation standards for ventilation systems, and requirements for the regular maintenance and repair of equipment should be central to the control standards. These measures ensure that operations are consistently conducted within safety parameters. The second line of defense is control measures, which are clear provisions on the standards and procedures of coal mine safety. These measures aim to ensure that coal mine operations are carried out within the prescribed safety range. If the identified risks do not meet the control standards, they are considered to be uncontrolled, which can easily lead to accidents. A necessary condition for a gas explosion in a coal mine is a gas concentration between 5% and 16%. Therefore, one control measure is to keep the gas concentration below 1%. Another control measure is to ventilate a certain area for up to 30 min before mining to ensure that the gas concentration is within the control standard. In addition, strengthening the safety training of miners is important to ensure that they are proficient in safety operation procedures. Clarifying control measures helps minimize human errors and ensures their effective implementation. Rectification measures are the third line of defense. When identified state risks escalate beyond control, timely corrective actions must be taken to return the system to a stable and safe state. As an example, if the gas concentration rises above the prescribed safety limit, gas detectors are triggered to issue an alarm. Management personnel should immediately lock the over-standard area, open the local ventilator, increase ventilation in the well, and reduce the gas concentration to ensure the safety of the mine environment. In the process of coal mine management, the “three-layer defense line” management system is adopted, which can effectively prevent and control risks at multiple levels. By formulating clear control standards, implementing specific control measures, and taking effective rectification actions in the event of potential safety hazards, the system establishes a multi-level safety protection network. This significantly reduces the risk of major safety accidents, such as gas explosions, during coal mine production, thereby enhancing the overall level of coal mine safety management.

Figure 11. Three lines of defense risk prevention and control framework.

4.3. Limitations and Prospects

Solid theoretical support for strengthening coal mine safety management and reducing the incidence of gas explosion accidents is provided by this study. It also provides useful insights and references for the application of root–state risk identification methods and complex networks in the field of accident cause analysis. The study has important theoretical value and practical significance, but there are still several limitations. (1) The accident causation problem studied here needs to be based on a large number of accident cases to support accurate decision-making for accident prevention. However, the existing coal mine gas explosion accident reports are not all comprehensive. Some reports only extract limited data and may ignore certain important factors, which hinders the accuracy of the auxiliary decision-making process. (2) Accident case analysis and manual coding are time-consuming, which may reduce efficiency to a certain extent. Although manual coding is feasible with a limited sample, it is challenging to apply it to a large number of accident reports and may introduce subjectivity into the research results. Future research can combine text mining tools and use a combination of qualitative and quantitative factor identification methods to improve the comprehensive efficiency of factor identification. (3) This study is based on accident data from Chinese coal mines, which may limit the generalizability of the results. Nevertheless, the proposed network-based risk analysis framework is constructed from universal topological metrics—degree centrality, betweenness centrality, clustering coefficient, and PageRank—which are not region-specific. Hence, the framework can be adapted to other coal mining systems, such as those in Australia, Poland, and India, with appropriate parameter calibration based on local operational and regulatory conditions. Future research using cross-national datasets, especially from European countries with different safety management practices, could further validate and extend the applicability of the proposed approach.

5. Conclusions

In this study, root–state risk identification (RSRI) theory is combined with complex network theory to construct a directed weighted causal network of coal mine gas explosion accidents. This method effectively bridges qualitative accident analysis and quantitative system modeling, and provides a new perspective for deeply understanding the structural mechanisms behind accident evolution. The proposed framework not only provides theoretical support for the risk assessment of coal mine gas explosion accidents but also lays a foundation for further application and verification in different countries, thus improving its universal applicability. The main conclusions are as follows:

(1)

The proposed root–state risk factor identification framework transforms descriptive information in accident reports into a structured causal network, which helps to clarify the direct and indirect relationships between risk factors. This method reduces the subjectivity in traditional accident cause analysis and provides a more complete system-level view.

(2)

A causal network model of coal mine gas explosion accidents is constructed, which provides a systematic analysis framework for understanding the formation and evolution of accidents. It reveals the important role of key factors and their paths in accident evolution and emphasizes the importance of identifying high-impact nodes in the early stage of accident propagation. Network simulation further shows that controlling key nodes and key links can significantly reduce system connectivity, which provides a basis for formulating more targeted security intervention measures.

(3)

Although the research data were obtained from gas explosion accident reports in China, the causal network construction method itself has strong portability. By adjusting the definition of factors, relationship structure, and weight setting, the framework can be applied to other mine types and complex industrial systems to identify high-risk factors and assist in formulating priority intervention strategies.

This study provides a structured perspective for understanding the systematic risk characteristics of gas explosion accidents and also provides methodological support for improving the pertinence and effectiveness of coal mine safety management. Future research can further expand data sources, explore multi-information source fusion analysis, and evaluate the applicability of this framework in different scenarios.