Research on Integration of Safety Policy System in Petrochemical Enterprises Based on Risk Hierarchical Control and Hidden Danger Investigation

Abstract

To enhance the risk management capacity of petrochemical enterprises, this paper presents a systematic and in-depth study of risk hierarchical control and hidden danger investigation technologies. Firstly, a risk hierarchical control system was developed based on text mining and Risk Breakdown Structure (RBS) theory, categorizing risk alarm levels into four tiers: no alarm, light alarm, medium alarm, and heavy alarm. Secondly, a hidden danger investigation and management system was established by integrating a three-dimensional hidden danger grading model with the Plan-Do-Check-Act (PDCA) closed-loop principle. Finally, a cooperative management technology system for risk and hidden dangers in petrochemical enterprises was constructed and validated using Shandong Luqing Petrochemical Enterprise as a case study. The results indicated that the comprehensive risk level of Shandong Luqing Petrochemical Enterprise is classified as II, with a yellow light warning signal. They demonstrated a positive correlation between the risk hierarchical control system and the hidden danger investigation and management system. The findings of this research provide valuable guidance for improving safety management in petrochemical enterprises.

1. Introduction

The petrochemical industry plays a crucial role in the national economy and is heavily reliant on resources and energy. Due to their unique industrial characteristics, petrochemical enterprises engage in product processing and the use of raw materials, which can pose significant risks during production and operation. Issues that arise in production or management may lead to safety incidents, threatening not only the development of these enterprises but also the safety of their personnel and assets [1,2].

According to statistics from China’s General Administration of Safety Supervision, more than 320 accidents occurred in the petrochemical industry between 2010 and 2014, resulting in over 2200 fatalities [3]. Additionally, the Ministry of Emergency Management (MEM) reported 1514 petrochemical accidents in China from 2013 to 2022, which led to a cumulative total of 1977 deaths. In 2022, there were 127 petrochemical accidents and 143 fatalities. In 2023, 115 chemical accidents were recorded, resulting in 159 deaths [4]. A statistical map illustrating petrochemical accidents in China from 2013 to 2023 is presented in Figure 1.

As seen from the accident time distribution graph of petrochemical accidents in China in 2023 (Figure 2), the accidents in March after the Spring Festival, July in summer, and December in winter are significantly higher than the monthly average.

According to the 2010 annual report from the U.S. Council of Labor Affairs, the petrochemical industry has the highest incidence of injured and sickened workers, highlighting significant deficiencies in workplace safety, including equipment failures [5]. In 2015, Norozi and colleagues [6] reported that over 198 work-related fatalities occurred in Iranian petrochemical companies over the past decade, primarily due to ineffective management systems that failed to prevent major accidents.

Relative to other chemical industries, safety issues in the petrochemical industry tend to be characterized by the following: (1) The petrochemical industry involves many flammable and explosive chemicals, resulting in a higher risk of potential explosions and fires. (2) The production process of petrochemicals is usually complex, involving multiple reaction steps and equipment, increasing the likelihood of operational errors and equipment failures. (3) Accidents in the petrochemical industry may have a serious impact on the surrounding environment, including contamination of air, water, and soil.

The term risk first appeared in the United States and was used in the insurance industry, and Britain became the first country to study the major hazard sources [7]. Subsequently, Germany and other countries have enacted a series of ordinances for the control of major sources of danger and increased the management of major sources of danger [8]. Stankovska believes that the two important steps of risk management are risk identification and control, and risk should be quantitatively evaluated in order to facilitate enterprises to better understand and control risks [9]. At present, foreign countries classify the hidden danger investigation and technical rectification of petrochemical enterprises as risk control, and risk analysis from a single aspect is not enough to prevent its occurrence, and the identification of risks needs to start from multiple angles and all-round [10,11,12]. With the in-depth understanding of risk control and the development of related disciplines, more and more experts begin to establish risk assessment control models. For example: AHP/ANP (Analytic Hierarchy Process/Analytic Network Process), Monte Carlo simulation, Bayesian network, probability-consequence impact model, fuzzy set theory, etc., mainly check the results of the application of risk assessment methods in equipment technical transformation projects [13]. Wu conducted research on the safety risk state index system of petrochemical enterprises based on AHP [14]. Shin S studied the Petrochemical Chemical Accident Risk Index (KCARI), and he verified that the model can be used as a tool for early screening and management of companies at high risk of chemical accidents [15]. Based on data preprocessing and extension analysis, Han proposed a pre-risk assessment model for the safety extension risk of petrochemical plants [16]. Pan proposed a risk analysis model based on Bayesian networks (BNs), which clarified the accident risk characteristics and root causes of petrochemical companies [17]. Miao studied the dynamic risk control model of coal chemical enterprises, combined the optimized neural network with the control chart, and constructed the dynamic risk classification control algorithm of coal chemical enterprises [18]. Xing applied Hazard and Operability Study (HAZOP) and Bayesian networks (BNs) to identify chemical process hazards and infer risk causation, enabling systematic risk assessment of chemical processes [19]. Wu proposed a dynamic risk assessment method based on bow tie analysis and Bayesian network analysis [20]. Hou proposed the domino effect pattern recognition of tank farm accidents based on the data mining method [21].

The concept of hidden troubles was first introduced in the publication “Occupational Safety and Health Terminology” [22]. In 2013, China proposed to establish a hidden danger investigation and management system, incorporating big data into enterprise management to enhance the efficiency of risk management and control within enterprises [23]. In 2014, Liu proposed risk control as the central approach and advocated for the Plan-Do-Check-Act (PDCA) cycle, applying these principles to the electric power industry, which greatly improved the management quality of electric power enterprises and reduced the occurrence of accidents [24]. In 2016, China first proposed a risk classification and control system alongside a hidden danger investigation and management system [25,26].

In recent years, many scholars have made new research contributions to the dual prevention mechanism of risk classification management and hidden danger investigation in petrochemical enterprises [27,28,29,30]. These studies mainly analyze the issues present in the risk management of petrochemical enterprises and propose effective hidden danger investigation and management measures in combination with the actual situation [31,32,33]. Based on AHP and FTOPSIS, Chen established a classification system for hidden danger investigation in chemical enterprises [34]. Zou researched and analyzed the existing safety management system of CM Chemical Company, guided by the theories of Heinrich’s law [35]. Other scholars conducted research on a chemical enterprise through field studies [36,37,38]. Zhang conducted field research on liquid ammonia refrigeration enterprises and established a hidden danger investigation information system for liquid ammonia refrigeration enterprises based on the data obtained and the results of analysis [39]. Zhang sorted out the current situation of chemical enterprises in Binzhou City, investigated the process flow and equipment of bromine enterprises, identified risks, and graded bromine storage tanks based on a three-dimensional risk classification model [40].

In summary, petrochemical risk management both domestically and internationally mainly focuses on the following aspects: first, risk analysis based on safety assessment models; second, according to the main problems of petrochemical enterprises, the corresponding management countermeasures are put forward. However, the study of the dual prevention mechanism of risk classification and hidden danger investigation in petrochemical enterprises has become the weak link of risk control in petrochemical enterprises. Therefore, it is necessary to further study a more scientific risk classification and control system and put forward a more reasonable integrated system of petrochemical risk control and hidden danger investigation.

This paper has analyzed the current situation of petrochemical enterprises, using Shandong Luqing Petrochemical Enterprise as a case study. It has proposed a more scientific system for the collaborative management of risks and hidden dangers in these enterprises. Based on data analysis, this research has conducted an in-depth study into the relationship between risk classification and hidden danger management.

Many existing models typically utilize simple two- or three-level risk classification, such as low-medium-high risk taxonomy and red-yellow-green signaling systems, whereas this paper makes risk identification more refined by introducing four-level risk classification (no alarm, light alarm, medium alarm, and heavy alarm). Compared with static assessment methods, such as Analytic Hierarchy Process (AHP) and Failure Mode, Effects Analysis (FMEA), and SWOT analysis, this paper combines the closed-loop principle of PDCA. This integration enables the hidden danger identification and management system to continuously improve and dynamically adjust management strategies based on real-time data and feedback, which improves the flexibility and effectiveness of management. Many existing models have a single hidden danger assessment, while the three-dimensional hidden danger grading model proposed in this paper considers multiple dimensions (e.g., probability of risk, degree of impact, and control capability), providing a more comprehensive perspective that facilitates thorough assessment and response to hidden dangers. The research in this paper has significant implications for the safety management of petrochemical enterprises.

2. Materials and Methods

2.1. Risk Hierarchical Control System

RBS (Risk-Based Supervision) theory follows hierarchical management and control, serving as the theoretical foundation for a hierarchical risk management and control system. The purpose is to manage the risks identified within an enterprise at different levels, thereby maximizing the utilization of enterprise resources. In this paper, we construct a risk hierarchy control system based on RBS theory.

The risk classification and control system for petrochemical enterprises consists of four key sectors: determining the scope of risk identification, risk identification, risk assessment, and risk control.

The first sector: determine the scope of risk identification in petrochemical enterprises. Including “man”, “machine”, “environment”, “management”, “materials”, and “process flow”.

The second sector pertains to risk identification, which employs three methodologies: Safety Check List (SCL), Job Hazard Analysis (JHA), and Preliminary Hazard Analysis (PHA). SCL is utilized to identify risks associated with production equipment, materials, and management. JHA concentrates on identifying risks inherent in processes and unsafe human behaviors, while PHA evaluates risks related to the operational environment.

The third sector involves conducting risk assessment research. This process utilizes text mining technology to establish risk assessment indicators in conjunction with the analytical hierarchy process and variable weight theory to determine both individual weights and comprehensive weights of the assessment indicators. In the grading of risk evaluation for petrochemical enterprises, risks are assessed across three dimensions: likelihood, severity, and sensitivity. The comprehensive risk alarm levels are categorized into four tiers: no alarm, light alarm, medium alarm, and heavy alarm. The thresholds are defined as follows: Level I (r in [0, 2.21]) indicates a normal state; Level II (r in (2.21, 2.83]) signifies a yellow light alarm state; Level III (r in (2.83, 3.64]) represents an orange medium-alarm state; and Level IV (r in (3.64, 5]) denotes a red heavy-alarm state.

The fourth sector: the development of risk management and control measures. Risk control measures mainly include management measures, training and education measures, individual protection measures, engineering measures, and emergency measures.

The flow chart of the risk control system of petrochemical enterprises is shown in Figure 3.

2.2. Hidden Danger Investigation and Management System

The 3D Hazard Rating Model integrates the difficulty of hazard management, the severity of potential consequences, and the rate of change of protective barriers to provide a comprehensive assessment of hazards. By categorizing hazards into major and general hazards, the 3D model helps managers prioritize the most serious to manage hazards, ensuring that resources are focused on the most important issues that need to be resolved and improving governance efficiency. Combined with the PDCA cycle theory, the 3D model supports the dynamic monitoring and management of hidden dangers. Enterprises can timely adjust the classification and management strategies for hidden dangers according to the actual situation, achieving closed-loop management and ensuring continuous improvement. Through systematic classification and management of hidden dangers, petrochemical enterprises can effectively reduce potential safety risks, minimize the likelihood of accidents, and protect the lives of employees and the property safety of the enterprise.

The hidden danger investigation and management system includes three parts: hidden danger investigation, hidden danger classification, and hidden danger closed-loop management. This paper uses the three-dimensional hidden danger classification model to carry out hidden danger classification and carries out hidden danger closed-loop management based on the PDCA cycle theory. According to the characteristics of hidden danger existing in the production of petrochemical enterprises, a three-dimensional hidden danger classification model is established, with the x-axis indicating the difficulty of hidden danger management, the y-axis indicating the severity of the consequences that may result from the hidden danger, and the z-axis indicating the rate of change of the protective barrier of the hidden danger, with the red area indicating the major hidden danger and the white area indicating the general hidden danger. The three-dimensional hidden danger classification model is shown in Figure 4. The PDCA cycle theory was originally proposed by the American quality management expert Hugh Hart in the 1930s and then adopted and publicized by the American quality management expert Dr. Deming, so it can be widely used in enterprise management and quality management. The PDCA cycle is divided into four stages, namely Plan, Do, Check, and Act. These four stages constitute a closed cycle management system, which runs through the entire process and specific work items of enterprise management [41].

3. Discussion on the Symmetry of Risk Hierarchical Control System and Hidden Danger Investigation and Management System

3.1. Design of Risk Hierarchical Control System

The risk hierarchical management and control system includes four links: determining the scope of risk identification, risk identification, risk assessment, and risk management and control, and follows dynamic operation and continuous improvement.

3.1.1. Scope and Method of Risk Identification

The identification of hazard sources for petrochemical enterprises should cover all facilities and equipment, activities of operators, and working environment in the production process of enterprises, including:

(1)

Projects in planning and construction, commissioning, and operation;

(2)

Conventional and unconventional operational activities;

(3)

Accidents and potential emergencies;

(4)

The activities of the operator;

(5)

The transportation process of raw materials and products;

(6)

Facilities and equipment, safety and protective equipment;

(7)

Changes in facilities, equipment, operators, processes;

(8)

Discard, abandonment, dismantling, and disposal;

(9)

Climatic, geological, and environmental impacts.

The division of production systems is mainly based on the 4M elements of accidents (man, machine, environment, management) four aspects of identification. Considering the particularity of petrochemical enterprises, the two factors of material and process flow are added, so the identification of risk points should be from the “man, machine, environment, management, material, process flow” six aspects. SCL is used to identify the risks of production equipment, materials, and management; see

Table 1. List of facilities and equipment.

Serial Number	Equipment Name	Category	Location	Affiliated Unit	Special Equipment or Not	Remark
1
2

. JHA is used to identify the risks of process and human unsafe behaviors, and PHA is used to identify the risks of the operating environment.

3.1.2. Establishment of Evaluation Index System Based on Text Mining Technology

Text mining refers to the process of mining information with deep value for research or decision-making from a large number of texts through statistics, machine learning, and other methods [42]. In this study, 1514 petrochemical accidents that occurred in China from 2013 to 2022 were used as a database, and 150 petrochemical accident reports were collected and collated. The identification of petrochemical accident risk factors based on text mining mainly includes the following aspects: the formation of text mining corpus; select a text mining tool; text mining technology is used for text preprocessing of corpus; the petrochemicals accident risk factor word cloud map was drawn. In order to eliminate the interference of irrelevant phrases in text mining, this study selects three parts of the accident investigation report: “accident location”, “accident process” and “accident cause” to form a text mining corpus. The flow chart of text mining is shown in Figure 5.

Before conducting text mining, it is essential to choose the appropriate mining tools. The most widely used open-source text mining tools include Weka (3.8.6), LingPipe (3.9.2), ROSTCM (6.0), ICTCLAS (5.0), Python (3.1.2), and the R language (4.3.0). The Python language provides a completely open and free environment, providing appropriate program analysis packages tailored to different analytical needs. Its functionalities encompass data analysis, statistical analysis, and result visualization. Commonly used text processing libraries such as PDF2, NLTK, and Python-docx facilitate the processing of PDF files, natural language text files, and the creation or updating of.doc files. These libraries support tasks such as information extraction, file separation, file integration, and annotation, offering comprehensive functions. Therefore, in this paper, the Python programming language will be used to carry out the research process, with the help of the PyCharm compiler to write the relevant programs.

In practice, the text content of a petrochemical accident investigation report has no uniform format standard and language, which belongs to unstructured text. Punctuation marks and invalid words are easy to mine by text mining tools as separate word segmentation. In order to avoid the influence of punctuation marks and invalid words on the analysis results, it is necessary to clean the original data [43]. Data cleaning refers to providing high-quality data sources for text mining results by processing meaningless data and defective data.

The main criteria for data cleansing include completeness, accuracy, readability, consistency, and repeatability. Completeness includes all the information in the dataset and does not leave out key fields. Readability includes that the cleaned text should be easy to understand and avoid using overly complex or technical terms. Deleting duplicates means deleting duplicate records to ensure that each event occurs only once in the dataset. Additionally, it is crucial to correct spelling and grammatical errors, remove common words that lack real meaning, and concentrate on the key information.

The process of data cleaning and word frequency statistics in this study is as follows: (1) Word frequency statistics are carried out on petrochemical accident investigation reports to form preliminary statistical results; (2) Punctuation marks and invalid words are cleaned by 3 professionals, and 3 preliminary cleaning results are formed; Projects a-c represent the three Professional staff. (3) Combine the 3 preliminary cleaning results to form the final statistical result of word frequency, and sort according to the frequency of word occurrence and word frequency; (4) Make word clouds and extract high-frequency words. The partial results of the final word frequency statistics (only the top 10 words in the order) are shown in

Table 2. Word Frequency statistics (top 10 digits).

Serial Number	Projects	Words	Frequency	Word Frequency	Projects	Words	Frequency	Word Frequency	Projects	Words	Frequency	Word Frequency
1	a	Fire fighting	589	0.9349	b	Safety	491	0.8524	c	Chemical	537	0.8703
2	a	Safety	476	0.7556	b	Petroleum	425	0.7378	c	Production	390	0.6321
3	a	Chemical	405	0.6429	b	Worker	387	0.6719	c	Environment	378	0.6126
4	a	Sense	366	0.5810	b	Production	352	0.6111	c	Equipment	320	0.5186
5	a	Process	292	0.4635	b	Irregularity	316	0.5486	c	Corrosion	301	0.4878
6	a	Break the law	263	0.4175	b	Project	294	0.5104	c	Process	298	0.4830
7	a	Production	245	0.3889	b	Detection	250	0.4340	c	Safety	276	0.4473
8	a	Project	187	0.2968	b	Management	217	0.3767	c	Worker	254	0.4117
9	a	Inspect	165	0.2619	b	Equipment	203	0.3524	c	Faults	229	0.3712
10	a	Endanger	138	0.2190	b	Job	179	0.3108	c	Programs	165	0.2674

.

Based on the above data, linear regression analysis was performed on frequency and word frequency, with frequency as the independent variable and word frequency as the dependent variable. The analysis of the results of the F-test shows that significance is presented at the level, rejecting the original hypothesis that the regression coefficient is 0, so the model meets the requirements. For variable covariance performance, VIF (variance inflation factor) is all less than 10, so the model has no multicollinearity problem and the model is well constructed. The formula of the model is as follows: Word frequency = 0.005 + 0.002 ∗ Frequency.

Using hierarchical cluster analysis, short branching items that are close to each other in the graph indicate a high degree of similarity in some characteristics. For example, “Chemical”, “Fire Fighting”, and “Safety” form a cluster, suggesting they share significant similarities in certain aspects. In contrast, “project”, “process”, and “equipment” have longer branches, indicating greater differences among them. The hierarchical clustering diagram is shown in Figure 6.

The establishment of risk evaluation indicators is the key to the risk classification and hidden danger investigation system of petrochemical enterprises, and a clear indicator system helps petrochemical enterprises to rank risks, focus resources on high-risk areas, and improve the efficiency of safety management.

Based on the above research, through the analysis of the previous literature, for petrochemical enterprises, the “personnel” factors mainly include the safety awareness and knowledge of the workers, the implementation of three-level safety education, emergency rescue ability, and psychological and physiological conditions. The “equipment” factors mainly include the safety status of the production equipment, whether the instrument and equipment are inspected regularly, whether there is a large source of danger in the equipment, whether there is safety protection equipment, and the failure rate of the power supply equipment. The “environmental” factors mainly include the safety of the production environment, the presence of safety signage in the production workshop, and the suitability of natural geographical conditions. The “management” factors mainly include the level of risk monitoring in chemical production, whether the enterprise manager has formulated the responsibility system for safe production, and the frequency of employee safety education and training. The “material” factors mainly include the proper storage of hazardous chemicals, ensuring compliance with storage requirements, and identifying toxic or corrosive raw materials. The “process” factors mainly include the rationality and feasibility of the production process methods, operating procedures, and instrumentation operating procedures. Through text mining and analysis of 150 petrochemical accidents from 2013 to 2022, an evaluation system is built based on six primary indicators: “personnel”, “equipment”, “environment”, “management”, “material”, and “process”, as shown in Figure 7:

3.1.3. Evaluation Index Weight Calculation

(1) AHP (The analytic hierarchy process)

The analytic hierarchy process (AHP) should first stratify and refine the research object, so as to establish a multi-level analysis model, compare the importance of the two factors, establish an importance judgment matrix, and obtain the weight of each index. The index with the largest weight is regarded as the largest influence factor. The specific calculation steps are as follows:

In constructing the hierarchical model, while the affiliations among the elements at each level are clearly defined, the relative importance of these elements remains to be determined. Therefore, a hierarchical ordering of the elements must be conducted through pairwise comparisons. If Z factor is affected by a factor X = {x₁, ……, x_n}, select two factors x_i and x_j, b_ij indicates the ratio of the influence of x_i and x_j on Z, and then organize the results of all b_ij comparisons to construct the judgment matrix B:

$$\mathrm{B}=\left[\begin{array}{ccc}{\mathrm{b}}_{11}& {\mathrm{b}}_{12}& \begin{array}{cc}\cdots & {\mathrm{b}}_{1\mathrm{n}}\end{array}\\ \begin{array}{c}{\mathrm{b}}_{21}\\ ⋮\end{array}& \begin{array}{c}{\mathrm{b}}_{22}\\ ⋮\end{array}& \begin{array}{c}\begin{array}{cc}\cdots & {\mathrm{b}}_{2\mathrm{n}}\end{array}\\ \begin{array}{cc}\ddots & ⋮\end{array}\end{array}\\ {\mathrm{b}}_{\mathrm{n}1}& {\mathrm{b}}_{\mathrm{n}2}& \begin{array}{cc}\cdots & {\mathrm{b}}_{\mathrm{n}\mathrm{n}}\end{array}\end{array}\right]$$

The values of b were determined using 1 to 9 and its reciprocal method as a scale, as shown in

Table 3. Meaning of scales 1 to 9.

Scale bij	Hidden Meaning
1	Factor i is as important as factor j
3	Factor i is slightly more important than factor j
5	Factor i is significantly more important than factor j
7	Factor i is more strongly important than factor j
9	Factor i is extremely important compared to factor j
2, 4, 6, 8	Comparisons of i and j fall between the above levels of comparison
inverse number	If the ratio of the importance of factors i and j is bij, then the ratio of factors j to i is 1/bij

.

In order to reduce the difficulty of comparing factors of different natures and improve the accuracy of the final importance, the factor pairwise comparison method is used in constructing the judgment matrix according to the scale.

The judgment matrix is normalized, and each row of the matrix after normalization is added:

$${\overline{W}}_{\mathrm{i}}={\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{b}_{ij}}(i,j=1,2,\cdots n)$$

(1)

The vector added to each row is normalized:

$$W_{\mathrm{i}}={\displaystyle \frac{W_{\mathrm{i}}}{{\displaystyle \sum _{i=1}^n{\overline{W}}_{\mathrm{i}}}}}(i,j=1,2,\cdots n)$$

(2)

The normalization process yields the feature vector W = (W₁, W₂, …, W_m)T

Once the feature vector that determines the largest feature root of the array is determined, it is normalized and denoted as W, and whether it can use the analytic hierarchy process needs to be tested for consistency.

Eigenvalue calculation formula:

$${\mathsf{\lambda }}_{\max }={\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}\frac{{\left(\mathrm{BW}\right)}_{\mathrm{i}}}{{\mathrm{nW}}_{\mathrm{i}}}}$$

(3)

where ${\lambda }_{\max }$ is the maximum characteristic root; $\mathrm{W}$ is the eigenvector of maximum characteristic root; $\mathrm{B}$ is the judgment matrix.

Consistency test formula:

$$CI={\displaystyle \frac{{\lambda }_{\max }-n}{n-1}}$$

(4)

where $CI$ is consistency index; ${\lambda }_{\max }$ is the maximum characteristic root; $n$ is unique non-zero characteristic root.

In the formula, CI is used for the consistency test, and the CI value is inversely proportional to consistency.

To measure the size of CI, the random consistency index RI is introduced:

$$RI={\displaystyle \frac{C{I}_1+C{I}_2+\cdots \cdots +C{I}_n}{n}}$$

(5)

CI and RI are compared to test whether the consistency of the judgment matrix is qualified, and the test coefficient CR is obtained. The formula is as follows:

$$CR={\displaystyle \frac{CI}{RI}}$$

(6)

If RI < 0.1, it can be determined that the inconsistency of the judgment matrix is acceptable and the influence weight on the calculated weight is small. Instead, the judgment matrix is modified until the value of RI meets the previous criteria.

After calculating the weight of the index, the overall hierarchical ranking of the index is carried out, and the consistency test is carried out. The method is the same as above.

(2) Variable weight theory

The Analytic Hierarchy Process method assigns indicator weights based on subjective judgments that rely on the expertise of professionals. This subjectivity may lead to evaluation outcomes that do not accurately reflect reality. Wang Peizhuang [44] proposes in his idea of variable weight that if he wants to eliminate the possible deviation between the constant weight and the actual, the weight of the indicators should be adjusted with the state value of the indicators and be adjusted appropriately, so that the most reasonable weight can be calculated. By incorporating a dynamic weight adjustment mechanism, the variable weight theory can be flexibly adjusted according to different situations or the subjective judgment of decision-makers, thereby reducing biases associated with subjective assessments in the AHP method.

The variable weight theory is that compared with the constant weight, when the index system has a very high or very low value, it can directly affect the determination of the index weight, resulting in a deviation between the calculated weight and the actual weight. Therefore, this paper introduces the variable weight theory to adjust the index weight dynamically to improve the rationality of the weight.

Variable weight vector $\mathrm{W}\left(\mathrm{X}\right)=\left[{\mathrm{W}}_1\left(\mathrm{X}\right),{\mathrm{W}}_2\left(\mathrm{X}\right)\cdots {\mathrm{W}}_{\mathrm{m}}\left(\mathrm{X}\right)\right]$. First, an m-dimensional punitive state variable weight vector is defined as a map $\mathrm{S}:\left[\mathrm{0,1}\right]\mathrm{m}\to \left[\mathrm{0,1}\right]\mathrm{m},\mathrm{X}\to \mathrm{S}\left(\mathrm{X}\right)=S_1\left(\mathrm{X}\right),\dots ,S_m\left(\mathrm{X}\right)]$, satisfy: (1) $X_i\ge X_j\Rightarrow S_i\left(\mathrm{X}\right)\le S_j\left(\mathrm{X}\right)$; (2) S_j(X) for each argument continuously $\mathrm{j}=\mathrm{1,2},\dots ,\mathrm{m}$; (3) For the constant weight variable W, normalization, monotonicity, and continuity are satisfied.

$$\mathrm{W}(\mathrm{X})={\displaystyle \frac{\mathrm{W}\cdot \mathrm{S}(\mathrm{X})}{{\mathsf{\Sigma }}_{\mathrm{j}=1}^{\mathrm{m}}[{\mathrm{W}}_{\mathrm{j}}\cdot {\mathrm{S}}_{\mathrm{j}}(\mathrm{X})]}}$$

(7)

The comprehensive weight matrix of the index is obtained by calculation: $\mathrm{W}={({\mathrm{W}}_{11},{\mathrm{W}}_{12}\cdots {\mathrm{W}}_{\mathrm{i}\mathrm{j}})}^{\mathrm{T}}$.

3.1.4. Risk Assessment and Risk Classification

Traditional risk assessment usually includes two dimensions: accident severity and accident possibility. Based on traditional two-dimensional risk assessment, the three-dimensional risk assessment model is based on the complexity of production accidents in petrochemical enterprises, and the sensitivity of accidents is different in different time and space. Sensitivity factors are introduced in the classification of risk assessment. Hazard source risk assessment was carried out from three dimensions: possibility (abcd in turn), seriousness (ABCD in turn), and sensibility (1234 in turn).

According to the above established risk index system of petrochemical enterprises, from the relationship between influencing factors (i.e., people, equipment, environment, management, materials and process factors, etc.) and different time spaces, based on the two-dimensional evaluation model of risk function, sensitivity factors were introduced to establish a three-dimensional risk classification model, as shown in Figure 8:

According to the classification of a two-dimensional matrix, the possibility, seriousness, and sensibility of risks are divided into five levels. The level set is set in

Table 4. Security risk level set.

Risk Evaluation Dimensions	Hierarchy	Evaluation Content	Grade Range	Median Grade
possibility	Class I	unlikely	[0, 1]	0.5
	Class II	seldom	(1, 2]	1.5
	Class III	infrequent	(2, 3]	2.5
	Class IV	non-recurrent	(3, 4]	3.5
	Class V	frequent	(4, 5]	4.5
seriousness	Class I	negligible	[0, 1]	0.5
	Class II	minute	(1, 2]	1.5
	Class III	usual	(2, 3]	2.5
	Class IV	severity	(3, 4]	3.5
	Class V	crux	(4, 5]	4.5
sensibility	Class I	insensitive	[0, 1]	0.5
	Class II	slightly sensitive	(1, 2]	1.5
	Class III	sensitivities	(2, 3]	2.5
	Class IV	more sensitive	(3, 4]	3.5
	Class V	Extremely sensitive	(4, 5]	4.5

:

n security experts evaluate the indicators based on the possibility, seriousness, and sensibility from five levels, then the degree of affiliation of the indicator X_ij corresponding to the level I indicator is Rp_ijI = NpI/n, where NpI is the number of experts who evaluated the likelihood of the indicator to be level I. Therefore, the likelihood affiliation vector of the indicator is Rp_ij = (Rp_ijI, Rp_ijII, Rp_ijIII, Rp_ijIV, Rp_ijV); the severity affiliation vector is RL_ij = (RL_ijI, RL_ijII, RL_ijIII, RL_ijIV, RL_ijV); and the sensitivity affiliation vector is Rs_ij = (Rs_ijI. Rs_ijII, Rs_ijIII, Rs_ijIV, Rs_ijV). The possibility, seriousness, and sensibility affiliation values are: ${\mathrm{r}}_{\mathrm{p}\mathrm{i}\mathrm{j}}={\mathrm{R}}_{\mathrm{p}\mathrm{i}\mathrm{j}}\times {\mathrm{M}}_{\mathrm{p}}^{\mathrm{T}}$; ${\mathrm{r}}_{\mathrm{L}\mathrm{i}\mathrm{j}}={\mathrm{R}}_{\mathrm{L}\mathrm{i}\mathrm{j}}\times {\mathrm{M}}_{\mathrm{L}}^{\mathrm{T}}$; ${\mathrm{r}}_{\mathrm{s}\mathrm{i}\mathrm{j}}={\mathrm{R}}_{\mathrm{s}\mathrm{i}\mathrm{j}}\times {\mathrm{M}}_{\mathrm{s}}^{\mathrm{T}}$.

Based on the above calculations, the affiliation matrix of the variable X_i is obtained:

$$r_{pi}=\left(\begin{array}{c}{r}_{pi1}\\ r_{pi2}\\ \dots \\ r_{pij}\end{array}\right) r_{Li}=\left(\begin{array}{c}{r}_{Li1}\\ r_{Li2}\\ \dots \\ r_{Lij}\end{array}\right) r_{si}=\left(\begin{array}{c}{r}_{si1}\\ r_{si2}\\ \dots \\ r_{sij}\end{array}\right)$$

(8)

Determine the three-dimensional risk weight as W = (W_p, W_L, W_s).

Based on the two-dimensional risk alert calculation, the formula for calculating the three-dimensional risk alert for a single indicator is:

$$r_{ij}=r_{pLs}=\sqrt{w_p{r}_{pij}^2+w_L{r}_{Lij}^2+w_s{r}_{sij}^2}$$

(9)

The combined risk alert is calculated as follows:

$$\mathrm{r}={\mathrm{W}}^{\mathrm{T}}\times {\mathrm{r}}_{\mathrm{i}j}=\left(\begin{array}{c}{\mathrm{W}}_{11}^{\prime },{\mathrm{W}}_{12}^{\prime }\cdots {\mathrm{W}}_{\mathrm{I}j}^{\prime }\end{array}\right)\times \left(\begin{array}{c}{\mathrm{r}}_{11}\\ {\mathrm{r}}_{12}\\ \dots \\ {\mathrm{r}}_{\mathrm{i}j}\end{array}\right)$$

(10)

According to the above formula, the single risk value and comprehensive risk value of 23 indicators were calculated, and the risk alarm level was calculated using the optimal segmentation method. The risk alarm level was divided into four levels: no alarm, light alarm, medium alarm, and heavy alarm, and the warning threshold ranges were [0, 1.87], (1.87, 2.94], (2.94, 3.79], (3.79, +∞]. The comprehensive risk alarm level is divided into four levels, and the threshold is set as follows: Level I r at [0, 2.21] indicates normal state; Level II r at (2.21, 2.83) indicates yellow light alarm status; Level III r is (2.83, 3.64), indicating orange medium alarm state; and Level IV r at (3.64, 5) indicates the red heavy alarm state.

3.1.5. Risk Management and Control

Risk control measures mainly include management measures, training and education measures, individual protection measures, engineering measures, and emergency measures.

(1) Management measures

The enterprise must establish a safety product management system that meets the requirements of safety product laws, regulations, and standards. Petrochemical enterprises should develop a “big, fast, and strict” self-inspection program for their own safety products, inspect the whole process of safety management and all process links of the product site, and organize workshops and teams to strictly investigate hidden dangers of accidents.

(2) Training and education measures

Strengthen the safety training of employees, mainly including new employees, old employees, and management personnel. Professional technicians must be equipped with appropriate professional certificates. Employee training is mainly for new employees as well as the three-level training for employees who return to work, and to ensure the length of training and the relevance of the training content.

(3) Individual protection measures

The production process of petrochemical enterprises will involve a lot of toxic chemical raw materials, which requires operators to wear the appropriate protective equipment in the production process. In the case of a complete set of protective equipment, operators must be urged to wear it correctly and in a timely manner.

(4) Engineering measures

Warning signs must be installed for toxic and hazardous jobs, and special signs must be installed for many special jobs. In the production of petrochemical enterprises, the supervision of technical measures is based on self-control by the technicians of the petrochemical enterprises on the one hand, and on the other hand, on the safety assessment of the petrochemical company, which is carried out by external specialists, in order to ensure the safety of the plant and equipment.

(5) Emergency measures

In the production of petrochemical enterprises, regular emergency drills are conducted, and the loopholes that existed during the drills are revised and supplemented in the emergency plan. Provide complete emergency equipment and facilities to ensure their integrity. Regular inspections and maintenance are performed to ensure the integrity of emergency equipment.

3.2. Design of Hidden Danger Investigation and Management System

3.2.1. Hidden Danger Investigation

(1) Type of investigation

The types of investigation mainly include: employee post level, team level daily, workshop level weekly or monthly, and company level type of hidden danger investigation.

(2) Investigation requirements

Regardless of any level of hidden danger, the investigation must be comprehensive, responsible to the person, and must ensure that all employees of the enterprise are involved in the investigation of hidden dangers, according to the investigation requirements of the regular investigation of hidden dangers.

(3) Investigation cycle

Each petrochemical enterprise type is different; there are different production workshops, different types of hidden dangers to conduct targeted investigation, including investigation cycle or frequency, that is, once a day, once a week, once a month, once a quarter, and once a year.

(4) Determine the investigation items

Prior to the implementation of a potential hazard study, the organization should select items from the list of potential hazards according to the type of potential hazard study materials, the number of employees, and the type of production process. Hidden danger investigation can be divided into on-site hidden danger investigation and basic management hidden danger investigation.

(5) Check result records

Organize the results of hidden danger inspection into a hidden danger inspection list. The hidden danger inspection list of the production site category mainly refers to the hidden danger inspection of the catalytic workshop with the facilities, equipment, and operation activities as the identification unit, and the contents of the hidden danger list include the name of equipment and facilities, operation name, inspection content, inspection standard, and so on. The list of hidden dangers in the basic management category mainly includes the name of hidden dangers, the contents of the inspection, the inspection standard, and the inspection cycle and type.

3.2.2. Classification of Hidden Danger

The systematic investigation and scientific grading of hidden danger is the premise of hidden danger management, and the consequences caused by accidents have the characteristics of chain type, so this paper puts forward the concept of the rate of change of hidden danger protection barrier 0, which is expressed by ΔG, and the quantitative calculation of ΔG is shown in Equation (9).

$$\mathsf{\Delta }G={\displaystyle \frac{Y_i}{G+Y_i}}$$

(11)

ΔG—denotes the rate of change of the protective barrier against hidden hazards;

Y_i—indicates that the i risk control measure fails;

G—represents the utility of risk control measures.

In the three-dimensional hidden danger classification model, the x-axis indicates the degree of difficulty in the rectification of hidden danger; the y-axis indicates the severity of hidden danger that may lead to accidents; and the z-axis changes in the rate of change of hidden danger. Protective barriers also directly affect the results of the classification of hidden danger and the determination of the effectiveness of hidden danger barriers firstly determines the effectiveness of the current risk management and control measures and then conducts an examination of the protective barriers of the various levels of hidden danger, determines the point of damage to the barriers and the damage degree of the system caused by this point, and finally determines the degree of barrier failure according to the determination criteria. Then, for each level of protective barriers of the hidden danger, the damage points of the barriers and the degree of system damage caused by the points are identified, and finally, the degree of barrier failure is determined according to the determination criteria. In this paper, the level of ΔG is classified into four levels, and the specific determination criteria are shown in

Table 5. Criteria for determining the rate of change of hidden protective barriers ΔG.

Rating	ΔG	Criteria for Determination
Z1	<0.1	No irregularities directly related to the production process
Z2	0.1–0.3	Minor damage to process barriers involved in operations
Z3	0.3–0.5	Moderate damage to the barrier function of the crypto
Z4	>0.5	Failure of more than half of the safety barrier functions, with a high likelihood of hazards evolving into accidents

, and the guidelines for determining the severity of the consequences of the accident and the difficulty of treatment are shown in

Table 6. Criteria for determining the severity of the consequences of accidents and the difficulty of treatment.

Possible Consequences of Accidents	Rating	Difficulty of Governance	Rating
Potential for large numbers of casualties or very large economic losses	y4	Large	x4
May cause injury, death or substantial economic loss	y3	Comparatively large	x3
May cause personal injury or general economic loss	y2	Normal	x2
Causing small economic losses	y1	Comparatively Small	x1

.

This paper divides hidden dangers into general hidden dangers and major hidden dangers. According to the three-dimensional classification model, general hidden dangers are divided into three levels: A, B, and C, as shown in Figure 3. The red area represents major hidden dangers, and the white part represents general hidden dangers. See

Table 7. Hidden danger grading coordinate.

Level of Danger	Region
Major hazard	(x4, y4, z4)
General Hazard Level A	(x4, y3, z4) (x4, y4, z3) (x4, y3, z3) (x3, y4, z4) (x3, y3, z4) (x3, y4, z3) (x3, y3, z3)
General Hazard Level B	(x4, y2, z4) (x3, y2, z4) (x2, y2, z4) (x2, y3, z4) (x2, y4, z4) (x4, y2, z3) (x3, y2, z3) (x2, y2, z3) (x2, y3, z3) (x2, y4, z3) (x4, y2, z2) (x3, y2, z2) (x2, y2, z2) (x2, y3, z2) (x2, y4, z2) (x4, y3, z2) (x4, y4, z2) (x3, y4, z2) (x3, y3, z2)
General Hazard Level C	Else

for a detailed classification of hidden dangers.

3.2.3. Closed-Loop Control of Hidden Dangers

In view of the hidden dangers of production safety, the hidden dangers control levels of petrochemical enterprises are divided into company level, factory level, workshop level, and team level, in which the company level is responsible for the investigation and control of major hidden dangers of enterprises; the factory level is responsible for general A-level hidden danger investigation and control; the workshop level is responsible for general B-level hidden danger investigation and control; and the team level is responsible for general C-level hidden danger investigation and control. For the investigation and management of hidden dangers, PDCA closed-loop management mode is adopted, and the main governance process is shown in Figure 9.

4. The Bidirectional Cooperative Relationship between Risk Hierarchical Control and Hidden Danger Investigation

In the collaborative risk and hidden danger management technology system of petrochemical enterprises, it mainly includes two barriers. The first barrier is the risk classification control system, and the second barrier is the hidden danger investigation and management system. The better the risk classification control work is done, which means that the better the hidden danger investigation work is done, the less the possibility of accidents. Hidden danger investigation and management is the re-treatment of the hazard source that failed to control the first barrier. In this stage, new hazards will also be identified and then fed back to the risk control system to control the new hazards. The relationship between the risk hierarchical management and control system and the hidden danger investigation and management system is shown in Figure 10.

According to the above research and analysis, based on the characteristics of petrochemical enterprises, combined with “11245” innovative safety management mode. The technical system flow chart of collaborative management of risks and hidden dangers in petrochemical enterprises was prepared, as shown in Figure 11.

“1” stands for one goal, namely, the prevention of production safety accidents.

“1” represents a set of systems, the dual management system of risk hierarchical control and hidden danger investigation and management.

“2” represents two lines of defense, namely, risk hierarchical control and hidden danger investigation and management.

“4” represents four principles, risk priority, systematic, full participation, and continuous improvement.

“5” means five aspects of implementation, namely the implementation of the main responsibility, the implementation of measures, the implementation of funds, the implementation of time, and the implementation of information.

5. Analysis of Simulation Results

5.1. Results of Risk Grading Assessment

The risk control and hidden danger management system of this study has been validated in eight petrochemical enterprises in China and achieved significant results. The specifics of the petrochemical companies are as follows: Sinopec Qingdao Petrochemical Refining and Chemical Company Limited (Qingdao, Shandong, China), Shandong Kaitai Petrochemical Company Limited (Zibo, Shandong, China), Shandong Huifeng Petrochemical Group Company Limited (Zibo, Shandong, China), Shandong Jincheng Petrochemical Group Company Limited (Zibo, Shandong, China), Shenghong Petrochemical Group Company Limited (Lianyungang, Jiangsu, China), Shandong Dongyue Group Chemical Company Limited (Zibo, Shandong, China), Shanxi Yongdong Chemical Company Limited (Yuncheng, Shanxi, China), and Shandong Shouguang Luqing Petrochemical Enterprises (Weifang, Shandong, China).

Shandong Huifeng Petrochemical Group Co., Ltd. (Zibo, Shandong, China). was founded in 1997 with more than 2000 employees. It is a large-scale modernized enterprise group integrating petrochemicals, new energy, new materials, heat supply, logistics services, retail terminals, and international trade, and the company owns 39 sets of various types of production units and 10 special lines for the railroad transportation of hazardous chemicals, which has a good risk management level at the good level. Sinopec Qingdao Refining and Chemical Co., Ltd. (Qingdao, Shandong, China). was established in 2004, with an annual production capacity of more than 8 million tons of gas, coal, and diesel oil products and more than 2 million tons of various types of petrochemical products such as liquefied petroleum gas, petroleum coke, polypropylene, styrene, mixed benzene, sulfur yellow, etc. The company’s chemical production scored low on the risk detection indicators. Given the more problems exposed by Shandong Shouguang Luqing Petrochemical Enterprise, this paper will focus on verifying the specific application of Shandong Shouguang Luqing Petrochemical Enterprise.

Shandong Shouguang Luqing Petrochemical Enterprise was established in August 2000, is a joint-stock private enterprise, covers an area of 5000 acres, is located in Shandong Weifang Shouguang Bohai Industrial Park, and has a total of more than two thousand employees, mainly in the production of diesel, polypropylene products, and gasoline. This paper primarily evaluates the safety production status of this petrochemical enterprise from the quantitative model. To conduct this evaluation, 150 safety technicians from the enterprise were randomly selected and divided into five groups of 30 individuals each. Each group consisted of the same personnel makeup: 10 safety section personnel, 5 middle management staff, 5 production workshop shift managers, and 10 first-line technicians from the production workshops. A scoring table containing selected safety indices was distributed to all 150 safety management and technical personnel within the facility. The evaluation experts are chosen based on the management personnel in charge of plant safety and technicians familiar with plant operations, and the risk evaluation judgment matrix is constructed based on the scoring of the evaluation experts. According to Formulas (1)~(7), the subjective weights and comprehensive weights of the indicators are calculated, and the calculation results are shown in

Table 8. Weights of evaluation indicators.

Target Layer	Criterion Layer	Weight Wi	Index Layer	Subjective Weight Wj	Comprehensive Weight Wij
Three-dimensional risk evaluation of Luqing petrochemical enterprise	Personnel factor X1	0.2051	X11	0.2437	0.0500
			X12	0.1865	0.0381
			X13	0.1653	0.0340
			X14	0.2011	0.0412
			X15	0.2034	0.0416
	Equipment factor X2	0.1422	X21	0.1789	0.0254
			X22	0.2250	0.0320
			X23	0.1978	0.0280
			X24	0.2231	0.0316
			X25	0.1752	0.0250
	Environmental factor X3	0.1267	X31	0.3619	0.0460
			X32	0.2972	0.0376
			X33	0.3409	0.0432
	Management factor X4	0.1553	X41	0.2215	0.0342
			X42	0.2079	0.0323
			X43	0.2613	0.0405
			X44	0.3093	0.0478
	Material factor X5	0.1734	X51	0.3329	0.0582
			X52	0.2968	0.0514
			X53	0.3703	0.0641
	Process factor X6	0.1983	X61	0.2981	0.0590
			X62	0.3320	0.0656
			X63	0.3699	0.0732

.

The three-dimensional risk evaluation matrix for chemical companies is:

$$\mathrm{X}=\left[\begin{array}{cccc}1& 2& 3& 4\\ 1/2& 1& 2& 4\\ 1/3& 1/2& 1& 3\\ 1/4& 1/4& 1/3& 1\end{array}\right]$$

The personnel factor judgment matrix is:

$${\mathrm{X}}_1=\left[\begin{array}{ccccc}1& 2& 2& 3& 4\\ 1/2& 1& 2& 2& 3\\ 1/2& 1/2& 1& 2& 3\\ 1/3& 1/2& 1/2& 1& 3\\ 1/4& 1/3& 1/3& 1/3& 1\end{array}\right]$$

The equipment factor judgment matrix is:

$${\mathrm{X}}_2=\left[\begin{array}{ccccc}1& 2& 3& 2& 4\\ 1/2& 1& 2& 3& 2\\ 1/3& 1/2& 1& 2& 3\\ 1/2& 1/3& 1/2& 1& 2\\ 1/4& 1/2& 1/3& 1/2& 1\end{array}\right]$$

The environmental factor judgment matrix is:

$${\mathrm{X}}_3=\left[\begin{array}{ccc}1& 3& 2\\ 1/3& 1& 2\\ 1/2& 1/2& 1\end{array}\right]$$

The management factor judgment matrix is:

$${\mathrm{X}}_4=\left[\begin{array}{cccc}1& 2& 3& 4\\ 1/2& 1& 2& 4\\ 1/3& 1/2& 1& 3\\ 1/4& 1/4& 1/3& 1\end{array}\right]$$

The material factor judgment matrix is:

$${\mathrm{X}}_5=\left[\begin{array}{ccc}1& 2& 3\\ 1/2& 1& 2\\ 1/3& 1/2& 1\end{array}\right]$$

The process judgment matrix is:

$${\mathrm{X}}_6=\left[\begin{array}{ccc}1& 3& 2\\ 1/3& 1& 2\\ 1/2& 1/2& 1\end{array}\right]$$

Weighting refers to the degree of importance of each indicator relative to the overall decision in a multi-indicator decision. Comprehensive weighting reduces the influence of subjective judgment to a certain extent through the introduction of variable weighting theory. The allocation of comprehensive weights reflects the influence of different indicators in the final evaluation or decision-making. Comprehensive weights can help enterprises prioritize each risk indicator, and after understanding the main risk indicators, enterprises can formulate targeted strategies and focus resources on high-risk areas. For example, the combined weight of X₆₃ (accuracy of process equipment operation) reaches 0.0732, which is at a high level, so enterprises need to pay attention to losses caused by errors in the operation of instruments and equipment.

From the above analysis, it can be seen that the personnel factor accounts for the highest percentage in each guideline layer of the risk hierarchical control system, with 20.51% of personnel factors, 19.83% of process factors, 17.34% of material factors, 15.53% of management factors, 14.22% of equipment factors, and 12.67% of environmental factors.

According to the score and the established judgment matrix, the comprehensive weight of each index is calculated, and the CR is less than 0.1. It is concluded that the weight of the three dimensions of risk is W = (0.243, 0.415, 0.342), and the importance of the three dimensions is ranked as seriousness, sensibility, and possibility. A consistency test was performed on the matrix, CR = 0.0072 < 0.1, passing the consistency test. Security experts evaluate risk assessment indicators according to possibility, seriousness, and sensibility, get the membership vector matrix of indicator X_i, and then calculate the membership value. The risk alarm of a single index and the comprehensive risk alarm are calculated, respectively, and the results are shown in

Table 9. Calculation result of single index risk alarm.

Target Layer	Criterion Layer	Index Layer	Possibility	Seriousness	Sensibility	Risk Alertness	Risk Level	Risk Alarm Level
Three-dimensional risk evaluation of Luqing petrochemical enterprise	Personnel factor X1	X11	2.25	2.10	1.65	2.36	II	light alarm
		X12	2.10	2.35	2.75	2.29	II	light alarm
		X13	2.35	1.80	2.60	2.27	II	light alarm
		X14	1.95	2.85	2.25	2.98	III	medium alarm
		X15	2.05	2.90	1.95	2.76	II	light alarm
	Equipment factor X2	X21	1.80	2.10	2.05	2.34	II	light alarm
		X22	2.15	1.80	2.10	3.01	III	medium alarm
		X23	1.65	2.35	1.65	2.42	II	light alarm
		X24	2.75	2.55	2.75	2.71	II	light alarm
		X25	2.60	2.15	2.60	2.99	III	medium alarm
	Environmental factor X3	X31	2.25	1.65	2.25	2.64	II	light alarm
		X32	1.95	2.75	2.75	2.48	II	light alarm
		X33	2.05	2.60	2.60	2.88	III	medium alarm
	Management factor X4	X41	2.10	2.25	2.25	2.29	II	light alarm
		X42	2.35	1.95	1.95	3.14	III	medium alarm
		X43	1.80	2.05	2.05	2.56	II	light alarm
		X44	2.85	2.10	2.10	2.97	III	medium alarm
	Material factor X5	X51	2.90	2.35	2.35	2.58	II	light alarm
		X52	2.10	1.80	1.80	2.71	II	light alarm
		X53	1.80	2.15	2.75	2.25	II	light alarm
	Process factor X6	X61	2.35	2.35	2.10	2.89	III	medium alarm
		X62	2.55	1.95	1.80	2.54	II	light alarm
		X63	2.05	2.05	2.35	2.67	II	light alarm

. The calculation result of the comprehensive risk alarm is as follows: the risk level is level II, and the yellow light alarm signal is issued. The evaluation result is consistent with the actual risk level of Luqing petrochemical enterprise.

5.2. Feasibility Verification of Risk Hierarchical Control System and Hidden Danger Investigation and Management System

This paper sorted out the work of risk classification control and hidden danger investigation and management of existing petrochemical enterprises, issued 150 risk classification control and hidden danger investigation rating tables, and recovered 136 with a recovery rate of 90.7%. The main participants were 23 personnel of the safety department of Luqing Petrochemical Enterprise, 47 middle managers, and 28 workshop group monitors. There are 52 front-line staff in the catalytic workshop, material transport workshop, polypropylene production workshop, maintenance workshop, and other workshops.

(1) Reliability test

(1) To calculate the reliability of half, the reliability coefficients of the two halves of a questionnaire are calculated separately. When two half reliability coefficients are the same, Spearman–Brown Formula (12) is used to calculate the reliability coefficient of the whole questionnaire. When the two half-reliability coefficients are different, Rulon Formula (13) or Flanagan Formula (14) are used for calculation.

$$R_{SB}={\displaystyle \frac{2r_{OB}}{1+r_{OB}}}$$

(12)

where $r_{OB}$ is the correlation coefficient of the two tests; $R_{SB}$ is the estimated or revised reliability of the questionnaire.

$$R_R=1-{\displaystyle \frac{S_{\overline{a}-\overline{b}}^2}{S_t^2}}$$

(13)

where $S_{\overline{a}-\overline{b}}^2$ is the variance of the difference between the scores of the two halves; $S_t^2$ is the total score variance.

$$R_F=2(1-{\displaystyle \frac{S_a^2-S_b^2}{S_t^2}})$$

(14)

where $S_a^2$ is the variance of both halves; $S_b^2$ is the variance of both halves; $S_t^2$ is the total score variance.

(2) KR20 formula: applies to survey data with scores of 0 and 1. (15) and (16) calculations can be used to apply binary data.

$$rKR20={\displaystyle \frac{K}{K-1}}[1-{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^{k}piqi}}{S_t^2}}]$$

(15)

$$rKR21={\displaystyle \frac{K}{K-1}}[1-{\displaystyle \frac{\overline{X}(K-\overline{X})}{K{S}_t^2}}]$$

(16)

where K is the number of questionnaire items; $S_t^2$ is the total score variance; $\overline{X}$ is the average of the total score of the questionnaire; $pi$ is the percentage of people who answered zero on question $i$; $qi$ is the percentage of people who answered 1 in question $i$.

(3) Cronbach’s α: It is applicable to both binary data and non-binary data and can be calculated by Equation (17).

$$\mathrm{\alpha }={\displaystyle \frac{\mathrm{K}}{\mathrm{K}-1}}[1-{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^{\mathrm{k}}{\mathrm{S}}_i^2}}{{\mathrm{S}}_{\mathrm{t}}^2}}]$$

(17)

where K is the number of problems measured; ${\mathrm{S}}_i^2$ is the variance of the score for question $i$; ${\mathrm{S}}_{\mathrm{t}}^2$ is the variance of the total score of the test.

(2) Validity test

In order to verify the structural validity of the questionnaire, the KMO test and Bartlett’s ball test were used to test the correlation of variables. In order to accurately test the relative value of the simple correlation coefficient between the original two variables and the coefficient of other partial correlation variables and the size of the error, Kaiser, Meyer, and Olkin proposed the sampling fitness test, also known as the KMO test. Check against the KMO metrics given by Kaiser.

In order to test the independence of variables, Bartlett’s spherical test is carried out. If common factors cannot be extracted for analysis, it means that variables are independent of each other. If the calculated p-value is less than 0.05, it can be analyzed. KMO and Bartlett’s spherical tests were conducted on first-level indicator variables, respectively, and each reliability test was greater than 0.8, indicating that the reliability quality of all research results was good, and the Cronbach α coefficient was 0.927. Therefore, the reliability test of this questionnaire passed. All variables, KMO = 0.746 > 0.7 and p value less than 0.05, meet the suitability test criteria. The questionnaire has structural validity, so the questionnaire results can be used for analysis.

SPSS (17.0) software was utilized to perform regression analysis on the scores of 136 experts regarding the double system, assessing the correlation between the variables. The risk and hidden danger data were entered into SPSS, as shown in Figure 12.

Machine learning techniques are applied to predict the risk values and hidden dangers based on the available values. A smaller relative error value indicates a better fit, with values generally below 20% considered acceptable. The average relative error of the risk model is 5.975% and the average relative error of the hidden danger model is 5.097%, which means that the model fits well. Specific prediction plots are shown in

Table 10. Risk and Hazard Data Prediction Chart.

Predicted Number of Steps	Risk	Hidden Trouble
1	84.134	83.192
2	84.362	83.428
3	84.590	83.663
4	84.818	83.899
5	85.047	84.136
6	85.275	84.372
7	85.504	84.609

.

The score of risk hierarchical control was set as the X-axis, and the score of hidden danger investigation and management was set as the Y-axis. After software regression analysis, the scatter diagram of the dual system was obtained, as shown in Figure 11. As can be seen from Figure 13, there is a positive correlation between the two in the range of 80 to 93 points. In order to ensure the statistical accuracy of the data, we chose a single value of a 95% confidence interval.

SPSS software was used for regression analysis of the two. The purpose of this analysis was to explore the impact of the quality of risk hierarchical control on the investigation and management of hidden dangers. Therefore, the score of risk hierarchical control was taken as the independent variable, and the score of hidden dangers investigation and management was taken as the dependent variable for regression analysis.

Only when the significance value is less than or equal to 0.05 is it statistically significant, indicating that the model can be used. The significance of the model established in this paper is equal to 0.000, and the significance value is less than 0.05, which indicates that the model is a statistical model and has research significance. Residual normality was used to test the regression model, and the check results were shown in Figure 14.

As can be seen from Figure 14, both sides of the curve in the figure are not completely symmetric, and there is a certain deviation. As can be seen from Figure 15, the distribution of scatter points is relatively scattered, indicating that the results of the normal distribution do not reach the most perfect state. In the real state, there is rarely an ideal normal distribution, as long as it is close to the normal distribution, it can accept its statistical significance. It can be seen that there is a positive correlation between risk hierarchical control and hidden danger investigation and management. In other words, the better the risk hierarchical control work is done, the hidden danger investigation and management work will also be improved accordingly.

6. Discussion and Conclusions

In order to improve the safety management level of petrochemical enterprises, combining the theoretical knowledge of safety system engineering, safety management, and applied statistical analysis, a technical policy system of risk classification and control and hidden danger investigation and management synergistic management for petrochemical enterprises is proposed. The main conclusions are as follows:

(1)

Based on RBS theory, the risk hierarchical management and control system of petrochemical enterprises is designed. Based on the text mining technology, 150 petrochemical account reports were collected and sorted out, and the risk assessment index system of petrochemical enterprises was constructed, which was divided into 6 first-level indicators and 28 second-level indicators. The weight of each index is calculated by AHP and variable weight theory, and the three-dimensional risk assessment method is used to evaluate the risk of petrochemical enterprises. The optimal segmentation method was used to calculate the risk alarm level, and the single indicator risk alarm level was divided into four levels: I (no alarm), II (light alarm), III (medium alarm), and IV (heavy alarm). The warning threshold ranges were [0,1.87], (1.87, 2.94], (2.94, 3.79], (3.79, +∞]. The comprehensive risk threshold is as follows: Level I [0,2.21], indicating the normal state. Level II (2.21, 2.83), indicates yellow light alarm status; Level III (2.83, 3.64), indicates orange medium alarm state; Level IV (3.64, 5) indicates red heavy alarm status.

(2)

Based on the theory of three-dimensional classification of hidden dangers and the principles of the PDCA cycle, the hidden danger investigation and management system of petrochemical enterprises is innovatively proposed. For the production safety hazards, the petrochemical hazard control level is divided into company level, factory level, workshop level, and team level. The company level is responsible for the investigation and management of major hidden dangers in the enterprise; the factory level is responsible for the investigation and management of general A-level hidden dangers; the workshop level is responsible for the investigation and management of general B-level hidden dangers; and the team level is responsible for the investigation and management of general C-level hidden dangers.

(3)

According to the characteristics of petrochemical enterprises, combined with the “11245” innovative safety management model, a technical policy system for the collaborative management of risk classification and control and hidden trouble detection and management in petrochemical enterprises has been proposed.

(4)

To more scientifically study the petrochemical enterprise risk classification control and hidden danger investigation management synergistic management technology policy system, Shouguang City, Shandong Province, Luqing petrochemical enterprises as an example to carry out empirical analysis, the comprehensive risk level of the second level, the enterprise, and the issuance of a yellow light alarm signal. The evaluation results are consistent with the actual risk level of Luqing Petrochemical Enterprise. 150 graded risk grading control and hidden danger investigation rating tables were issued, and the positive correlation between graded risk control and hidden danger investigation management was verified based on the theory of principal component analysis.

(5)

Through the implementation of a risk and hidden danger collaborative management technology system, petrochemical enterprises can effectively reduce the incidence of accidents, thereby reducing the emission of harmful substances and improving the quality of the ecological environment. This approach not only enhances the enterprise’s sense of environmental responsibility but also strengthens its positive role in the sustainable development goals, helping to realize the environmental protection objectives of the United Nations Sustainable Development Goals. In addition, by identifying and addressing potential hazards promptly, petrochemical enterprises can significantly reduce their risk exposure to neighboring residents, ensure community safety, enhance community trust and support, and contribute to the broader sustainable development goals.

(6)

Although this study has some results, some issues still need to be analyzed in depth and scrutinized. Different experts may be based on personal experience and cognitive differences, and this subjectivity may affect the accuracy and reliability of the final risk grading. Although the variable weight theory is introduced to reduce subjective judgment, it should be necessary to construct a more reasonable basis for scoring so that the data can be more objective and accurate, and the introduction of multi-level and cross-field expert panels for scoring can be considered in future research to enhance the objectivity of the assessment process. The implementation of the risk hierarchical control and hidden danger investigation system in different enterprises may face many challenges, including differences in the corporate culture, insufficient staff training, and resource allocation. Future research could focus on developing implementation guidelines and case studies to help organizations of different sizes and types overcome these challenges. Future research could provide insights into risk management in petrochemical companies by comparing risk management systems with those in other high-risk industries and analyzing the strengths and weaknesses of different models.