A natural gas pipeline third-party damage accident early warning method based on a dynamic Bayesian network

Abstract

The application provides a natural gas pipeline third-party damage accident early warning method based on a dynamic Bayesian network. The method comprises the following steps: first, third-party damage accident factor identification; second, risk accident scene building; third, accident consequence analysis, an ESD is used to build a sequence diagram about the natural gas pipeline failure consequences caused by the third-party damage; fourth, node probability calculation; fifth, DBN dynamic quantitative modeling coupled with the ESD and the FT; sixth, third-party accident probability dynamic analysis based on the DBN model; and seventh, risk control measures are summarized according to the failure probability analysis results. The method can obtain the pipeline failure risk probability by calling historical data, calculating the occurrence probability of each factor by using a fuzzy comprehensive evaluation method, and reasoning the input data information of the Bayesian network as the data information, and the probability information is combined with the accident consequence analysis method to perform early warning.

Description

Technical Field

[0001] This invention relates to the technical field of early warning methods for natural gas pipeline damage accidents, and in particular to an early warning method for third-party damage accidents to natural gas pipelines based on dynamic Bayesian networks. Background Technology

[0002] With the increasing demand for natural gas in the energy sector, pipeline transportation has undoubtedly replaced other modes of transport, becoming the most efficient and economical way to transport media such as oil and natural gas. my country has already built over 80,000 kilometers of natural gas pipelines. Historical data shows that third-party damage has become one of the main factors threatening the structural integrity of buried pipelines. Therefore, effective early warning systems for natural gas pipeline accidents and the ability to predict their potential consequences can effectively improve the overall operational reliability of pipelines.

[0003] Currently, the main methods used for natural gas pipeline failure risk assessment include the analytic hierarchy process (AHP), Kent score, and fault tree analysis. However, these methods often rely on the subjective opinions of experts, leading to inconsistent and uncertain risk assessment results, which in turn contributes to inaccuracies in pipeline risk assessments. At present, the domestic industry can only achieve preliminary static assessments of pipeline third-party damage risks using traditional methods, failing to adequately address the highly arbitrary, unpredictable, and uncertain characteristics of third-party accidents.

[0004] In recent years, to realize the feasibility of deductive reasoning in accidents, Bayesian networks, as a probabilistic reasoning technique for handling uncertainties, have been frequently used in risk analysis for events such as leaks, fires, explosions, drilling operations, and maintenance activities. Compared with traditional risk assessment methods, the advantages of Bayesian networks are mainly: 1) Bayesian networks are not affected by subjective factors in accident probabilistic reasoning, reducing interference from human factors in the quantitative analysis and calculation part, and minimizing the uncertainty brought about by subjective factors; 2) Bayesian networks can be used to handle datasets with incomplete information. Compared with traditional methods that cannot handle missing or biased data, the mapping method of the Bayesian network model reflects the probabilistic relationship model between data in the entire database. The absence of a certain input data does not interfere with the accurate reasoning of the entire model; 3) Bayesian networks have unique parameter learning and probability update capabilities. When new observational evidence is used as input information, the Bayesian network model can use Bayes' theorem to achieve real-time updates of probability information, ensuring the real-time nature of the output probability calculation results.

[0005] The above analysis clearly demonstrates that Bayesian network models effectively address the challenges of risk assessment in pipeline third-party damage and failure incidents, which is one reason why they have become a primary tool for risk assessment. However, simply analyzing failure incidents is insufficient for fundamentally achieving early warning; it's necessary to consider the temporal progression of risk factors to provide timely warnings. Therefore, dynamic Bayesian networks are chosen. Dynamic Bayesian networks do not refer to networks whose topology or parameters change dynamically, but rather to a network model that combines static Bayesian networks with time factors, resulting in time-series data processing capabilities. Compared to conventional BN models, DBN models offer the following advantages:

[0006] (1) Consider the temporal characteristics of nodes in the model. In the DBN model, the temporal characteristics of node factors are fully considered. The hidden Markov theorem in reliability theory is used to set the state transition equation, so as to calculate the probability of occurrence of the self-loop node in each subsequent time segment, so as to achieve an effective early warning function.

[0007] (2) While considering the temporal sequence of nodes, it retains the characteristics of the conventional BN model. It retains the characteristics of conventional BN in data analysis, probability updates, etc., and can also be combined with FTA model and ESD model to complete the transition from qualitative analysis to quantitative analysis, and complete the analysis of accident causal relationships.

[0008] In recent years, international research on pipeline third-party damage accidents has mainly focused on combining Bayesian networks with other qualitative analysis methods to achieve a shift from qualitative to quantitative assessment. Currently, Bayesian networks are an effective analytical tool for uncertainties in probabilistic systems, widely used in various accident analyses due to their excellent performance in handling uncertainties related to various states, correlations, and faults of system variables. They are often combined with other analytical methods (bowtie models, fault tree models, Delphi techniques, etc.) to enhance their effectiveness in accident risk assessment. However, conventional Bayesian network models cannot explain the time dependence of events and cannot provide early warnings based on the dynamic characteristics of nodes at different time periods. Furthermore, it is necessary to fit the most probable consequences based on existing data to achieve early control and prevention. Therefore, a new risk model is needed for risk warning and consequence prediction analysis of pipeline damage caused by third-party accidents.

[0009] The above background information is intended to aid in understanding the inventive concept and technical solution of this invention. It does not necessarily belong to the prior art of this patent application. In the absence of clear evidence that the above content was disclosed before the filing date of this patent application, the above background information should not be used to evaluate the novelty of the technical solution of this application. Summary of the Invention

[0010] To address at least one of the technical problems mentioned in the background section, the present invention aims to propose a method for third-party damage risk assessment and consequence prediction of natural gas pipelines based on dynamic Bayesian networks. This method establishes a risk warning model, calculates the probability of occurrence of various factors by retrieving historical data and using fuzzy comprehensive evaluation, and uses this as input data for Bayesian networks to infer the pipeline failure risk probability. This probability information, combined with accident consequence analysis methods, enables early warning.

[0011] To achieve the above objectives, the present invention provides the following technical solution.

[0012] A method for early warning of third-party damage accidents in natural gas pipelines based on dynamic Bayesian networks includes:

[0013] The first step is to identify the factors contributing to the third-party damage incidents. This involves identifying the basic events from pipeline monitoring reports, existing data from EGIG and PHMSA, and opinions from experts in the field, and to initially establish a basic factor analysis system.

[0014] The second step is to build risk and incident scenarios, and establish the interrelationships between the basic factors in the first step through Financial Analysis.

[0015] The third step is to analyze the consequences of the accident and use ESD to construct a sequence diagram of the potential failure consequences of third-party sabotage on the natural gas pipeline.

[0016] The fourth step is to calculate the node probability.

[0017] The fifth step is to perform dynamic quantitative modeling of DBN by coupling FT and ESD. The two qualitative analysis models, FT and ESD, are mapped to obtain the corresponding BN model. The self-circulating nodes in the model are determined according to the temporal characteristics of the nodes, and the temporal factors in the model are determined to complete the construction of the DBN model.

[0018] Step 6: Dynamic analysis of third-party accident probability based on the DBN model;

[0019] The seventh step is to summarize risk control measures based on the failure probability analysis results.

[0020] The first step mainly involves two categories: unintentional third-party damage and intentional third-party damage.

[0021] The analysis of intentional third-party damage involves two factors: deliberate sabotage of pipelines and drilling to steal gas.

[0022] Unintentional third-party damage is analyzed from two aspects: natural factors and human factors.

[0023] In the second step, all the basic events (BEs) obtained from the first step are used as the initial events in the FT model, and the pipeline leakage caused by third-party damage is used as the top event. The two are connected by a tree network and intermediate nodes are added at appropriate positions to ensure the coherence of the model.

[0024] In the third step, the failure types considered in the third-party damage mechanism include deformation, surface damage, and fracture.

[0025] The fourth step includes the calculation of basic events and the probability of consequences. The probability of basic events (BEs) required by this method is obtained through two methods: one is to directly search for the probability of occurrence in literature databases such as pipeline investigation reports, EGIG, and PHMSA; the other is to use the fuzzy comprehensive evaluation method to convert the qualitative evaluation of the event by experts into a quantitative evaluation and then calculate the probability of occurrence for BEs that cannot be found.

[0026] Trapezoidal fuzzy numbers (TZFN) are selected to describe the expert judgment language. Then, based on the expert hierarchy weights and certain rules, they are converted into fuzzy numbers, and the probability of occurrence is calculated.

[0027] In the fifth step, the dynamic quantitative modeling of DBN is based on the following assumptions:

[0028] 1) The basic events are independent of each other;

[0029] 2) The sub-nodes of the same factor are conditionally independent;

[0030] 3) Pipeline failure accidents are only affected by the analyzed third-party factors.

[0031] Based on the above assumptions, the Fourier Transform (FT) model is transformed into a Block Native (BN) model using the Bobbio mapping function. In the graph mapping, the basic, intermediate, and top events of the FT are represented as leaf nodes, child nodes, and the root node in the BN, respectively. The connections between events at each level in the fault tree model are represented as arcs in the BN. In the numerical mapping, the occurrence probability of the basic events is assigned to the corresponding leaf nodes as prior probabilities; the probabilities of child nodes and the root node are calculated using the Edit Conditional Probability Table (CPT).

[0032] DBN, as an extension of BN in time series, ensures that any node in a time slice depends only on its parent node in the same time slice and similar nodes in adjacent time slices. The models used in this method are all two-time-slice DBNs, with the number of time slices extended from 2 to N. Their joint probability distribution is expressed as:

[0033]

[0034] Where is the i-th node at time t. For the model The parent node.

[0035] In step six, there are three types of analysis and diagnosis:

[0036] 1) Reliability analysis; By adjusting the input parameters of key factors by ±10%, observe whether the risk assessment fluctuations of the entire system are normal;

[0037] 2) Structural importance analysis: Importance analysis is performed using the probability ratio of variation (RoV), and the RoV is calculated using the prior and posterior probabilities of each leaf node;

[0038] 3) Predictive analysis: Using DBN to predict the probability of basic events and failure consequences in time series.

[0039] 2) In structural importance analysis, during model operation, events with higher prior probabilities indicate higher frequency of occurrence, and events with higher RoV contribute more to pipeline failure accidents. The formula is shown in equation (15):

[0040]

[0041] In the formula, Let φ be the posterior probability of leaf node i. BE (X i ) is the prior probability of leaf node i.

[0042] Compared to existing third-party damage risk assessment methods, this invention considers the dynamic characteristics of basic events over time, comprehensively considers various possible events of accident consequences, and uses analytical methods to predict the probability of the consequences occurring in the next period of time. This is more in line with the current design concept of accident risk early warning. By inputting historical data and calculated probabilities into the model, the most likely failure situation under the current circumstances can be obtained, providing a basis for early warning of pipeline third-party damage accidents and strengthening pipeline safety management.

[0043] Considering the limitation of the Batch Normalization (BN) model—the absence of loops in its causal paths, meaning the data in the model cannot provide early warning based on time sequence—this method employs the Dependency-Based Normalization (DBN) model to overcome the limitations of traditional BN models in terms of time independence. It can capture not only static dependencies between variables but also dependencies along the time dimension. For consequence risk prediction, the ESD model is used for accident scenario propagation analysis. ESD shares some commonality with BN networks in terms of conditional relationships, logic gates, rules, and graphical symbols, and can be analyzed as an additional component of the BN model. Hierarchical Bayesian analysis is performed using historical data to obtain the probability of consequences occurring at corresponding time points, realizing the dynamic characteristics of the overall system from accident cause analysis to consequence prediction over time.

[0044] The present invention also provides the application of the above method in early warning of third-party damage accidents in natural gas pipelines.

[0045] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined to obtain specific implementation methods.

[0046] The raw materials or reagents involved in this invention are all commercially available products, and the operations involved are all routine operations in the field unless otherwise specified.

[0047] The beneficial effects of this invention are as follows:

[0048] This method, based on Financial Theory (FT) and Emergency Distress Analysis (ESD), incorporates time-series information analysis of relevant nodes into the Block Network (BN) model, forming a DBN model for early warning of third-party pipeline accidents. Simultaneously, after an accident, ESD methods are used to analyze possible consequences, and historical data is used for probability analysis, providing a reliable basis for early warning of pipeline accidents.

[0049] This method replaces conventional event tree analysis with ESD (Emergency Detection and Disaster Reduction) in the early warning process of pipeline third-party risk accidents, providing a more complete and comprehensive analysis of accident consequences. Furthermore, the ESD model can be better integrated with subsequent Bayesian Network (BN) model analysis, minimizing data loss. The method couples FTA (Freedom of Action) and ESD methods to build a dynamic Bayesian Network (DBN) model for third-party damage accidents in natural gas pipelines. It fully considers the time-related temporal factors in third-party damage, uses the state transition equation in Hidden Markov Models to predict the probability of occurrence, and uses this as the basis for the model's early warning function.

[0050] The present invention adopts the above-mentioned technical solution to achieve the above objectives, which makes up for the shortcomings of the prior art, is reasonably designed, and is easy to operate. Attached Figure Description

[0051] To make the above and / or other objects, features, advantages and examples of the present invention more apparent and understandable, the accompanying drawings are described below:

[0052] Figure 1 This is a flowchart of a method for early warning of third-party damage to natural gas pipelines;

[0053] Figure 2 is a schematic diagram of the accident tree for third-party damage to natural gas pipelines (Figure 2(a) is the accident tree for third-party damage to natural gas pipelines; Figure 2(b) is the accident tree for intentional damage caused by drilling and gas theft; Figure 2(c) is the accident tree for unintentional damage caused by natural factors; Figure 2(d) is the accident tree for unintentional damage caused by human factors; Figure 2(e) is the accident tree for human factors caused by external engineering activities; Figure 2(f) is the accident tree for human factors caused by deficiencies in management systems).

[0054] Figure 3 This is a schematic diagram of an ESD model illustrating the consequences of third-party damage to natural gas pipelines.

[0055] Figure 4 It is a mapping rule that maps the Fourier Transform (FT) to a Batch Normalization (BN) model;

[0056] Figure 5 It is a DBN with dual time slices;

[0057] Figure 6 It is the DBN model for the risk of third-party damage to natural gas pipelines. Detailed Implementation

[0058] Those skilled in the art can refer to the content of this document and appropriately replace and / or modify the process parameters to achieve the desired results. However, it should be particularly noted that all similar replacements and / or modifications are obvious to those skilled in the art and are considered to be included in this invention. The products and preparation methods described in this invention have been described through preferred examples, and those skilled in the art can obviously modify or appropriately change and combine the products and preparation methods described herein without departing from the content, spirit, and scope of this invention to realize and apply the technology of this invention.

[0059] Unless otherwise specified, the materials, methods, and examples described herein are exemplary and not limiting. While similar or equivalent methods and materials can be used to implement or test the invention, suitable methods and materials are described herein.

[0060] The key terms included in this application include those listed below.

[0061] Third-party sabotage: refers to actions by personnel unrelated to pipeline operation that directly cause accidental damage to oil and gas pipeline facilities, such as the use of mobile equipment, tools, vehicles, or intentional damage.

[0062] Risk assessment is the process of quantitatively evaluating the probability of an event's impact and loss on the surrounding environment, residents' lives, and property before or after the event occurs (but before the event concludes). In other words, risk assessment quantifies the likelihood of an event or event causing impact or loss. In this paper, risk assessment refers to identifying third-party hazards in a natural gas pipeline leak and inferring the risk level of the pipeline failure accident based on the probability of each hazard.

[0063] FTA: Fault Tree Analysis, is a method that identifies the hazards of a system by constructing a tree from top to bottom and connecting events according to their relationships, using a graphical path.

[0064] ESD: Event Sequence Diagram, which uses events, conditions, logic gates, parameters, constraints, rules, and other elements, along with graphical symbols, to represent the occurrence of a fault and the possible development of its impact.

[0065] FCEM: Fuzzy Comprehension Evaluation Method. Based on the membership theory of fuzzy mathematics, it transforms qualitative evaluation into quantitative evaluation, that is, it uses fuzzy mathematics to give an overall evaluation of an object constrained by multiple factors.

[0066] HBA: Hierarchical Bayesian analysis, a technique for analyzing the probability of risk, used to address uncertainty caused by a lack of or no data.

[0067] DBN: Dynamic Bayesian Network, is a Bayesian network that links different variables together with adjacent time steps. This is often referred to as a "two-time-slice" Bayesian network.

[0068] The present invention will be described in detail below with reference to specific embodiments.

[0069] Figure 1 This is a flowchart of a method for early warning of third-party damage to natural gas pipelines.

[0070] like Figure 1 As shown in the figure, this embodiment provides a method for early warning of third-party damage accidents to natural gas pipelines, and its detailed steps are as follows.

[0071] The first step is to identify the factors that could cause the incident to be sabotaged by a third party.

[0072] The first step aims to identify as many third-party sabotage risk factors as possible that could lead to pipeline accidents. In this step, these factors are categorized into intentional sabotage and unintentional sabotage from the perspective of third-party sabotage of natural gas pipelines.

[0073] This method primarily identifies basic events from existing data such as pipeline monitoring reports, EGIG (Emergency Data Organization), and PHMSA (Pipeline and Dangerous Goods Administration), as well as opinions from field experts. Based on the above analytical perspectives and data sources, a preliminary basic factor analysis system based on intentional and unintentional damage is established. The specific contents of each factor are as follows:

[0074] (1) Analysis of factors of intentional damage: Intentional damage refers to the behavior of people damaging pipelines out of their own subjective consciousness. We will analyze the factors from two aspects: intentional damage to pipelines and drilling to steal gas.

[0075] a) Deliberate sabotage of pipelines: The main influencing factors considered in this method for deliberate sabotage incidents include social relations, pipeline management level, and public security conditions along the pipeline route;

[0076] b) Drilling to steal gas: Compared with intentional sabotage, drilling to steal gas also involves the intention of people to seek economic benefits. Therefore, the behavior of drilling to steal gas needs to be analyzed from three factors: legal management, daily inspection, and public education.

[0077] (2) Analysis of Unintentional Damage Factors: This method explains unintentional damage as an act where the damage is not intentional or proactive, but rather an unconscious act that exceeds the pipeline's capacity, leading to a failure. Therefore, it analyzes the issue from two perspectives: natural factors and human factors.

[0078] a) Natural factors: In this method, natural factors can be understood as two aspects: the pipeline's own factors and the surrounding natural environment factors. Among them, the pipeline's own factors refer to the annual changes in the pipeline's service life, wall thickness, and surrounding soil cover as the service life increases;

[0079] b) Human Factors: In cases of unintentional sabotage, human factors mainly refer to situations where the factor requires employee participation during initial construction or later operation and maintenance, or where human actions may lead to the factor being damaged or unable to function properly. This assessment method primarily includes three aspects: pipeline protection warnings, external engineering activities, and deficiencies in management systems.

[0080] The second step is to build risk and incident scenarios.

[0081] Fault Tree (FT) is a commonly used model for qualitative analysis. It effectively connects accidents with the various factors that cause them, using a tree diagram to describe the logical relationships between potential accidents and their causes. In this method, all basic events (BEs) analyzed in the first step are used as the initial events in the FT model, and third-party sabotage leading to pipeline leakage is used as the top event. The two are connected in a tree network, with intermediate nodes added at appropriate locations to ensure model coherence. The FT model constructed from the basic events of a third-party sabotage accident is shown in Figure 2, including... Figure 2(a) , 2(b) Table 1 shows the meanings of the accident tree symbols for third-party damage to natural gas pipelines: 2(c), 2(d), 2(e), 2(f);

[0082] Table 1. Meaning of symbols in the accident tree for third-party damage to natural gas pipelines.

[0083]

[0084]

[0085] The third step is to analyze the consequences of the accident.

[0086] To identify the potential consequences of pipeline damage caused by third-party sabotage, an ESD model was developed to describe the scenario propagation process from the onset of pipeline failure. Natural gas pipelines are susceptible to interference from third-party factors. Considering the forms of third-party sabotage, this method considers three failure types: deformation, surface damage, and fracture. Deformation is the least severe consequence, causing structural damage to the natural gas pipeline. In this failure state, the pipeline integrity can still be maintained, and a natural gas leak will not be induced. Surface damage does not affect the pipeline's operation itself but damages the anti-corrosion materials on the pipeline surface, eventually leading to corrosion leaks over a long period. Fracture is the most serious failure, directly causing pipeline integrity failure and resulting in a gas leak. In summary, it is clear that both surface damage and fracture accidents will lead to gas leakage and safety hazards, posing a fire and explosion risk when exposed to surrounding sparks or ignition sources. The ESD analysis of third-party sabotage in natural gas pipelines is as follows: Figure 3 As shown.

[0087] The fourth step includes calculating the probability of basic events and consequences.

[0088] Third-party damage risk assessment of pipelines requires event probabilities. However, these data are often limited and incomplete. This method obtains the probabilities of the basic events (BEs) required by the proposed approach using two methods: first, by directly searching for occurrence probabilities in pipeline survey reports, EGIG, PHMSA, and other literature databases; second, for BEs for which no occurrence probabilities were found, a fuzzy comprehensive evaluation method is used to convert the qualitative evaluations of the events by experts into quantitative evaluations, thereby calculating the occurrence probabilities.

[0089] A brief description of the fuzzy comprehensive evaluation method is as follows: The fuzzy set of the probability space is represented by a fuzzy number between 0 and 1, which can be assigned the probability of an event. Qualitative knowledge or judgment is transformed into quantitative numerical reasoning, and fuzzy numbers are used to dilute the fuzziness and subjectivity of expert judgment language. Among fuzzy numbers, triangular fuzzy numbers (TFN) and trapezoidal fuzzy numbers (TZFN) are more effective because they can generate the failure probability of basic events based on the scores of the evaluation set. This paper chooses trapezoidal fuzzy numbers (TZFN) to describe expert judgment language. First, the expert evaluation expressions for basic events are divided into Very Low (VL), Low (L), Fairly Low (FL), Medium (M), Fairly High (FH), High (H), and Very High (VH). The meaning and specification scale of the fuzzy number set are shown in Table 2.

[0090] Table 2. Scale of Fuzzy Number Sets

[0091]

[0092] Then, based on the expert level weights and certain rules, the results are converted into fuzzy numbers. Considering that the opinions of different experts have different weights, a weighted average method is used to summarize the judgments of multiple experts, as shown in equation (1).

[0093]

[0094] Among them, P i It is the aggregate fuzzy number of input event i; P i,j W represents the fuzzy number of expert j's input for event i; j is the weight of expert j, and m is the number of experts.

[0095] The evaluation opinions are comprehensively processed by the fuzzy set cut-off set to obtain the relation function of the fuzzy number W corresponding to each fuzzy language, as shown in equation (2):

[0096]

[0097] Then, the fuzzy probability score (FPS) of the aggregated fuzzy number is obtained by calculating the scores of the left and right fuzzy sets, as shown in equations (3), (4), and (5).

[0098] FPS Left =sup x [f w (x)∧f max (x)] (3)

[0099] FPS Right =sup x [f w (x)∧f min (x)] (4)

[0100] FPS(P i ) = [FPS Right (P i +1-FPS Left (P i )] / 2 (5)

[0101] Finally, FPS is converted into failure probability using equations (6) and (7) for further quantitative analysis.

[0102] PF i =1 / 10 k (6)

[0103] k = 2.301 × [(1-FPS) / FPS] 1 / 3 (7)

[0104] The above is the basic event probability algorithm in the FT model. Similarly, the probability of occurrence of accident consequences also needs to be calculated. Considering that there are many reliable sources of historical accident data, the HBA method is chosen for calculation.

[0105] In the HBA framework, three main parameters γ, α, and β are constructed, where parameters α and β are called hyperparameters and are recorded as h(γ|α,β), which follow a diffusion or non-informative distribution; the uncertain parameter γ follows a generalized distribution. The above is the first-stage prior calculation in the HBA framework.

[0106] The second-stage distribution calculation mainly addresses the uncertainty of parameters α and β. The second-stage distribution is updated using common data D from different sources, and the posterior analysis of the hyperparameters is obtained, as shown in Equation (8).

[0107]

[0108] In the formula, h(α,β|D) is the posterior probability of the hyperparameter, and the likelihood function of l(D|α,β) can be expressed as equation (9):

[0109] l(D|α,β)=∫l(D|γ)h(γ|α,β)dγ (9)

[0110] After obtaining the posterior probability distributions of α and β, the posterior distribution h1 of γ is updated by calculating the average of α and β, as shown in equation (10):

[0111] h1(γ|D)=∫∫h(γ|α,β)h(α,β|D)dαdβ (10)

[0112] When more new observational information D * Once input into the HBA analysis framework, h1(γ|D) can be further rewritten as h1(γ|D,D) according to Bayes' theorem. * )

[0113]

[0114] The fifth step is to perform dynamic quantitative modeling of DBN that couples FT and ESD.

[0115] Bayesian networks (BNs) are increasingly used to handle probabilistic problems and in reliability modeling of systems based on uncertainty knowledge. Compared with traditional methods such as fault trees (FT), event trees (ET), and bowtie models (BT), Bayesian networks (BNs) take into account the conditional dependencies between factors and describe the relationships between nodes. A significant advantage of Bayesian networks (BNs) is that the conditional probabilities assigned to nodes throughout the model can be updated based on the input of new observation data. BNs utilize the "d-separation" criterion and the chain rule for quantitative analysis. According to the "d-separation" criterion, all leaf nodes are conditionally independent, while other nodes depend on their direct parent nodes. Based on conditional independence and the chain rule, the variables included in the BN model are U = {A1, A2, ..., A...}. n The joint probability distribution P(U) of} is

[0116]

[0117] Among them, Pa(A) i ) is a variable U = {A1, A2, ..., A n The parent set of}.

[0118] The primary application of Batch Normalization (BN) in incident analysis is as an inference engine used to update the prior probability of an event given new information, known as evidence e. When new evidence is observed, the posterior probability of a variable is obtained:

[0119]

[0120] Equation (13) can be used for both probability prediction and probability update. In predictive analysis, the conditional probability of the form P(accident|event) is calculated, representing the probability of a specific accident occurring given that a major event has occurred or not occurred. On the other hand, in update analysis, the form of P(event|accident) is evaluated to show the probability of a specific event occurring given that an accident has occurred.

[0121] In the Bayesian network-based pipeline third-party damage risk assessment method of this paper, the constructed Bayesian network should be based on the following assumptions:

[0122] 1) The basic events are independent of each other;

[0123] 2) The sub-nodes of the same factor are conditionally independent;

[0124] 3) Pipeline failure accidents are only affected by the analyzed third-party factors.

[0125] Combination Figure 6 To state Hypothesis 2), legal management, public education, and routine patrols all have an impact on the occurrence of borehole gas theft; however, given that borehole gas theft has occurred, legal management has no impact on public education and routine patrols (the same applies to changing nodes, and there is no impact between child nodes); that is, given that gas theft has occurred, the three child nodes below it are independent of each other.

[0126] Based on the above assumptions, the Fourier Transform (FT) model is transformed into a Block Native (BN) model using the Bobbio mapping function. In the graph mapping, the basic, intermediate, and top events of the FT are represented as leaf nodes, child nodes, and the root node in the BN, respectively. The connections between events at each level in the fault tree model are represented as arcs in the BN. In the numerical mapping, the occurrence probability of the basic events is assigned to the corresponding leaf nodes as prior probabilities; the probabilities of child nodes and the root node are calculated using the Edit Conditional Probability Table (CPT). Figure 4 The mapping rules for mapping the Fourier Transform (FT) to a Batch Normalization (BN) model are shown.

[0127] Dynamic Bayesian Networks (DBNs), as a time-series extension of the BN model, can be understood as a generalization of Markov processes. In a DBN, any node in a time slice depends only on its parent node in the same time slice and similar nodes in adjacent time slices. The models used in this method are all two-time-slice DBNs, with the number of time slices extended from 2 to N. Their joint probability distribution is expressed as:

[0128]

[0129] Where is the i-th node at time t. For the model The parent node.

[0130] For ease of processing, assume that the DBN model satisfies two conditions:

[0131] 1) The network topology does not change over time; that is, except for the initial moment, the variables and their probabilistic dependencies are the same at other times.

[0132] 2) It satisfies the first-order Markov condition, that is, given the state at the current moment, the state at the future moment depends only on the current moment and is independent of the state at the past moment.

[0133] After satisfying the above conditions, a dynamic Bayesian network can be viewed as an expansion of a Bayesian network over a time series. The current time slice is t, and the next time slice is t+1. In this method, the key factors affecting natural gas pipeline leakage are identified through ESD model analysis as pipeline rupture (C3) and corrosion perforation (C4). The probability of pipeline leakage due to pipe corrosion caused by pipeline surface protection failure gradually increases over time; therefore, corrosion perforation (C4) in time slice (t+1) is affected by its state in time slice (t). Similarly, natural gas leakage in pipelines is not only affected by C3 and C4 in the same time segment, but also by the state in other time slices (t). A dual-time-slice DBN is formed using these two methods. Figure 5 The results were interpreted by constructing a dual-time-slice DBN for the time-related factors in the consequence analysis.

[0134] The accident consequence analysis (ESD) model also incorporates OR and AND gate conditional logic relationships and input parameters during its construction. Therefore, while preserving these gate logic relationships, the entire model can be effectively integrated with the DBN model. Furthermore, this method utilizes Genie 2.3 software to construct a Bayesian network. The constructed DBN model is shown below. Figure 6 As shown.

[0135] Step 6: Dynamic analysis of third-party accident probability.

[0136] The fifth step involves constructing a complete DBN model of basic events and consequences of third-party damage, followed by dynamic assessment of accident probabilities based on this model. The main analytical and diagnostic approaches employed here are threefold: 1) reliability analysis; 2) structural importance analysis; and 3) predictive analysis.

[0137] 1) Reliability Analysis: When developing a new model, it is necessary to ensure its robustness. This is verified by changing the model's input parameters to allow the output to vary within an appropriate range. This method involves adjusting the input parameters of key factors by ±10% and observing whether the fluctuations in the risk assessment of the entire system are normal.

[0138] 2) Structural Importance Analysis: When updating probabilities in the BN model, knowledge of the top event or outcome event is generally used as evidence. This method uses the Probability Variation Ratio (RoV) for importance analysis, calculating RoV using the prior and posterior probabilities of each leaf node. During model operation, events with higher prior probabilities indicate higher frequency of occurrence, and events with higher RoV contribute more to pipeline failure accidents. The formula is shown in equation (15):

[0139]

[0140] In the formula, Let φ be the posterior probability of leaf node i. BE (X i ) is the prior probability of leaf node i.

[0141] 3) Predictive Analysis: The DBN (Browser-Based Network) is used to predict the probability of basic events and failure consequences in time series data. Based on current data and historical data, the most likely failure consequences and the most probable basic events leading to failure are obtained. Furthermore, the BN network can update its probabilities in real time as new data is input into the system, ensuring the timeliness of the analysis and prediction results.

[0142] Step 7: Risk management measures.

[0143] Based on the assessment and analysis results of step six, and in conjunction with the principles of accident safety early warning, the most effective preventive measures for the current situation are proposed. At the same time, based on DBN's prediction of possible future failure events, targeted early warning and investigation are carried out by comprehensively considering the probability of occurrence and the severity of harm.

[0144] The conventional techniques described in the above embodiments are existing technologies known to those skilled in the art, and therefore will not be described in detail here.

[0145] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for early warning of third-party damage accidents in natural gas pipelines based on dynamic Bayesian networks, characterized in that... include: The first step is to identify the factors contributing to the third-party damage incidents. This involves identifying basic events from pipeline monitoring reports, existing data from the European Gas Pipeline Accident Data Organization (EGIG), the Pipeline and Hazardous Materials Safety Authority (PHMSA), and opinions from experts in the field, and to initially establish a basic factor analysis system. The second step is to build risk and incident scenarios, and establish the interrelationships between the basic factors in the first step using a fault tree analysis diagram (FT). The third step is to analyze the consequences of the accident by using an event sequence diagram (ESD) to construct a sequence diagram of the consequences of the natural gas pipeline failure caused by third-party sabotage. The fourth step is the calculation of node probabilities, including the calculation of basic event and consequence probabilities. The calculation of basic event probabilities is obtained through one of the following two methods: First, the probability of occurrence is obtained by directly searching the pipeline investigation report, EGIG, and PHMSA literature databases; second, for basic events whose probability of occurrence cannot be found, the fuzzy comprehensive evaluation method is used to convert the qualitative evaluation of the event by experts into a quantitative evaluation and then calculate the probability of occurrence. The probability of consequences was calculated using the HBA method; The fifth step is to dynamically and quantitatively model the Fault Tree Analysis Graph (FT) and Event Sequence Graph (ESD) in a coupled dynamic Bayesian Network (DBN). The two qualitative analysis models, FT and ESD, are used to obtain the corresponding Bayesian Network (BN) models based on relevant mapping relationships. The self-looping nodes in the model are determined based on the temporal characteristics of the nodes, and the temporal factors in the model are determined to complete the construction of the dynamic Bayesian Network (DBN) model. Step 6: Dynamic analysis of third-party incident probability based on dynamic Bayesian network (DBN) model; Step 7: Summarize risk control measures based on the failure probability analysis results; This method is based on fault tree analysis graph (FT) and event sequence graph (ESD). It incorporates the temporal information analysis of relevant nodes into the Bayesian network (BN) model to form a dynamic Bayesian network (DBN) model for early warning of pipeline third-party accident risks. At the same time, after an accident, the ESD method is used to analyze the possible consequences and conduct probability analysis based on historical data, providing a basis for early warning of pipeline accidents. Based on the time-related temporal factors in third-party sabotage, the state transition equation in the Hidden Markov Model is used to predict the probability of occurrence, and this is used as the basis to realize the early warning function of the model.

2. The method according to claim 1, characterized in that: The first step mainly divides damage into unintentional third-party damage and intentional third-party damage; Intentional third-party damage is analyzed from two aspects: deliberate sabotage of pipelines and drilling to steal gas; and / or unintentional third-party damage is analyzed from two aspects: natural factors and human factors.

3. The method according to claim 1, characterized in that: In the second step, all the basic events (BEs) obtained from the first step are used as the initial events in the fault tree analysis graph (FT), and the pipeline leakage caused by third-party damage is used as the top event. The two are connected by a tree network, and intermediate nodes are added at appropriate positions to ensure the coherence of the model.

4. The method according to claim 1, characterized in that: In the third step, the failure types considered in the third-party damage mechanism are deformation, surface damage, and fracture.

5. The method according to claim 1, characterized in that: The fourth step includes the calculation of basic events and consequences probabilities. The probabilities of basic events (BEs) required by this method are obtained through two methods: First, by directly searching the pipeline investigation reports, the European Gas Pipeline Accident Data Organization (EGIG), and the Pipeline and Hazardous Materials Safety Authority (PHMSA) literature databases to obtain the probability of occurrence. Second, for basic events (BEs) whose probability of occurrence cannot be found, the fuzzy comprehensive evaluation method is used to convert the qualitative evaluation of the events by experts into a quantitative evaluation and then calculate the probability of occurrence. Trapezoidal fuzzy numbers (TZFN) are selected to describe the expert judgment language, and then converted into fuzzy numbers according to the expert hierarchy weights and certain rules, and further calculated the probability of occurrence. Specifically, the weighted average method is used to summarize the judgments of multiple experts, and the evaluation opinions are comprehensively processed by the fuzzy set cutoff to obtain the relationship function of the fuzzy numbers W corresponding to each fuzzy language. Then, the fuzzy probability score of the aggregated fuzzy number is obtained by calculating the scores of the left and right fuzzy sets.

6. The method according to claim 1, characterized in that: In the fifth step, the dynamic quantitative modeling of the Dynamic Bayesian Network (DBN) is based on the following assumptions: 1) The basic events are independent of each other; 2) The sub-nodes of the same factor are conditionally independent; 3) Pipeline failure accidents are only affected by the analyzed third-party factors; Based on the above assumptions, the fault tree analysis graph (FT) is converted into a Bayesian network (BN) model through the Bobbio mapping function. In the graph mapping, the basic events, intermediate events, and top events of the fault tree analysis graph (FT) are represented as leaf nodes, child nodes, and root nodes in the Bayesian network (BN), respectively. The connections between events at each level in the fault tree analysis graph (FT) are represented as arcs in the Bayesian network (BN). In numerical mapping, the probability of occurrence of basic events is assigned to the corresponding leaf nodes as prior probabilities; for child nodes and root nodes, probabilities are calculated by editing the conditional probability table (CPT).

7. The method according to claim 1, characterized in that: In step six, there are three types of dynamic analysis: 1) Reliability analysis; By adjusting the input parameters of key factors by ±10%, observe whether the risk assessment fluctuations of the entire system are normal; 2) Structural importance analysis: Importance analysis is performed using the Probability Ratio of Variation (RoV). The RoV is calculated using the prior and posterior probabilities of each leaf node. During model operation, events with higher prior probabilities have higher frequency of occurrence, and events with higher RoV contribute more to pipeline failure accidents. The formula is shown in equation (15). In the formula, Let Φ be the posterior probability of leaf node i. BE (X i ) is the prior probability of leaf node i; 3) Predictive analysis: Using dynamic Bayesian networks (DBN) to predict the probability of basic events and failure consequences in time series.

8. The application of the method according to any one of claims 1-7 in early warning of third-party damage accidents in natural gas pipelines.

Images