Introduction

Currently, railway construction is extending into areas with extreme geology and ecological sensitivity on an unprecedented scale. China is at a critical stage of high-quality railway development. This has made “environment-engineering” coupling risks a new research focus and posed potential risks for tunnel construction1]– [2. However, railway construction in complex and dangerous areas is often difficult to fully analyze and understand, with risks being highly uncertain, sudden, complex, and destructive3,4,5,6. They not only affect railway system operations but also pose significant threats to construction personnel7]– [8. Therefore, to ensure no major safety accidents occur in railway construction in complex areas, it is essential to first enhance the ability to identify and control safety risks and build a more effective safety risk control system9.

Risk identification, as the primary link of risk management, refers to the use of appropriate methods to identify risk factors in project implementation, which runs through the entire lifecycle of railway engineering construction10. It is the prerequisite and foundation for safety risk assessment and control11. To this end, it is of great significance to scientifically, continuously, and comprehensively identify potential risk factors in each stage of railway engineering construction. Many researchers have conducted in-depth research on safety risk identification. Goh et al.12 introduced the method of Case Based Reasoning (CBR) in building safety risk identification, which improved the efficiency and quality of risk identification. Sharafat et al.13 proposed the use of the Bow-tie method as a novel risk analysis approach to systematically assess and manage risk factors related to tunnel boring machines (TBMs) under complex geological conditions. Dai & Zhao14 integrated shallow-buried tunnel engineering, and based on expert questionnaire surveys, improved the Analytic Hierarchy Process (AHP) by optimizing the transmission matrix algorithm to construct a multi-factor influence weight matrix. As it is also used in construction engineering for risk assessment15. They also employed a multi-level fuzzy comprehensive evaluation method to conduct a fuzzy assessment of the influencing factors. Khademi et al.16 utilized the Fuzzy Analytic Hierarchy Process (Fuzzy AHP) to identify the main risk factors in TBM construction. Leitner17 analyzed common risk assessment models in the Slovak railway system and pointed out that identifying risks requires clarifying the definition and composition of risks. He analyzed the causes and development sequence of risk events and proposed five steps for risk identification and risk event development. Zou et al.18 prioritized risks based on their impact on typical project objectives, including cost, time, quality, safety, and environmental sustainability, and pinpointed 25 key risk factors. Vishwas et al.19 ranked the risk factors in construction and analyzed them using a combination of qualitative and quantitative methods. Liu et al.20 proposed the Root State Hazard Identification (RSHI) method to address the shortcomings of traditional methods that lack systematicity and comprehensiveness. The method was used to identify hazards in coal mine underground risk management. The results showed that the RSHI method can comprehensively identify root and state hazards related to coal mine underground risk management. Li et al.21 explored a critical hazard identification method for railway accident prevention and proposed a new accident causal network to simulate the interaction between hazards and accidents. The results showed that the critical hazard identification method based on integer programming combined with the proposed weighted direction accident causal network considering length had the best performance in accident prevention. In addition, the PTF-VIKOR method based on prospect theory has shown significant advantages in the risk assessment of prefabricated building construction22. This method takes into account the psychological and behavioral characteristics of decision-makers, and can more accurately reflect the risk preferences and uncertainties in the actual decision-making process23. Prospect theory has been widely applied in risk assessment. Chang and Zhao24 proposed a PTF-VIKOR model based on prospect theory, which establishes a risk assessment index system for prefabricated building construction from five aspects: personnel, equipment, management, technology, and environment, and uses interval-valued Pythagorean fuzzy numbers (IVPFNs) for weighting, overcoming the limitations of traditional rationality assumptions. Chang et al.25 also proposed the IVTSFS-CPT-EDAS model to assess the impact of subway tunnel construction on existing buildings, which has been proven effective through case validation and improves the accuracy of assessment. These studies show that prospect theory has significant advantages in risk assessment.

The above research shows that safety risk factor identification in railway tunnel construction plays a vital role in risk control. Current studies mainly focus on static weights or unidirectional causality. The cases studied are concentrated in urban subways, coal mines, or general mountain tunnels26. There is a lack of systematic quantification in ecologically sensitive areas, and no clear risk factor prioritization list has been proposed. Railway tunnel construction in ecologically sensitive areas faces challenges such as large engineering scale, complex construction environment, long construction period, and high construction requirements. Risk factors are often hidden in various stages of construction and interact with each other, making risk identification and analysis difficult27.

This study, based on the new Xi’ning-Chengdu Railway Xi’ning-Huangshengguan section, investigates risk factor identification and optimization for railway tunnel construction in ecologically sensitive areas. First, it extracts the core influencing factors for railway construction risk identification. Then, using the decision-making and experimental evaluation laboratory and interpretive structural model methods, it constructs an adversarial multi-level hierarchical structure model (DEMATEL-AISM). This model calculates the importance and coupling relationships of various influencing factors in railway construction risk identification, reveals the hierarchical relationships between factors, and improves the theoretical system of railway construction risk identification. The research approach of this article is shown in Fig. 1.

Fig. 1
figure 1

Flow chart of DEMATEL-AISM research on influencing factors.

Full size image

Risk factors for railway tunnel construction in ecologically sensitive areas

The Xi’ning-Chengdu Railway section from Xi’ning to Huangshengguan is located in Ruoergai County, Aba Prefecture, Sichuan Province. The line traverses the transitional zone between the eastern edge of the Qinghai-Tibet Plateau and the Loess Plateau, with a general north-south orientation. It passes through the Kahaerqiao Wetland Nature Reserve at multiple points, has a total length of approximately 35.4 km, and is a railway project in an ecologically sensitive area.

The Kahaerqiao Wetland Nature Reserve has a flat and open terrain, with an altitude ranging from 3,400 to 3,800 m. Peat soil, which is widely distributed, is characterized by high moisture content, low strength, and flammability, making it highly sensitive to construction activities. Improper construction practices can easily trigger grassland fires, especially during dry seasons and periods of high wind speed, significantly increasing fire risks. Additionally, improper excavation and backfilling of peat soil during construction, or inadequate maintenance measures, can lead to peat soil loss, thereby damaging surface vegetation and the wetland ecosystem. The reserve also features a well-developed water system and abundant water resources. However, factors such as wastewater from construction, hazardous chemical leaks, and improper disposal of domestic waste can cause water pollution and pose threats to the health of the wetland ecosystem. Furthermore, excessive precipitation, inaccurate estimation of tunnel water inflow, and unreasonable design of waterproofing and drainage systems can lead to a drop in the water level, impacting the survival environment of wetland ecosystems and wild flora and fauna. Noise, exhaust emissions, and waste generated by construction activities can also destroy the habitat of wildlife, disrupting their normal life and reproduction. Improper design of borrow pits and disposal sites, as well as illegal dumping of construction waste and spoil, can directly damage wild vegetation, reduce vegetation coverage, and impair ecological functions. In summary, the construction of the ecological sensitive area of the Xi’ning-Chengdu Railway from Xi’ning to Huangshengguan faces risks in multiple aspects, including grassland fires, peat soil loss, water resource pollution, destruction of wildlife habitats, damage to wild vegetation, and a decline in water levels. These risks are interrelated and influence each other. Effective risk control measures must be implemented during construction to ensure construction safety and the protection of the ecological environment.

Based on existing literature data and similar engineering, analyze the clear and potential influencing factors in the environmental risk identification process of railway construction in ecologically sensitive areas, and merge the influencing factors with similar meanings28. The influencing factors of railway engineering risk identification, including grassland fires, peat soil loss, water resource pollution, damage to wildlife ecological environment, damage to wild vegetation, and decrease in water level, are preliminarily summarized in Table 1. In the ecologically sensitive Ruoergai Wetland region, “construction season” typically refers to the period from mid-April to mid-October each year. During this interval, temperature and precipitation conditions are relatively favorable, allowing concrete curing, tunnel excavation, and support installation to meet quality requirements. Outside this window, construction safety, quality, and schedule are significantly impacted.

Table 1 Factors influencing risk identification in railway engineering.
Full size table

This study further selected risk identification influencing factors by designing a questionnaire and combining the knowledge and experience of experts and scholars in relevant research directions, as well as on-site workers with practical experience. Survey respondents to determine the degree of influence of various influencing factors on risk identification, and assign values of 0 to 4 points for non conformity, non conformity, relatively conformity, and very conformity, respectively.

In this study, a total of 23 questionnaires were distributed, with 21 valid responses received, indicating a response rate of 91.30%. The survey targeted experts in railway engineering, environmental protection, and risk management, encompassing fields such as geological engineering, environmental science, civil engineering, and safety engineering. Respondents had an average work experience of over 7 years.

Through statistical analysis of the collected data, 17 common factors were identified as the core influencing factors for risk identification in railway construction within ecologically sensitive areas. These factors are: surface peat soil (S1), surface vegetation (S2), construction season (S3), climate condition (S4), building type (S5), building land occupation and height (S6), engineering line position (S7), construction methods (S8), construction equipment and materials (S9), operation level (S10), design of waterproofing and drainage system (S11), spoil dump design (S12), pollutant handling (S13), garbage handling (S14), inventory of hazardous chemicals (S15), environmental awareness (S16), and regulatory system (S17).

Construction of risk factor model based on DEMATEL-AISM

Methodology

This study tackles the issue of risk factor identification and optimization in railway tunnel construction within ecologically sensitive areas. Using the DEMATEL-AISM method, which combines the Decision Making Trial and Evaluation Laboratory (DEMATEL) and the Adversarial Interpretive Structural Model (AISM), we’ve built a multi-level hierarchical model. This model dynamically quantifies the interdependencies among risk factors. DEMATEL analyzes the mutual influences of factors, while AISM structures these influences hierarchically. It helps in understanding the complex relationships among risk factors and provides a systematic way to identify and optimize them, improving the risk management in railway tunnel construction projects.

Decision-making Trial and Evaluation Laboratory analysis29 (DEMATEL) establishes logical relationships and direct influence matrices among system elements. It calculates each element’s degree of influence and being influenced, enabling the computation of centrality and cause degree. These metrics determine causal relationships and each element’s position in the system30]– [31. In this study, expert questionnaires collected ratings for 17 risk factors, constructing direct influence matrix O. After normalization, normalized direct influence matrix N and comprehensive influence matrix T were derived. Matrix T reflects direct and indirect influences among factors. Adversarial Interpretive Structural Model (AISM) integrates adversarial thinking into Interpretive Structural Modeling (ISM). It analyzes elements and their interdependencies in complex systems32. In this study, AISM was used for risk factor analysis. Fuzzy reachable matrix (FR) and reachability matrix (R) were constructed, and threshold λ was selected by considering factors like hierarchy and loop elements, ensuring the model’s effectiveness for key risk factor identification.

Given the numerous and complex risk factors in the Xi’ning-Chengdu Railway Xi’ning-Huangshengguan section, and their interdependent and evolving relationships, this study uses the DEMATEL-AISM model to identify and assess root risk factors in ecological sensitive railway tunnel construction. Based on the obtained risk factors for such areas, this model clarifies their structural hierarchy. The specific process is shown in Fig. 2.

Fig. 2
figure 2

Flow chart of DEMATEL-AISM research on influencing factors.

Full size image

Among them, O is the Direct impact matrix; N is the Normalized direct effect matrix; T is the Comprehensive effect matrix; FR is a Fuzzy reachable matrix; R is the Reachability matrix; R’ is the Contracted reachability matrix; S’ is the Skeleton matrix; S is the General skeleton matrix; WS is the comprehensive loop-labeled general skeleton matrix; D|C|M|R represents the Influence degree, Affected degree, Cause degree, and Centrality degree.

DEMATEL model

(1) Direct impact matrix and Comprehensive effect matrix.

The DEMATEL algorithm is a solution for solving complex system problems by determining the relationships between multiple factors. Based on the risk identification influencing factors selected in this study, combined with a survey questionnaire, the key influencing factors were quantitatively scored, and the degree of influence between each influencing factor was statistically analyzed. A 0–4 evaluation scale was used (0 = no influence, 1 = small influence, 2 = general influence, 3 = strong influence, 4 = strong influence) to measure the degree of influence between elements within the system, as shown in Table 2. Construct a direct impact matrix for each risk identification influencing factor, O=[Oij]n×n. As shown in Table 3, it is specified that when i = j, Oij = 0.

Table 2 Expert evaluation semantic scale.
Full size table
Table 3 Direct impact matrix O.
Full size table

Firstly, Eq. (1) is used to normalize the direct impact matrix, which means dividing all numbers by the maximum sum of each row in the relationship matrix to calculate the normalized direct effect matrix N, as shown in Table 4.

Table 4 Normalized direct effect matrix N.
Full size table
$$N={\left( {\frac{{{O_{ij}}}}{{Max\left( a \right)}}} \right)_{17 \times 17}}$$
(1)

In the formula, a is the set of sums for each row.

Secondly, the comprehensive impact matrix T=[tij]17 × 17 can be obtained through Eq. (2). Further quantify the degree to which each risk identification influencing factor affects other factors, and obtain a comprehensive effect matrix T as shown in Table 5.

$$T=\left( {N+{N^2}+{N^3}+ \cdots +{N^{\text{k}}}} \right)=\sum\limits_{{k=1}}^{\infty } {{N^k}=N{{\left( {1 - N} \right)}^{ - 1}}}$$
(2)

In the formula, \({\left( {1 - N} \right)^{ - 1}}\) is the inverse matrix of \(\left( {1 - N} \right)\).

Table 5 Comprehensive effect matrix T.
Full size table

(2) Calculation of influence degree, affected degree, cause degree, and centrality degree.

According to the comprehensive impact matrix, calculate the influence degree D and the affected degree C using Eq. (3). D is the sum of the elements in each row of the comprehensive effect matrix T, which means the comprehensive impact value of a certain element on other elements. The larger the value, the greater the impact degree33. C is the sum of the elements in each column of the comprehensive effect matrix T, which means the comprehensive impact value of a certain element by other elements. The larger the value, the greater the degree of influence. Calculate the centrality degree and cause degree using Eq. (4), and the calculation results are shown in Table 6. Thus, draw a scatter plot with centrality as the horizontal axis and causality as the vertical axis, as shown in Fig. 3. Among them, centrality is a positive indicator, and its value reflects the level of importance. The sign of causal degree reflects the impact of a certain element on other elements. A value greater than 0 indicates a greater impact on other elements, that is, causal elements; A value less than 0 indicates that it is more influenced by other factors, that is, the outcome factor34.

$${D_i}=\sum\nolimits_{{j=1}}^{n} {{t_{ij}}} ,{C_i}=\sum\nolimits_{{j=1}}^{n} {{t_{ij}}}$$
(3)
$${M_i}={D_i}+{C_i},{R_i}={D_i} - {C_i}$$
(4)
Table 6 Calculation results of cause degree and centrality degree of various factors.
Full size table
Fig. 3
figure 3

Scatter plot of various influencing factors cause degree and centrality degree.

Full size image

AISM model

(1) Constructing Fuzzy Reachable Matrix FR.

The fuzzy reachable matrix FR is obtained by multiplying the fuzzy multiplication matrix FB with the maximum and minimum operators until the product remains unchanged. The calculation process of the fuzzy reachable matrix is shown in Eq. (5).

$$F{B^{(k - 1)}} \ne F{B^k}=F{B^{(k+1)}}=FR$$
(5)

The fuzzy multiplication matrix FB represents the comprehensive influence matrix, where all the main diagonals become 1. The calculation formula is shown in Eq. (6).

$$FB=T+I$$
(6)

In the formula, I is the identity matrix, which is a Boolean square matrix with only diagonal 1.

The fuzzy operator adopts the Chad operator, which is the maximum minimum operator, as shown in Eq. (7).

$${c_{ij}}=\sum\limits_{{k=1}}^{n} {{b_{ij}} \wedge } {b_{kj}}=\left( {{b_{i1}} \wedge {b_{1j}}} \right) \vee \left( {{b_{i2}} \wedge {b_{2j}}} \right) \vee \left( {{b_{i3}} \wedge {b_{3j}}} \right) \cdots \vee \cdots \left( {{b_{in}} \wedge {b_{nj}}} \right)$$
(7)

The fuzzy reachable matrix FR can be calculated as shown in Table 7.

Table 7 Fuzzy reachable matrix FR.
Full size table

(2) Determine threshold λ.

The set of non repeating values in the fuzzy reachable matrix FR, except for the number 0, is called the threshold set. Within the range of intercept values of (0,1], 54 structures are obtained, and the hierarchical statistics are shown in Table 8. Among them, the intercept values of (0,1] in the fuzzy reachable matrix FR are taken, and the obtained intercept matrix is the reachability matrix.

Table 8 Topology layer level statistics corresponding to values.
Full size table

The selection of variable λ is crucial for identifying the key factors affecting the construction of railway tunnels in ecologically sensitive areas. According to Table 8, the statistical charts of 54 structures number of levels, number of connected domains, number of loops, the maximum number of elements contained in the loop, and the number of structures are shown in Figs. 4, 5, 6, 7 and 8, respectively.

Fig. 4
figure 4

Bar chart of changes in the number of levels.

Full size image
Fig. 5
figure 5

Bar chart of changes in the number of connected domains.

Full size image
Fig. 6
figure 6

Bar chart of changes in the number of loops.

Full size image
Fig. 7
figure 7

Bar chart of changes in the maximum number of elements contained in the loop.

Full size image
Fig. 8
figure 8

Bar chart of changes in the number of structures.

Full size image

After multiple data simulations, the following five thresholds have been summarized λ The principle of determination.

1) The more levels there are, the better. Hierarchical division of influencing factors, the more levels there are, the better it can reflect the interaction relationship between elements. As shown in Fig. 4, the number of levels gradually increases with the increase of intercept values (i.e. structural numbering), followed by a decrease.

2) The fewer isolated systems, the better. As shown in Fig. 5, the number of independent regions increases with the increase of intercept values.

3) The more circuits there are, the better. From Fig. 6, it can be seen that as the intercept value increases, the trend of the number of loops in the system first increases and then decreases. After the 28th structure, the number of loops is all 0.

4) There is no major circuit (the maximum circuit containing more than 5 elements belongs to the major circuit). From Fig. 7, it can be seen that each structure has a loop, and the maximum number of elements contained in all loops shows a decreasing relationship as the intercept value increases.

5) The larger the number of structures, the better. From Fig. 8, it can be seen that the number of structures is not necessarily related to the size of the intercept value.

Based on the above five principles, select the top five structures with the highest rankings, as shown in Table 9.

Table 9 Top five structures.
Full size table

The selection of the optimal threshold value (λ) was based on a comprehensive evaluation of several key factors, including the number of levels, the number of connected domains, the number of loops, and the maximum number of elements within the loops. Adhering to the five established principles, a detailed analysis was conducted on the top five structures listed in Table 8. The structure corresponding to serial number 24 achieved a level count of 5, which aligns with the principle that a higher number of levels allows for a more detailed depiction of the relationships between elements. Compared to other structures, it offers a more refined hierarchical division. Additionally, this structure has 3 connected domains, which is a relatively moderate number, avoiding the pitfalls of excessive fragmentation due to too many domains or the masking of associations due to too few. Furthermore, it has 1 loop with a maximum of 3 elements, which reflects the cyclical feedback relationships between risk factors without creating overly complex large loops. Considering all these factors, the structure numbered 24 performed well across multiple key indicators, leading to the selection of λ = 0.29751 as the optimal threshold value.

(3) Build adjacency matrix.

Based on the comprehensive effect matrix T and intercept λ construct the adjacency matrix A, which reflects the influence relationship of various factors in matrix form. The calculation method is shown in Eq. (8):

(8)

Among them: threshold λ = 0.29751, the adjacency matrix A calculated using the feature threshold method is shown in Table 10.

Table 10 Adjacency matrix A.
Full size table

(4) Establish reachability matrix and general skeleton matrix.

For the adjacency matrix A, it is first transformed into a multiplication matrix B through Eq. (9), and then multiplied by B (Eq. (10)) until the matrix remains unchanged, resulting in the reachable matrix R as shown in Table 11.

$$B=A+I$$
(9)
$${B^{(k - 1)}} \ne {B^k}={B^{(k+1)}}=R$$
(10)
Table 11 Reachability matrix R.
Full size table

To perform node contraction on the reachability matrix R, where loops in the reachability matrix are treated as single nodes, resulting in the contracted reachability matrix R’. The resulting contracted reachable matrix R’ is illustrated in Table 12.

Table 12 Contracted reachability matrix R’.
Full size table

Then perform edge reduction operation to remove duplicate paths. Shrink the edges of R ‘to obtain the skeleton matrix S’, and substitute the loop elements into the general skeleton matrix S as shown in Table 13. The method is as shown in Eq. (11):

$${S^{\prime}}={R^{\prime}} - {({R^{\prime}} - I)^2} - I{S^{\prime}}+I={R^{\prime}} - {({R^{\prime}} - I)^2}$$
(11)
Table 13 General skeleton matrix S.
Full size table

The results of the Analytic Hierarchy Process and the general skeleton matrix S can establish a hierarchical structure model of risk factors for railway tunnel construction in ecologically sensitive areas, with adversarial comprehensive impact coefficients. Firstly, replace the influence values in the comprehensive impact matrix T with the skeleton matrix S to obtain the general skeleton matrix TS matrix with comprehensive values, as shown in Table 14. Then, by replacing the values with “1” in the loop relationship, obtain the comprehensive loop-labeled general skeleton matrix, as shown in Table 15.

Table 14 General skeleton matrix TS with comprehensive values.
Full size table
Table 15 Comprehensive loop-labeled general skeleton matrix WS.
Full size table

Hierarchical extraction

For any reachable matrix, there is a reachable set R, an antecedent set Q, and a common set T, where T = R∩Q. Taking the adjacency matrix A as an example, the reachable set of ei is denoted as R(ei), which refers to all elements with a row value of 1 where the factor is located. The antecedent set of ei is denoted as Q(ei), which refers to all elements in the column where the factor is located with a value of 1. The common set of ei is denoted as T(ei), i.e. R(ei)∩Q(ei). The UP type hierarchical graph, which is a result first hierarchical division, has an extraction rule of T(ei) = R(ei). As long as the reachable set is the same as the common set, extract the relevant elements. Place the extracted elements above each time and place them in order from top to bottom. A DOWN type hierarchical chart, which is a cause first hierarchical division, is extracted based on the rule of T(ei) = Q(ei). The extracted elements are placed below each time, and the extracted elements are placed in order from bottom to top. Perform hierarchical extraction according to S + I, and the results are shown in Table 16.

Table 16 Adversarial level extraction results.
Full size table

Drawing topology hierarchy diagram

Based on the relationships between elements and the extraction results of confrontation levels, a directed hierarchical topological hierarchy diagram can be created. The reachable relationship of influencing factors during the construction process is represented by directed line segments. Bidirectional arrows indicate the formation of loops, signifying mutual reachability relationships. Furthermore, the lower the level, the more fundamental the influencing factors, while the higher the level, the more direct the influencing factors35. The topology hierarchy diagrams for UP and DOWN types are depicted in Fig. 9.

Fig. 9
figure 9

Topological hierarchy diagram.

Full size image

Model result analysis

Cause degree and centrality degree analysis

Centrality is an indicator of the degree to which various influencing factors in a system affect project construction. The results in Fig. 3; Table 6 show that the top 5 risk factors in the centrality of railway tunnel construction in ecologically sensitive areas are environmental awareness (S16), regulatory system (S17), construction methods (S8), construction equipment and materials (S9), and design of waterproofing and drainage system (S11), indicating that these factors have a relatively high impact on the construction of railway tunnels in ecologically sensitive areas. The five influencing factors with lower centrality are climate condition (S4), engineering line position (S7), operation level (S10), inventory of hazardous chemicals (S15), and construction season (S3), indicating that these factors have a lower impact on the construction of railway tunnels in ecologically sensitive areas. However, this does not mean that these factors are not important, but rather have lower importance compared to other factors.

Cause degree is used to distinguish the causal and outcome attributes of factors. If the cause degree is greater than 0, this factor is a causal attribute, and the larger its value, the more significant its impact on other factors. If the cause degree is less than 0, this factor is a result attribute, and the larger its absolute value, the greater its degree of influence from other factors. The results in Fig. 3; Table 6 show that the top five factors with the strongest causal attributes among all influencing factors are climate condition (S4), construction season (S3), engineering line position (S7), surface peat soil (S1), and building land occupation and height (S6). The top five ranking with the strongest result attributes are design of waterproofing and drainage system (S11), pollutant handling (S13), spoil dump design (S12), regulatory system (S17), and garbage handling (S14). When controlling project construction risks, attention should be paid to effective control of high-altitude factors, as controlling these factors can indirectly control other factors through their influence; For outcome factors, not only can they be controlled themselves, but also the causal factors that affect these outcome factors can be controlled.

Activity system analysis

According to the extraction rules, the manifestation layer factor of the extraction results under the UP result priority rule is the direct cause of the system, while the root order of the extraction results under the DOWN reason priority rule is the fundamental cause of the system, which is the meaning of AISM. Compared to traditional ISM models, AISM can more accurately reflect the interaction relationship between risk factors in railway tunnel construction in ecologically sensitive areas and the strength of the relationship due to the simultaneous adoption of two extraction rules. From Fig. 9, it can be seen that both the UP and DOWN structural models contain five levels, from bottom to top, from cause to result. This system is an extensible system with active factors such as surface peat soil (S1), construction season (S3), building land occupation and height (S6), engineering line position (S7), spoil dump design (S12), pollutant handling (S13), and inventory of hazardous chemicals (S15). There are isolated elements in the form of building (S5) and operational level (S10). After removing isolated elements, The UP type causal series are {S9, S11, S12, S13}>{S3, S8}>{S14, S16, S17}>{S1, S2, S6, S7, S15}>{S4}, while the DON type causal series are {S9, S11}>{S8, S12, S13}>{S14, S16, S17}>{S2}>{S1, S3, S4, S6, S7, S15}.

Loop analysis

A loop, also known as strong connectivity, is formed when several mutually causal factors are interconnected end to end. All elements within a loop can be categorized as a subsystem. The garbage handling (S14), environmental awareness (S16), and regulatory system (S17) in Fig. 9 form a loop, where surface peat soil (S1), surface vegetation (S2), building land occupation and height (S6), engineering line position (S7), and inventory of hazardous chemicals (S15) are the direct reasons for the appearance of loop S14, S16, and S17. In the hierarchical topology diagram, different causal factors interact with other factors as a whole. Therefore, in the process of constructing railway tunnels in ecologically sensitive areas, integrated control and overall management of factors in the loop should be carried out to minimize the adverse effects of risk factors on construction.

Hierarchy analysis

The risk factors in the construction of railway tunnels in ecologically sensitive areas can be divided into three levels and five layers from top to bottom, namely the surface level (layer 0), the middle level (layers 1, 2, and 3), and the root level (layer 4). The hierarchical factors are the union of the top-level elements, which only receive arrows, meaning they are all affected and do not affect other elements. In the hierarchical topology diagram of risk factors in the construction process of railway tunnels in ecologically sensitive areas, the top layer factors that exclude isolated factors are merged to form the visible factors, which are construction equipment and materials (S9), design of waterproofing and drainage system (S11), spoil dump design (S12), and pollutant handling (S13). These factors are the most direct influencing factors. To swiftly and effectively mitigate risks, it is essential to begin with the control of surface-level factors. In order to quantify the overall risk reduction rate that may result from controlling these surface factors, a two-step weighting method is employed. Based on the Mi values from Table 6, the 17 factors are normalized to obtain their relative weights.

\({w_i}=\frac{{{M_i}}}{{\sum {{M_i}} }}\), Among them, \(\sum {{M_i}} =135.56\).

The combined weight of the four surface-level factors—construction equipment and materials (S9), design of waterproofing and drainage system (S11), spoil dump design (S12), and pollutant handling (S13)—is :

$${w_{surf}}=\frac{{\left( {8.897+8.351+8.315+8.162} \right)}}{{\sum {{M_i}} }} \approx 0.249$$

The author added a Likert question in the questionnaire: “If S9, S11, S12, and S13 are fully controlled, what percentage do you think the overall project risk can be reduced?” The analysis of the statistical results shows that the average response from 21 experts is 2.48. Therefore, the risk reduction rate \(\delta\) is calculated as follows:

$$\delta ={w_{surf}}*0.249=0.62$$

Additionally, it should be noted that this estimation is only for the effect of prioritizing the control of surface - level factors based on the results of this questionnaire. The actual effect still needs to be verified on - site.

In the surface-level factors, construction equipment and materials (S9), design of waterproofing and drainage system (S11), spoil dump design (S12), and pollutant handling (S13) are easily observable and measurable. The construction team can prioritize these surface-level factors and take specific measures such as selecting appropriate construction equipment and materials, optimizing the design of the drainage system, and properly handling pollutants to quickly and effectively reduce construction risks. These decisions are clearly executable in practice. The construction team can develop detailed operation plans and execution processes based on the specific needs and conditions of the project. For example, in selecting construction equipment and materials, the team can evaluate the performance, cost, and applicability of different equipment and materials according to the engineering requirements and technical specifications, and choose the best solution for the current project. In optimizing the design of the waterproofing and drainage system, a rational drainage layout and flood control measures can be designed based on geological conditions and climatic characteristics to ensure smooth and safe drainage during the construction process. Through these specific implementation measures, the risk control of surface-level factors can be effectively implemented, thereby achieving a significant reduction in project risks.

The intermediate factor set, which refers to the factors located in the middle of the hierarchical topology model graph, sends upward directed line segments to the upper level factors to influence them. There are many intermediate factors in this system, and this factor set spans three levels, with a total of 10 factors, including surface peat soil (S1), surface vegetation (S2), construction season (S3), building land occupation and height (S6), engineering line position (S7), construction methods (S8), garbage handling (S14), inventory of hazardous chemicals (S15), environmental awareness (S16), and regulatory system (S17), The above factors are the core risk factors for the construction of railway tunnels in ecologically sensitive areas. They are located at the center of the entire system and are influenced by the root cause, playing a crucial role in connecting the apparent and root cause factors. They should be given priority consideration in the risk factor control process. In practice, project managers can develop corresponding management strategies and operational procedures based on the characteristics of these factors. For example, regarding construction season (S3), managers can reasonably schedule construction progress according to climate forecasts and engineering requirements, avoiding high-risk working periods under adverse climatic conditions and selecting suitable seasons for key construction phases to mitigate the adverse effects of the natural environment on construction. Meanwhile, by enhancing training and education on environmental awareness (S16), the environmental consciousness and sense of responsibility of construction personnel can be improved, encouraging them to voluntarily adopt environmental protection measures during the construction process. This indirectly influences surface-level factors and achieves effective control of project risks. In addition, optimizing construction methods (S8) by introducing advanced construction techniques and processes can improve construction efficiency and quality, reducing construction risks. Rational planning of the land occupation and height of building (S6) and engineering alignment (S7) can minimize ecological damage and reduce environmental risks.

The root level is the fifth layer, which includes only climate condition (S4). This layer serves as the foundation of the entire risk factor system and has a profound impact on factors in all other layers. As a root cause, climate condition (S4) determines the types and degrees of risks that the project may face in different seasons and climatic environments. In practice, project managers can develop long-term project plans and risk management strategies based on climate condition forecasts. For example, in rainy seasons, flood prevention and drainage measures should be prepared in advance; in cold regions, effective insulation and anti-freezing measures should be taken to ensure the safety of construction personnel and the normal operation of construction equipment. By incorporating climate conditions into project planning and decision-making processes, project risks can be fundamentally reduced, and the success rate of the project can be increased.

Discussions

The present study introduces the DEMATEL-AISM model to dynamically quantify the risk factors of railway tunnel construction in ecologically sensitive areas and constructs a five-layer hierarchical structure to reveal the interdependencies among these factors. This approach not only addresses the limitations of existing research in terms of methodological integration but also offers a novel theoretical framework for risk identification in railway tunnel construction within ecologically sensitive areas. Compared with traditional risk identification models such as AHP, FMEA, and FTA, the DEMATEL-AISM method demonstrates significant advantages. Unlike AHP, DEMATEL-AISM can handle bidirectional interdependencies among factors, precisely locating key risk factors through the calculation of influence and influenced degrees, and clearly presenting the hierarchical relationships among risk factors. This enables decision-makers to formulate more targeted risk control strategies based on dynamic quantification results. Unlike FMEA, which mainly focuses on failure modes of products or processes, DEMATEL-AISM takes a systemic approach to comprehensively analyze the interactions among risk factors, providing a complete risk profile from macro to micro levels for railway tunnel construction, which aids in early identification and management of potential risks. Compared with FTA, which constructs a fault tree to analyze the causes of system failures, DEMATEL-AISM uses fuzzy reachable and reachability matrices to stratify risk factors. It can not only identify direct risks but also trace back to root causes and predict potential risks, providing more precise directions for risk prevention and control. Moreover, compared with the single ISM model, DEMATEL-AISM further quantifies the weights between factors by introducing the DEMATEL method, making the model more scientific and practical. These advantages enable DEMATEL-AISM to effectively integrate multi-source data in the risk assessment of complex ecological sensitive railway tunnel construction, providing more accurate risk prioritization and control strategies, and injecting new vitality into project risk management.

The DEMATEL-AISM model proposed in this study is not only theoretically innovative but also of significant practical value. It provides decision-makers, contractors, and regulatory bodies with a scientific and systematic evaluation framework to identify and manage risks in railway tunnel construction in ecologically sensitive areas. In practice, decision-makers can use this model to identify key risk factors and develop corresponding risk management strategies. For example, during the project planning and design phase, fundamental risk factors such as climate conditions, construction seasons, and engineering alignment should be given priority. During the construction process, direct influencing factors like construction equipment and materials, drainage system design, spoil dump design, and pollutant handling should be controlled first. By controlling these key factors, the overall risk in the construction process can be effectively reduced. Moreover, the model can also be used for real-time risk monitoring and dynamic risk assessment. By regularly updating expert rating data, changes in risk factors can be monitored in real-time, and risk management strategies can be adjusted promptly. This not only enhances the scientific nature and effectiveness of risk management but also provides a strong guarantee for the smooth implementation of the project.

Conclusions

The construction of railway tunnels in ecologically sensitive areas is influenced by various risk factors, and there are complex interactions among these risk factors. Compared with traditional ISM, this article adopts a more innovative DEMATEL-AISM method to study the intrinsic relationships between these factors. Based on the analysis of the model results above, the following conclusions can be drawn:

(1) The main factors that have a high impact on railway tunnel construction in ecologically sensitive areas include environmental protection awareness (S16), regulatory system (S17), construction methods (S8), construction equipment and materials (S9), and design of waterproofing and drainage system (S11). These factors are of great theoretical and practical significance in risk control and are the core factors that project managers should focus on.

(2) The fundamental risk factors affecting the construction of railway tunnels in ecologically sensitive areas include climate condition (S4), construction season (S3), engineering line position (S7), surface peat soil (S1), and building land occupation and height (S6). When controlling project construction risks, attention should be paid to effective control of the above factors.

(3) Garbage handling (S14), environmental awareness (S16), and regulatory system (S17) form a loop. In the process of constructing railway tunnels in ecologically sensitive areas, integrated control and overall management of the factors in the loop should be carried out to minimize the adverse effects of risk factors on construction.

(4) The four most direct influencing factors are construction equipment and materials (S9), design of waterproofing and drainage system (S11), spoil dump design (S12), and pollutant handling (S13). Effective control of railway tunnel construction risks in ecologically sensitive areas can be achieved by controlling these surface level factors.

(5) Surface peat soil (S1), surface vegetation (S2), construction season (S3), building land occupation and height (S6), engineering line position (S7), construction methods (S8), garbage handling (S14), inventory of hazardous chemicals (S15), environmental awareness (S16), and regulatory system (S17) are the core risk factors for railway tunnel construction in ecologically sensitive areas, and should be given priority consideration in the risk factor control process.

Based on the above conclusions, the following suggestions can be proposed for risk management and control of railway tunnel construction in ecologically sensitive areas:

(1) Integrate management of garbage handling, environmental awareness, and regulatory system to ensure effective control of these key factors throughout the construction process.

(2) During the project planning and design phase, fully consider fundamental risk factors such as climate condition, construction season, engineering line position, surface peat soil, and building land occupation and height.

(3) During the construction process, emphasis should be placed on controlling the direct influencing factors. Construction risks can be reduced by selecting appropriate equipment and materials, designing effective drainage systems, and properly handling waste and pollutants.

Despite the achievements made in risk factor identification and optimization, this study still has some limitations. For example, the data mainly comes from expert ratings, and further collection of actual construction data is needed in the future to verify the accuracy and reliability of the model. Although the model performs well in railway tunnel construction in ecologically sensitive areas, its applicability to other types of projects still needs further validation. In addition, to facilitate understanding and application, the model has been simplified to some extent, which may not be able to fully capture the complex dynamics of all risk factors. Future research can consider extending the model to other systems (such as urban subways, bridges, etc.) to verify its universality, integrating real-time risk monitoring data to develop a dynamic risk assessment system to enhance the dynamic adaptability of risk management, and using empirical accident data to validate the effectiveness of the framework and further optimize model parameters.