Oil and gas field ground construction safety risk assessment method based on dynamic bayesian network

Abstract

This invention relates to the field of safety risk management technology for oil and gas field surface engineering construction, and discloses a method for assessing safety risks in oil and gas field surface construction based on dynamic Bayesian networks. First, it systematically identifies risk factors for all construction processes based on accident causation analysis and a WBS-RBS coupling matrix. Then, it constructs a social network analysis model to quantify the correlation strength of risk factors and determine key risk indicators. Next, it uses the K2 algorithm to optimize the network structure and introduces a time dimension, distinguishing between dynamic and static nodes, and constructs a multi-time-slice dynamic Bayesian network model. Finally, it uses real-time hazard inspection data as evidence input to the model, predicts accident probabilities through forward inference, diagnoses key disaster-causing factors through sensitivity analysis, and outputs risk levels. This method solves the problems of incomplete risk identification, static assessment models, and disconnection from actual construction practices in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of construction safety risk management technology for oil and gas field surface engineering, and in particular to a method for assessing the safety risks of oil and gas field surface construction based on dynamic Bayesian networks. Background Technology

[0002] Oil and gas field surface construction is a crucial stage in oil and gas extraction. Its construction is characterized by numerous sites, long distances, wide geographical coverage, complex processes, high concentrations of hazardous media, and frequent overlapping operations. The construction scope encompasses multiple professional fields, including temporary site facilities, civil engineering, installation engineering, crossing and bridging projects, electrical, automation, and communication engineering, and road and bridge engineering. It involves high-risk procedures such as excavation, hoisting, working at heights, hot work, confined space operations, pipeline welding, and pressure testing. The construction site environment is complex and variable, and the work involves flammable, explosive, toxic, and high-pressure media. Accidents often result in mass casualties and significant property damage. Therefore, conducting safety risk assessments for oil and gas field surface construction is of paramount importance for ensuring personnel safety, smooth project implementation, and the sustainable development of enterprises.

[0003] Currently, domestic and international scholars have conducted extensive research on safety risk assessment for oil and gas field surface construction, resulting in various assessment methods such as the safety checklist method, risk matrix method, event tree analysis method, fault tree analysis method, and hierarchical analysis method. However, existing methods are mostly static assessment models, which are difficult to effectively depict the dynamic evolution of risk factors with changes in construction progress, environmental conditions, personnel status, and management intensity, and cannot achieve real-time judgment and early warning of construction safety status. In terms of risk factor identification, existing research mostly relies on expert experience or single data sources, lacking systematic risk identification methods based on big data of accident cases. This leads to incomplete coverage of risk factors, strong subjectivity in hierarchical classification, and difficulty in fully reflecting the complex risk characteristics of multi-factor coupling and nonlinear transmission in oil and gas field surface construction. In addition, some risk assessment methods have not been deeply integrated with construction procedures (such as work breakdown structures), resulting in a disconnect between theoretical models and actual field applications, and failing to provide accurate support for hazard identification and management.

[0004] Therefore, how to systematically and comprehensively identify safety risk factors and quantify their correlations throughout the entire construction process, how to introduce a time dimension to achieve dynamic risk evolution modeling and probability prediction, and how to closely integrate the assessment model with on-site construction procedures and hazard inspection data have become urgent problems to be solved in the current safety risk assessment of oil and gas field surface construction. Summary of the Invention

[0005] The purpose of this invention is to provide a method for assessing the safety risks of oil and gas field surface construction based on dynamic Bayesian networks, which solves the problems in the prior art of making it difficult to comprehensively and systematically identify construction risk factors and quantify their correlation, being unable to dynamically predict the probability of risk evolution, and having an assessment model that is out of touch with actual construction.

[0006] To achieve the above objectives, this invention provides a method for assessing the safety risks of oil and gas field surface construction based on dynamic Bayesian networks, comprising the following steps: Based on historical accident cases, safety risk factors are classified into five categories through accident causation analysis: unsafe human behavior, unsafe conditions of objects, unsafe environmental conditions, management deficiencies, and technical deficiencies. A Work Breakdown Structure (WBS) and a Risk Breakdown Structure (RBS) are constructed, and a mapping relationship between construction procedures and risk factors is established through the WBS-RBS coupling matrix to achieve comprehensive identification of risk factors in all construction procedures. The risk factors and preset accident types are used as network nodes, and the relationships between risk factors and between risk factors and accident types are used as directed weighted edges. A social network analysis (SNA) model is constructed, and the weighted degree, eigenvector centrality, betweenness centrality, authority, and hub degree of the nodes are calculated to determine key risk indicators. Based on the structure of the SNA model, the initial network structure is optimized using the K2 algorithm, and the maximum likelihood estimation method is used for parameter learning to establish a static Bayesian network model. Based on the static Bayesian network model, a time dimension is introduced. Dynamic nodes and static nodes are distinguished by analyzing whether each risk factor changes with construction progress, personnel status, environmental conditions, or management intensity. A state continuation or state transition mechanism is set for dynamic nodes. Multiple time slices are set, and a dynamic Bayesian network structure containing conditional probability tables within time slices and state transition probability tables between time slices is constructed. The EM algorithm is used to learn parameters for some missing time series data to obtain a dynamic Bayesian risk assessment model. The dynamic Bayesian risk assessment model inputs real-time hazard inspection data obtained at the construction site as evidence. It calculates the probability of various accidents occurring in multiple time units in the future through forward reasoning, and reverse diagnoses key disaster-causing factors through sensitivity analysis, outputting risk level and early warning information.

[0007] This invention provides a method for assessing safety risks in oil and gas field surface construction based on dynamic Bayesian networks. This method achieves systematic and quantitative identification of risk factors throughout the entire construction process. It can dynamically depict the evolution of risks with factors such as construction progress, personnel status, and environmental conditions, and deeply integrates with on-site hazard inspection data. This solves the problems of incomplete risk identification, static assessment models, and disconnection from actual construction in existing technologies, and significantly improves the accuracy, timeliness, and engineering applicability of risk assessment. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0009] Figure 1 This is a diagram of a collapse accident according to the present invention. Figure 2 This is a diagram of mechanical injury accidents according to the present invention. Figure 3 This is the SNA network structure diagram of the causes of oil and gas field surface construction in this invention. Figure 4 This is a distribution diagram of the causal network nodes of oil and gas field surface construction accidents according to the present invention. Figure 5 This is the Bayesian network structure diagram of the initial oil and gas field surface construction safety risk of the present invention. Figure 6 This is a Bayesian network structure diagram of the safety risk of oil and gas field surface construction optimized by the K2 algorithm of this invention. Figure 7 This is a static Bayesian network model diagram of oil and gas field surface construction according to the present invention. Figure 8 This invention presents a dynamic Bayesian network structure diagram of safety risks in oil and gas field surface construction. Figure 9 This is a dynamic Bayesian network model diagram of the safety risks of oil and gas field surface construction in this invention. Figure 10 This is a schematic diagram of the calculation results of the dynamic Bayesian network model of the present invention. Figure 11 This is a flowchart of the safety risk early warning mechanism of the present invention. Figure 12 This is the information push hierarchy diagram of the present invention. Figure 13 This is the overall system architecture diagram of the present invention. Figure 14 This is a schematic diagram of the closed-loop management mechanism of the present invention. Figure 15 This is the SNA network structure diagram of nodes R1-10 of the present invention. Figure 16 This is a flowchart of the steps of the oil and gas field surface construction safety risk assessment method based on dynamic Bayesian network of the present invention. Detailed Implementation

[0010] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0011] Please refer to Figures 1 to 16 This invention provides a method for assessing the safety risks of oil and gas field surface construction based on dynamic Bayesian networks, comprising the following steps: S101: Based on historical accident cases, safety risk factors are classified into five categories through accident causation analysis: unsafe human behavior, unsafe conditions of objects, unsafe environmental conditions, management deficiencies, and technical deficiencies. A Work Breakdown Structure (WBS) and a Risk Breakdown Structure (RBS) are constructed, and a mapping relationship between construction procedures and risk factors is established through the WBS-RBS coupling matrix to achieve comprehensive identification of risk factors in all construction procedures. Specifically, the first step was to collect and organize safety accident cases that had occurred in the field of oil and gas field surface construction over the years. Due to the lack of public disclosure or difficulty in obtaining information on some accidents, this implementation method is an incomplete statistical analysis, collecting a total of 115 cases of oil and gas field surface construction accidents. Accident collection employed two methods: first, using octopus crawler technology to automatically retrieve historical accident cases from official websites such as the Ministry of Emergency Management's website and the Safety Management Network; second, consulting accident cases in published books. Through these methods, an original case database containing 115 accidents was established.

[0012] Based on accident causation theory, a causal analysis was conducted on the aforementioned historical accident cases, classifying the causal factors into five categories: unsafe acts by people, unsafe conditions of equipment, unsafe environmental conditions, and management and technical deficiencies. Unsafe acts by people include 12 items: lack of safety awareness, lack of awareness of compliance with rules and regulations, lack of awareness of safety management responsibility, poor physical condition, poor psychological condition, lack of qualification certificates, improper operation, violation of regulations, failure to conduct gas detection and analysis, failure to take hazard avoidance measures, inadequate risk identification, and insufficient experience. Unsafe conditions of equipment include 8 items: lack of safety protective equipment, lack of safety warning signs, lack of protective fencing, unreasonable selection of machinery and equipment, insecure installation of machinery and equipment, defects in machinery and equipment, overloading of machinery and equipment, and lack of a safety use permit for machinery and equipment. Unsafe environmental conditions include haphazard material storage, poor nighttime lighting, toxic gases, and ground conditions. The five categories of risk factors are: poor base environment, narrow environment, steep slope, adverse weather, and height suspension, totaling 8 items; management deficiencies include: inadequate safety education, incomplete safety management system, lack of a safety responsibility system, lack of dedicated safety management personnel, inadequate emergency management, lack of dedicated command personnel, inadequate on-site safety management, and inadequate construction organization, totaling 8 items; and technical deficiencies include: lack of construction permit, unqualified construction unit, lack of construction plan or construction plan that does not meet requirements, lack of safety operating procedures or safety operating specifications that do not meet requirements, lack of safety technical briefing, inadequate engineering measures, lack of mastery of new technologies and processes, failure to construct according to the construction plan, and failure to operate according to safety operating specifications, totaling 9 items. These five categories comprise a total of 45 specific risk factors.

[0013] Based on this, a Work Breakdown Structure (WBS) is constructed, decomposing oil and gas field surface construction into three levels according to construction specialties: the first level is oil and gas field surface construction; the second level is divided into six categories according to specialties: site temporary facilities, civil engineering, installation engineering, crossing engineering, electrical / automation / communication engineering, and road and bridge engineering; the third level is further subdivided into 50 work units, including but not limited to site leveling and temporary road paving, campsite transportation and hoisting, prefabrication, foundation pit excavation and dewatering, rubble foundation construction, pile foundation work, dynamic compaction, formwork work, rebar tying, concrete work, main structure masonry work, and roof waterproofing. Work operations, decoration and renovation work, station and yard road work, surveying and setting out, work zone cleaning, pipeline laying, trench excavation, construction lane relocation and construction, material hoisting and transportation, assembly and welding, weld heat treatment, corrosion protection and heat insulation, pipeline laying in trenches, overhead pipeline positioning, jointing and splicing, trench backfilling and landform restoration, light rail and cableway, pump and motor installation, container installation, tower installation, furnace installation, flare fabrication and installation, steel structure fabrication and installation, ball passing, diameter measurement, pressure testing and drying, storage tank installation, pipeline purging and cleaning, directional drilling crossing, large-scale excavation crossing, pipe jacking crossing, shield tunneling crossing, tunnel crossing, pipeline crossing, electrical engineering, automation and communication, road work and bridge work.

[0014] Simultaneously, a Risk Decomposition Structure (RBS) is constructed, using the aforementioned five categories and 45 specific risk factors as the third-level risk nodes. A coupling matrix is ​​constructed using the WBS-RBS coupling method: the columns of the matrix represent the b-th work unit within the a-th task decomposed in the WBS, and the rows represent the d-th risk factor within the c-th risk decomposed in the RBS. Referring to specific safety inspection items in relevant standards, each work unit and each risk factor's intersection is individually assessed: if the risk factor exists in the work unit, it is marked as "1"; if it does not exist or its probability of existence is extremely low, it is marked as "0". Through this binary labeling, a complete WBS-RBS coupling matrix is ​​formed, establishing a precise mapping relationship between construction procedures and risk factors, achieving comprehensive risk identification across all construction procedures and scenarios.

[0015] The partial correspondences are shown in Table 1 below: Table 1. Coupling of WBS-RBS in Oil and Gas Field Surface Construction

[0016] This mapping relationship provides a standardized and structured data foundation for subsequent social network analysis, Bayesian network modeling, and on-site hazard investigation and management.

[0017] S102: Using the risk factors as network nodes and the relationships between risk factors as directed weighted edges, construct a Social Network Analysis (SNA) model, and calculate the weighted degree, eigenvector centrality, betweenness centrality, authority, and hub degree of the nodes to determine key risk indicators. Specifically, structural dynamics methods were used to deeply analyze the 115 historical accident cases collected in step S101, systematically identifying the relationships between the risk factors of each accident, and constructing and extracting the causal network diagrams corresponding to each accident. For example, the causal network of a certain collapse accident is as follows: Figure 1 As shown, the causal network of a certain mechanical injury accident is as follows: Figure 2 As shown.

[0018] After extracting the causal networks of each accident, all networks are integrated and transformed into a side table format of "Source node + Target node + Weight". Some data of the transformation results are shown in Table 2.

[0019] Table 2 Topology Transformation Results

[0020] In this system, source nodes correspond to causal factors, target nodes correspond to consequence factors, and each risk factor is replaced by a number. The frequency of occurrence of the correlation between factors is used as the weight. For example, the weight of "R1-1→R1-11" is 21, the weight of "R1-2→R1-8" is 18, and the weight of "R4-2→R1-11" is 22.

[0021] Import the above side table into Gephi software to construct an SNA directed weighted network model of risk factors for oil and gas field surface construction, such as... Figure 3 As shown, the network contains 54 nodes, including 43 risk factor nodes, 9 accident nodes (including fire, explosion, falling object, fall from height, vehicle injury, mechanical injury, poisoning / asphyxiation, collapse, and others), and 196 directed edges. The thickness of the edges represents their weight; the greater the weight, the thicker the edge. Figure 3 It can be clearly seen that the width of the connecting lines between R1-1 and R4-2, R4-2 and R2-3, R2-3 and R5-7, and R1-9 and R1-2 is significantly greater than that of other edges, indicating that the connection strength and influence between these nodes are relatively higher.

[0022] Based on this, multiple centrality indicators of network nodes are calculated to determine key risk indicators, including: (1) Node weighted degree. The weighted degree includes the weighted in-degree, weighted out-degree, and weighted total degree. Its core definition is the sum of the weights of all edges connected to the node. The node degree statistics are shown in Table 3.

[0023] Table 3 Node Degree Statistics

[0024] As can be seen from Table 3, the weighted total value of R1-1 (lack of safety awareness) is 120, the weighted total value of R1-11 (inadequate risk identification) is 121, the weighted total value of R4-1 (inadequate safety education) is 218, and the weighted total value of R4-2 (incomplete safety management system) is 258. Figure 4 The distribution of node degrees is shown. The weighted degree of the above-mentioned nodes is significantly higher than that of other nodes, making them core related nodes in the risk network.

[0025] (2) Eigenvector centrality. This index measures the indirect influence and radiation effectiveness of a node in the network. The calculation formula is shown in Equation (1): (1). Among them, This represents the connection relationship between node i and node j; Let represent the importance of node i; c is a scaling constant. The eigenvector centrality values ​​of the nodes are shown in Table 4.

[0026] Table 4. Centrality values ​​of node eigenvectors

[0027] As can be seen from Table 4, nodes R1-9 (no gas detection and analysis), R2-1 (lack of safety protection equipment), R2-2 (lack of safety warning signs), R2-3 (lack of enclosure and fencing), R2-4 (unreasonable selection of mechanical equipment), R2-5 (unsecured installation of mechanical equipment), R2-6 (defective mechanical equipment), R3-1 (disorganized material stacking), and R5-6 (inadequate engineering measures) have high eigenvector centrality.

[0028] (3) Betweenness centrality. This metric measures the frequency with which a node appears on the shortest path between all other nodes in the network. The calculation formula is shown in equation (2): (2), of which The total number of shortest paths from node vs to node vt; This refers to the number of nodes that pass through node vi in ​​the shortest path from node vs to vt. Node betweenness centrality values ​​are shown in Table 5.

[0029] Table 5. Node Betweenness Centrality Values

[0030] As can be seen from Table 5, the betweenness centrality of R1-8 (violations of rules and regulations) is 20.25, the betweenness centrality of R4-7 (inadequate on-site safety management) is 44.53, the betweenness centrality of R1-10 (failure to take risk avoidance measures) is 11.67, and the betweenness centrality of R1-11 (inadequate risk identification) is 16.37.

[0031] (4) Authority and Hubiness. Authority and hubiness are a pair of complementary metrics based on the HITS algorithm. Authority measures the degree to which a node is pointed to by other nodes with high hubiness, while hubiness measures the degree to which a node points to other nodes with high authority. The calculation formulas are shown in equations (3) to (6): (3) (4) Among them, authority is used Indicates that the degree of hub is used

[0032] When the iteration converges, the pivot degree and authority degree vectors respectively satisfy the eigenvalue equations. (5) (6) Where λ is ( The principal eigenvalues ​​of ), h is The principal eigenvector, a is The main eigenvectors.

[0033] The node authority and hub values ​​are shown in Table 6.

[0034] Table 6. Node Authority and Hub Function Values

[0035]

[0036] As can be seen from Table 6, the pivotality value for R1-10 (without risk mitigation measures) is 0.2836. For example... Figure 15 As shown, this node can simultaneously point to multiple accident nodes such as A2 (explosion), A3 (object strike), A5 (vehicle damage), A6 (mechanical injury), A7 (poisoning and asphyxiation), and A9 (other) through directed edges, becoming a common risk source for various safety accidents. Meanwhile, the authority value of R2-5 (loosely installed mechanical equipment) is 0.06298, and the authority value of R1-12 (lack of experience) is 0.02194. These nodes are pointed to by multiple hub nodes and are the core carriers of the final risk consequences.

[0037] Through quantitative analysis of the five dimensions mentioned above—weighted degree, eigenvector centrality, betweenness centrality, authority, and hubness—key risk nodes such as lack of safety awareness, inadequate risk identification, insufficient safety education, and unsound safety management systems can be accurately identified. The correlation strength, scope of influence, and role in risk transmission of each risk node can be clarified, thereby determining core risk indicators and providing data support and theoretical basis for the subsequent construction of static and dynamic Bayesian network models.

[0038] S103: Based on the structure of the SNA model, the initial network structure is optimized using the K2 algorithm, and the maximum likelihood estimation method is used for parameter learning to establish a static Bayesian network model. Specifically, firstly, the SNA network topology obtained in step S102 is used as the initial Bayesian network structure. The static Bayesian network is then drawn using GeNIe software to obtain the initial Bayesian network structure for the safety risks of oil and gas field surface construction, as shown below. Figure 5 As shown, the initial network structure is complex and contains redundant edges, which increases the difficulty of subsequent parameter learning.

[0039] To optimize the network structure, the K2 algorithm is used for structure learning. The K2 algorithm, proposed by Cooper and Herskovits in 1992, is a Bayesian network structure learning algorithm based on greedy search and a Bayesian scoring criterion. Its core objective is to efficiently mine causal relationships between variables under limited sample and prior knowledge conditions, generating a network topology with good fit and moderate complexity. The algorithm first constructs a suitable scoring function as the criterion for judging the quality of the network structure; then, it uses a hill-climbing method as a search strategy to solve for the network structure B_s with the maximum posterior probability based on a given dataset D. That is, the optimal network structure is determined by finding B_s that maximizes the posterior probability. The formula for calculating the posterior probability is shown in equation (7): (7) right Taking the logarithm gives us: (8) The scoring function obtained from the K2 algorithm is: (9) (10) Where P(Bs) is the prior probability of the model, and Score(Bs|D) is the model's score; n is the number of variables. The total number of possible values ​​for the parent node set of variable xi This represents the total number of possible values ​​for variable xi. Let be the number of samples in dataset D where the parent node set πi of variable xi takes the j-th value, and the variable's state is the k-th value.

[0040] The nodes in a Bayesian network structure are generally binary discrete variables, where 0 represents that the risk factor did not occur and 1 represents that the risk factor occurred. The risk factors from the 115 accident cases collected in step S101 were statistically analyzed, and a 0-1 matrix was constructed. Some data are shown in Table 7.

[0041] Table 7: 0-1 matrix of construction accidents at some oil and gas field surfaces.

[0042] Based on the initial network model obtained from SNA, the K2 algorithm was used to optimize the model using the pgmpy library in the Python programming language. During the algorithm execution phase, three core constraints were set: first, the number of nodes was limited to 56; second, improvements were required based on the existing network structure; and third, each risk factor node was ensured to have a connection. The optimized Bayesian network structure is as follows. Figure 6 As shown, this structure removes redundant edges, resulting in a simpler and more rational network topology.

[0043] After optimizing the network structure, parameter learning is performed, which involves determining the Conditional Probability Table (CPT) for each node. The core function of the CPT is to quantify the influence of the parent node's state on the child node's state, forming the basis for uncertain reasoning. This step employs Maximum Likelihood Estimation (MLE) for parameter learning. The MLE algorithm, based on the maximum likelihood theorem, primarily finds the model parameters that maximize the probability of the data occurrences using complete observation data. Its core steps include: defining the likelihood function... Let represent the observed dataset, consisting of independent and identically distributed samples whose common probability distribution is defined by the parameter vector. If determined, then its likelihood function It can be represented as: (11) Log-likelihood function: Taking the log-likelihood function simplifies the calculation.

[0044] (12) Finding the derivative: Take the derivative of the logarithmic function, set the derivative to 0, and solve for the parameters. .

[0045] (13) Solving for the parameters: Solving the equations yields estimated values ​​for the parameters.

[0046] (14) This method requires no prior knowledge or subjective intervention, relies entirely on sample data, has a simple computational logic, strong interpretability, high computational efficiency, and can accurately capture the statistical patterns of the samples.

[0047] The optimized static Bayesian network structure and conditional probability table are input into the GeNIe software for parameter learning of the maximum likelihood estimation algorithm, resulting in a static Bayesian network model for the safety risks of oil and gas field surface construction. Figure 7 As shown. The conditional probabilities of all nodes can be obtained using the GeNIe software.

[0048] After the model was built, the static Bayesian network model was validated using k-fold cross-validation. k-fold cross-validation is a commonly used model performance evaluation method in machine learning. Its core logic is to reduce sample bias caused by a single partition and improve the reliability of the validation results by splitting and training the samples multiple times. The core process is as follows: the complete sample set is randomly and uniformly divided into k independent subsets (folds), where k-1 subsets are used as the training set for model parameter calibration and training, and the remaining subset is used as the test set to evaluate the model's predictive performance; this process is repeated k times, with different subsets used as the test set each time; finally, the average of the k test results is taken as the final performance evaluation index of the model. In this implementation, k is set to 5, i.e., a 5-fold cross-validation method is used for validation calculation. The validation accuracy is shown in Table 8.

[0049] Table 8. Prediction accuracy of static Bayesian network models for different accidents.

[0050] The static model achieved an overall average prediction accuracy of 90.48% for nine types of accidents, indicating that the static Bayesian network model can effectively describe the static correlation between risk factors and has basic safety risk prediction capabilities, laying a solid foundation for the subsequent introduction of the time dimension to construct a dynamic Bayesian network model.

[0051] S104: Based on the static Bayesian network model, a time dimension is introduced. Dynamic nodes and static nodes are distinguished by analyzing whether each risk factor changes with construction progress, personnel status, environmental conditions or management intensity. A state continuation or state transition mechanism is set for dynamic nodes. Multiple time slices are set to construct a dynamic Bayesian network structure containing conditional probability tables within time slices and state transition probability tables between time slices. The EM algorithm is used to learn parameters for some missing time series data to obtain a dynamic Bayesian risk assessment model. Specifically, firstly, all nodes in the static Bayesian network are analyzed one by one. Considering the temporal characteristics of oil and gas field surface construction, dynamic nodes that change over time are identified, and their changes over time are clarified. The detailed analysis is as follows: (1) Human factors: including 12 factors from R1-1 to R1-12. Among them, R1-1 (lack of safety awareness), R1-2 (lack of awareness of compliance with rules and regulations), and R1-3 (lack of awareness of safety management responsibility) can be dynamically improved through safety education and training, and their status changes with the frequency and effect of training; R1-7 (improper operation) and R1-8 (violation of rules and regulations) are operational behaviors, which are affected by the intensity of on-site management and the degree of work fatigue, and the standardization of operation varies in different construction stages; R1-9 (failure to conduct gas detection and analysis), R1-10 (failure to take risk avoidance measures), and R1-11 (inadequate risk identification) are process operations, which are dynamically adjusted according to the work location and work content; R1-4 (poor physical fitness) and R1-5 (poor psychological fitness) can be improved through medical treatment, medication and rest, and are also dynamic states; R1-6 (failure to obtain qualification certificate) and R1-12 (insufficient experience) require long-term learning and accumulation in the later stage, and are static nodes. Among them, R1-7, R1-8, R1-9, R1-10, and R1-11 are all habitual actions that will form path dependencies and continue in the state.

[0052] (2) Material factors: including 8 factors from R2-1 to R2-8. Among them, R2-1 (lack of safety protection equipment), R2-2 (lack of safety warning signs), and R2-3 (lack of enclosures and barriers) can be dynamically improved through on-site replenishment and change in real time with the allocation of materials; R2-5 (insecure installation of mechanical equipment) is affected by the installation process and subsequent maintenance, and there is a state transition from "secure" to "loose"; R2-7 (overload of mechanical equipment) changes dynamically with the amount of construction tasks; R2-4 (unreasonable selection of mechanical equipment) is determined in the construction plan design stage and belongs to static attributes; R2-6 (defects in mechanical equipment) is an inherent attribute of the equipment before it leaves the factory or enters the site and belongs to static attributes; R2-8 (mechanical equipment has not obtained a safety use permit) does not change dynamically with the construction process and belongs to static attributes. Among them, the adjustment of the amount of construction tasks in R2-7 has inertia, and the overload state will continue if there is no plan optimization.

[0053] (3) Environmental factors: including 8 factors from R3-1 to R3-8. Among them, R3-1 (disorganized material stacking) can be dynamically improved through on-site organization; R3-2 (poor nighttime lighting) can be dynamically optimized by adding lighting equipment; R3-3 (toxic gases) changes dynamically with ventilation conditions and work activities, and there is a state switch of "exceeding the standard - meeting the standard"; R3-7 (adverse weather) is a time-sensitive environmental factor, which appears or disappears dynamically over time; R3-4 (poor foundation environment), R3-5 (narrow environment), R3-6 (steep slope), and R3-8 (high-altitude suspension) are inherent attributes and belong to static nodes. Among them, R3-1 and R3-3 require special treatment by personnel, and the state continues; the addition of lighting equipment for R3-2 has a lag effect, and the state remains unchanged without rectification.

[0054] (4) Management Factors: These include eight factors from R4-1 to R4-8. Among them, R4-1 (inadequate safety education) will change dynamically with implementation, such as being transformed into "adequate" after regular training; R4-4 (lack of dedicated safety management personnel) and R4-6 (lack of dedicated command personnel) can be dynamically adjusted according to the construction risk level; R4-7 (inadequate on-site safety management) and R4-8 (inadequate construction organization) are affected by the construction progress and the complexity of the operation, and the management intensity varies at different construction stages; R4-2 (incomplete safety management system) and R4-3 (lack of established safety responsibility system) belong to the top-level design of the project preparation stage, and once determined, they will not be frequently adjusted during the construction period, and are considered static attributes; R4-5 (inadequate emergency management) is prepared in the early stage of construction and will not change in a short period of time, and is considered a static node. Among them, the implementation of the safety education plan of R4-1 has inertia, and the state remains unchanged if there is no adjustment; the adjustment of the intensity of on-site management of R4-7 has lag, and the state remains unchanged if there is no supervision.

[0055] (5) Technical factors: These include nine factors from R5-1 to R5-9. Among them, R5-5 (no safety technical briefing), R5-8 (construction not in accordance with the construction plan), and R5-9 (operation not in accordance with safety operation specifications) are process-related and change with the progress of construction procedures; R5-6 (inadequate engineering measures) can be dynamically improved through construction rectification; R5-1 (no construction permit) and R5-2 (construction unit's qualifications do not meet the requirements) are determined by the company's own conditions and relevant departmental certifications, and will not change dynamically during construction, thus belonging to static attributes; R5-7 (lack of mastery of new technologies and processes) will improve with training and practical experience accumulation, and will not change in the short term, thus belonging to static attributes; R5-3 (no construction plan developed) and R5-4 (no safety operation procedures developed) are technical preparation documents in the early stage of construction, and will not be dynamically adjusted after formal construction unless there are major design changes, thus belonging to static attributes. Among them, the improvement of engineering measures in R5-6 requires adjustment of the plan, and the poor measures at the current stage will lead to problems with the measures in the next time period.

[0056] (6) Accident nodes: The occurrence status of nine types of accidents, A1 to A9, changes with the dynamic evolution of risk factors, which is manifested as the state transition of the result node in the dynamic Bayesian network.

[0057] After completing the state transition analysis of the nodes, the dynamic Bayesian network structure for oil and gas field surface construction was drawn using the temporalplate canvas in GeNIe software. Nodes with static attributes were drawn outside the temporalplate area, while nodes with dynamic attributes were drawn within the temporalplate area. A time step of 8 was set to a period of one week, meaning that one day was considered a time segment except for the initial state. The constructed dynamic Bayesian network structure is as follows: Figure 8 As shown.

[0058] The core parameters of a dynamic Bayesian network include the Conditional Probability Table (CPT) and the Transition Probability Table (TPT). The CPT describes the probability distribution of a node's own state given the state of its parent node, quantifying the dependency of a node's state at a specific time step on its parent node's state. The TPT characterizes the probability of a dynamic node's state change between different time steps. Due to the introduction of time series data, the data required for parameter learning must include multi-time-slice information. This implementation method extracts relevant data from the week preceding the events of 115 accidents, using this data as the dataset for dynamic Bayesian network parameter learning. Partial data is shown in Table 9.

[0059] Table 9 Data set one week before the accident

[0060] Because some accidents lack complete data from the week prior to their occurrence, resulting in partial data gaps in the dataset, the Expectation-Maximization (EM) algorithm is employed for parameter learning. The EM algorithm is an iterative parameter estimation algorithm for data with missing information. Its core idea is to iteratively approach the optimal parameter values ​​by alternating between "expectation steps (E-step)" and "maximization steps (M-step)." For DBN parameter learning, the EM algorithm treats unobserved node states (including node states corresponding to missing data and hidden state transitions across time slices) as latent variables. It iteratively calculates the posterior expectation of these latent variables and then maximizes the likelihood function based on the expectation to obtain converged parameter estimation results. The specific steps are as follows: Initial parameterization: Assume the initial parameters are (15) Step E: Calculation : (16) M-step: Update parameters : (17) Repeated iterations: Repeat the E-step and M-step until the parameters converge.

[0061] The dynamic Bayesian network structure of oil and gas field surface construction and the previous week's dataset were input into GeNIe software. The EM algorithm was used for parameter learning to obtain a dynamic Bayesian network model of safety risks in oil and gas field surface construction. Figure 9 As shown. After establishing the dynamic Bayesian network model, the conditional probabilities of all nodes at different time slices can be obtained in the GeNIe software.

[0062] The dynamic Bayesian network model was validated using k-5 fold cross-validation, and the validation accuracy is shown in Table 10.

[0063] Table 10 Prediction accuracy of dynamic Bayesian network model for different accidents

[0064] The dynamic model achieved an overall average prediction accuracy of 92.75% for nine accident types, an improvement over the static model, validating the effectiveness of the dynamic Bayesian network model. Subsequent steps will involve applying this dynamic Bayesian risk assessment model to practical engineering projects.

[0065] S105: Input the real-time hazard inspection data obtained at the construction site as evidence into the dynamic Bayesian risk assessment model, calculate the probability of various accidents occurring in multiple time units in the future through forward reasoning, and diagnose key disaster-causing factors through sensitivity analysis, and output risk level and early warning information.

[0066] Specifically, firstly, real-time hazard inspection data from the construction site is acquired. To facilitate integration with the dynamic Bayesian network model, a hazard database is pre-established, associating hazard details with corresponding risk factors. This implementation method uses a specific project as an example, collecting all safety inspection data from the project's construction site on a specific day of a specific month of a specific year. Some hazard records are shown in Table 11. Table 11 Hazard Records from Safety Inspections

[0067] This includes the type of hazard (human factors, material factors, environmental factors, technical factors, and management factors), the specific content of the hazard, and the corresponding risk factor node number. For example, workers operating on an uninspected scaffolding platform in the No. 2 pipe gallery of the purification plant are working, corresponding to risk factor R1-1 (lack of safety awareness); the protective cover of the secondary terminal block of the welding machine in the hazardous waste temporary storage room of the purification plant is damaged, corresponding to R2-6 (mechanical equipment defect); the excavation of the optical cable trench at the well station was carried out without a pre-work safety analysis, a pre-shift meeting, or the provision of an emergency escape call device, corresponding to R4-1 (inadequate safety education), etc.

[0068] The aforementioned hazard information is used as evidence input into the dynamic Bayesian risk assessment model. Specifically, in the GeNIe software, the node corresponding to the hazard information is assigned the state of "occurred" (State=1) at the initial time (t=0), while the states of all other nodes are set to "not occurred" (State=0). Figure 10 As shown in Table 12, after model update calculation, the probability values ​​of occurrence of 9 types of accidents within the next 7 time units can be obtained.

[0069] Table 12 Probability of 9 Types of Accidents in the Next 7 Days

[0070] As shown in Table 12, at the initial stage of hazard triggering at time t=1, the probability of A2 (explosion) is 21.79%, and the probability of A6 (mechanical injury) is 24.46%, both considered high-risk accidents; the probability of A3 (object strike) is 13.26%, considered a medium-risk accident; and the remaining accidents are low-risk accidents. The evolution of the probabilities of various accidents over time shows different trends: A2, A6, A8, and A9 show a rapid decrease; A4 and A7 show a slow decrease; A1 and A3 show a fluctuating and stable trend; and A5 shows a slight increase.

[0071] After obtaining the probability of an accident, a risk level classification is performed. The maximum value among the nine calculated accident probabilities is used as the core criterion. Referring to relevant standards and specifications for safety management of oil and gas field surface construction, the risk level of the construction site is divided into four levels, as shown in Table 13. Table 13 Risk Level Classification Standards

[0072] The specific method for determining the risk level is as follows: The occurrence probabilities of the nine accident categories (A1 fire, A2 explosion, A3 falling object, A4 fall from height, A5 vehicle injury, A6 machinery injury, A7 poisoning and suffocation, A8 collapse, A9 other) calculated and output by the dynamic Bayesian network model are compared. The maximum value is taken, and the overall risk level of the current construction safety is determined according to the probability range shown in Table 13. For example, an accident occurrence probability greater than 0.3 is classified as Level I major risk, 0.03 to 0.3 as Level II relatively high risk, 0.003 to 0.03 as Level III general risk, and less than 0.003 as Level IV low risk. Risk acceptance criteria and control principles are shown in Table 14. Table 14 Risk Acceptance Criteria and Control Principles

[0073] Level I risk is unacceptable and requires high attention and necessary control measures; Level II risk is undesirable and should be taken seriously with effective control measures; Level III risk is acceptable and effective control measures are advisable; Level IV risk is acceptable and no control measures are required. This implementation method uses the maximum probability among the nine types of accidents as the core judgment criterion to determine the current construction safety risk level.

[0074] Simultaneously, sensitivity analysis is used to diagnose key disaster-causing factors. Sensitivity analysis is a key method for quantitatively identifying core input variables. In the dynamic Bayesian network model, probability importance is calculated for discrete risk nodes, as shown in equation (18): (18), where a, b, c, and d are coefficients determined by the Bayesian network structure and the current evidence. If the absolute value of the derivative D is large, it indicates that the parameter has a greater impact on the posterior probability, and vice versa.

[0075] Nine types of accidents were designated as target nodes, and sensitivity analysis was performed sequentially. The key influencing factors for different accidents were determined by the depth of the node's color; the deeper the red of the node, the greater its influence on the corresponding target accident node. The analysis results show that: for A1 (fire), R1-1 (lack of safety awareness) and R1-9 (failure to conduct gas detection analysis) had the greatest impact; for A2 (explosion), R3-1 (disorganized material storage) was the most critical trigger; for A3 (object strike), unsafe conditions of objects (R2-2, R2-4, R2-7, R2-8) were dominant; for A4 (fall from height), R3-8 (suspended environment at height) was the most critical triggering factor; for A5 (vehicle injury), R1-1 and R1-10 (failure to take evasive action) were the key risk sources; for A... For A6 (mechanical injury), the core causes are R1-12 (insufficient experience), R2-5 (loose installation of mechanical equipment), R3-6 (steep slope), and R4-6 (lack of dedicated command personnel); for A7 (poisoning and asphyxiation), R3-3 (toxic and harmful gases) is the most critical cause; for A8 (collapse), R5-6 (inadequate engineering measures) is the most critical cause; for A9 (other), it is mainly affected by factors such as R1-7 (improper operation), R1-10 (lack of hazard avoidance measures), R3-5 (narrow environment), and R4-2 (inadequate safety management system).

[0076] Based on the output of risk levels and early warning information, this method further embeds a dynamic Bayesian risk assessment model into a dual prevention mechanism framework to construct a security risk management platform. Specifically: First, a standardized hazard database is established. This database is based on the WBS-RBS coupling matrix constructed in step S101. Based on the three-layer decomposition structure of the WBS, it is first categorized into six major work types (temporary site facilities, civil engineering, installation engineering, crossing and bridging engineering, electrical / automation / communication engineering, and road and bridge engineering), and then further subdivided into 50 work units (i.e., the third layer of the WBS). For each work unit, its specific inspection content (including work management, protective isolation, tools and equipment, personal protective equipment, work techniques, emergency preparedness, etc.) is analyzed. Each inspection content is mapped to one of 45 risk factors, clarifying the hazard level of each inspection content (divided into general hazards and major hazards), and determining the mapping relationship between hazards and risk factors, so that hazards discovered in daily safety inspections can be directly correlated to the status of specific risk factors.

[0077] Secondly, an intelligent hazard identification mechanism is established. Based on the characteristics of the inspection items in the hazard database, they are divided into document-based inspection items and on-site environmental verification inspection items. For document-based inspection items (such as construction plans, safety technical briefings, personnel qualifications, equipment certificates, etc.), online uploading, hierarchical review in the background, and system archiving are achieved through the safety management platform. For on-site environmental verification inspection items (such as working environment, protective facilities, equipment status, temporary power supply, etc.), a fusion mode of "intelligent monitoring + manual inspection" is adopted. By using fixed-point monitoring by construction site cameras, mobile inspection by robotic dogs, and high-altitude inspection by drones, combined with large-scale model image recognition, the inspection content in the hazard database is automatically matched to achieve rapid identification and reporting of hazards.

[0078] Then, a dual early warning system is established. This system issues warnings based on both the level of hazard and the level of safety risk, with two levels: red and orange. The early warning mechanism process is as follows: Figure 11 As shown. Regarding hazard level early warning, major hazards directly trigger a red alert, while general hazards, if not rectified satisfactorily or not rectified within the specified time, trigger an orange alert; failure to rectify within the specified time escalates to a red alert. Regarding risk level early warning, Level I and II risks directly trigger a red alert, Level III risks trigger an orange alert, and Level IV risks do not trigger an alert. Early warning information is pushed through multiple channels, including platform messages and SMS, in a hierarchical manner, as shown in the following hierarchy: Figure 12 As shown: Red alert information is pushed to the group's safety management level, project safety management level, and project department safety personnel; Orange alert information is pushed to the project safety management level and project department safety personnel; Level IV low risk information is only pushed to project department safety management personnel.

[0079] Finally, based on the above mechanisms, an integrated security risk management platform (such as a data acquisition layer, a technical support layer, and a user layer) is constructed. Figure 13(As shown). The data acquisition layer adopts a dual acquisition mode combining manual and intelligent inspection equipment. It acquires hazard data in real time through intelligent devices such as cameras, drones, and robotic dogs, as well as manual input by on-site safety management personnel. The technical support layer integrates a hazard database, an intelligent hazard identification mechanism, a dynamic Bayesian risk assessment model, and a dual early warning system. The user layer is divided into a group safety management terminal and a project safety management terminal. The project safety management terminal is used by project safety management personnel and on-site safety officers, while the group safety management terminal allows group safety management personnel to implement unified control over multiple project sites. Through this platform, intelligent management of the entire process of hazard identification, risk assessment, early warning push, and closed-loop handling is achieved: the platform records information such as early warning trigger time, early warning level, hazard content, responsible person for push, response status, rectification measures, and rectification results, forming a hazard handling ledger; after rectification, the on-site safety officer uploads rectification photos, which are automatically verified by the platform. If the verification is successful, the early warning status is automatically lifted. If the rectification fails to meet the standards, the early warning continues to be triggered until the hazard rectification is completed, thus achieving closed-loop management of "early warning-response-rectification-verification-lift" (e.g., Figure 14 (As shown).

[0080] Through the above steps, dynamic assessment of safety risks in oil and gas field surface construction, diagnosis of key disaster-causing factors, and graded early warning output based on dynamic Bayesian networks were achieved. The deep integration of dynamic Bayesian network model and dual prevention mechanism was also completed, providing a complete practical solution for dynamic management and control of safety in oil and gas field surface construction, and effectively improving the accuracy, timeliness and engineering applicability of risk assessment.

[0081] Thus, firstly, based on historical accident cases and the WBS-RBS coupling matrix, a systematic and structured identification of risk factors across all construction processes was achieved, solving the problems of incomplete risk coverage and reliance on expert experience in traditional methods. Secondly, social network analysis quantified the correlation strength and transmission path between risk factors, providing an objective topological structure for Bayesian network modeling. Building upon this, a dynamic Bayesian network model was constructed by introducing a time dimension, effectively depicting the dynamic evolution of risk factors with changes in construction progress, personnel status, and environmental conditions, enabling rolling predictions of future multi-time-step accident probabilities. Finally, by using real-time hazard inspection data as evidence input to the model, combined with forward reasoning and sensitivity analysis, dynamic assessment of accident probabilities and reverse diagnosis of key disaster-causing factors were achieved, outputting tiered early warning information. This method significantly improves the accuracy, timeliness, and engineering practicality of safety risk assessment for oil and gas field surface construction, promoting a shift in safety management from post-event response to pre-event early warning.

[0082] The above-disclosed embodiments are merely one or more preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes for implementing the above embodiments and equivalent changes made in accordance with the claims of this application still fall within the scope of this application.

Claims

1. A method for assessing safety risks in oil and gas field surface construction based on dynamic Bayesian networks, characterized in that, Includes the following steps: Based on historical accident cases, safety risk factors are classified into five categories through accident causation analysis: unsafe human behavior, unsafe conditions of objects, unsafe environmental conditions, management deficiencies, and technical deficiencies. A Work Breakdown Structure (WBS) and a Risk Breakdown Structure (RBS) are constructed, and a mapping relationship between construction procedures and risk factors is established through the WBS-RBS coupling matrix to achieve comprehensive identification of risk factors in all construction procedures. The risk factors and preset accident types are used as network nodes, and the relationships between risk factors and between risk factors and accident types are used as directed weighted edges. A social network analysis (SNA) model is constructed, and the weighted degree, eigenvector centrality, betweenness centrality, authority, and hub degree of the nodes are calculated to determine key risk indicators. Based on the structure of the SNA model, the initial network structure is optimized using the K2 algorithm, and the maximum likelihood estimation method is used for parameter learning to establish a static Bayesian network model. Based on the static Bayesian network model, a time dimension is introduced. Dynamic nodes and static nodes are distinguished by analyzing whether each risk factor changes with construction progress, personnel status, environmental conditions, or management intensity. A state continuation or state transition mechanism is set for dynamic nodes. Multiple time slices are set, and a dynamic Bayesian network structure containing conditional probability tables within time slices and state transition probability tables between time slices is constructed. The EM algorithm is used to learn parameters for some missing time series data to obtain a dynamic Bayesian risk assessment model. The dynamic Bayesian risk assessment model inputs real-time hazard inspection data obtained at the construction site as evidence. It calculates the probability of various accidents occurring in multiple time units in the future through forward reasoning, and reverse diagnoses key disaster-causing factors through sensitivity analysis, outputting risk level and early warning information.

2. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, Construct a Work Breakdown Structure (WBS) and a Risk Breakdown Structure (RBS), and establish a mapping relationship between construction procedures and risk factors through a WBS-RBS coupling matrix. Specifically, this includes: The Work Breakdown Structure (WBS) includes the breakdown of the oil and gas field surface into six major categories according to construction specialties: site temporary facilities, civil engineering, installation engineering, crossing and bridging engineering, electrical, automation and communication engineering, and road and bridge engineering, which are further subdivided into multiple work units. The Risk Breakdown Structure (RBS) includes multiple specific risk factors subdivided into five risk factor categories. The WBS-RBS coupling matrix is ​​used to binarize the intersection of each work unit and each risk factor to form a risk mapping relationship.

3. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, Constructing a Social Network Analysis (SNA) model specifically includes: The nodes also include nine types of accident nodes: fire, explosion, falling object, fall from height, vehicle injury, mechanical injury, poisoning and suffocation, collapse and others. The weight of the directed weighted edge is the frequency of the correlation between risk factors in historical accident cases.

4. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, The K2 algorithm is used to optimize the initial network structure, specifically including: The structure of the Social Network Analysis (SNA) model is used as the initial network structure input, and optimization constraints are set. The optimization constraints include: limiting the total number of network nodes, requiring improvements based on the SNA model structure, and ensuring that each risk factor node has a connection relationship. A scoring function based on the Bayesian scoring criterion is adopted, and a hill-climbing method is used as the search strategy. With the goal of maximizing the scoring function, the network topology that maximizes the posterior probability is iteratively searched under the optimization constraints, and redundant edges are removed.

5. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, Dynamic and static nodes are distinguished by analyzing whether various risk factors change with construction progress, personnel status, environmental conditions, or management intensity. Specifically, this includes: Factors whose status changes with the construction process are classified as dynamic nodes, including: human factors such as lack of safety awareness, lack of compliance with rules and regulations, improper operation, failure to conduct gas detection and analysis, failure to implement risk avoidance measures, and inadequate risk identification; material factors such as lack of safety protective equipment, lack of safety warning signs, lack of enclosures, insecure installation of machinery and equipment, and overloading of machinery and equipment; environmental factors such as disorderly stacking of materials, poor nighttime lighting, toxic gases, and adverse weather; and management factors such as inadequate safety education, lack of dedicated safety management personnel, inadequate on-site safety management, and inadequate construction organization. Factor nodes whose state can be considered constant during the construction period are classified as static nodes; Set up a state continuation mechanism for dynamic nodes that perform habitual operations, and a state transition mechanism for dynamic nodes that can be improved in the short term.

6. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, Set multiple time slices, specifically including: The time step is determined by the construction dynamic monitoring cycle. The state of risk factors within each time step is modeled as a time slice. Markov dependencies are established between adjacent time slices through a state transition probability table. The initial time slice is used as the starting state for risk assessment. The number of subsequent time slices is set according to the prediction requirements. Each time slice contains the state probability distribution of all dynamic and static nodes within that period.

7. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, The real-time hazard inspection data acquired at the construction site is used as evidence and input into the dynamic Bayesian risk assessment model, specifically including: A hazard database is pre-established, linking hazard content with corresponding risk factors. During routine safety inspections, if a hazard is discovered, the risk factor corresponding to the hazard is assigned the status of "occurred" at the initial moment; if the hazard is not discovered, it is assigned the status of "not occurred".

8. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, The probability of various accidents occurring within multiple future time units is calculated through forward reasoning, and key disaster-causing factors are diagnosed through sensitivity analysis. Risk levels and early warning information are then output, including: The maximum value among the nine types of accident probabilities is used as the core judgment criterion. The maximum value is compared with the preset risk level threshold to determine the current construction safety risk level, and the corresponding risk acceptance criteria are matched. The level of early warning information push is determined according to the risk level.

9. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, Sensitivity analysis was used to diagnose key disaster-causing factors, specifically including: Nine types of accidents are set as target nodes. The probability importance of each risk factor node to the target accident node is calculated in turn. The influence is represented by the color of the node, and the core risk factors that have the greatest impact on each type of accident and their transmission path are identified.

10. The method for assessing safety risks of oil and gas field surface construction based on dynamic Bayesian networks as described in claim 1, characterized in that, After outputting the risk level and warning information, it also includes: The dynamic Bayesian risk assessment model is embedded into the dual prevention mechanism framework to establish a safety risk management platform that includes a hidden danger database, an intelligent hidden danger identification mechanism, risk classification and control, and a dual early warning system, for hidden danger identification, risk assessment, early warning push and closed-loop handling.

Images