A natural gas purification production safety quantitative evaluation and decision-making method based on root cause analysis and cascade failure propagation

Abstract

The present application relates to natural gas purification technical field, especially in a kind of based on root cause analysis and cascading failure propagation natural gas purification production safety quantification evaluation and decision-making method.The present application first adopts nuclear independent component analysis and neural network granger causality analysis to construct the directed topology network consisting of initial risk node, propagation node and termination node;Then the node relationship is classified as causal relationship, "and" relationship and "or" relationship, respectively calculate cascading failure probability, and the safety event probability is sorted to determine the priority disposal target;Further construct the comprehensive decision-making utility function of fusion node uniqueness score and repair cost score, generate risk blocking sequence;When blocking is invalid, the maximum mutual information coefficient and transmission entropy are used to adaptively update local topology network and dynamically optimize blocking sequence.The present application realizes the cascading failure evaluation and closed-loop risk control of natural gas purification production whole process unit, significantly improves the real-time performance and accuracy of safety decision-making.

Description

Technical Field

[0001] This invention relates to the field of natural gas purification technology, and in particular to a method for quantitative assessment and decision-making on the safety of natural gas purification production based on root cause analysis and cascading failure propagation. Background Technology

[0002] With the rapid development of my country's natural gas industry, large-scale natural gas purification plants, as the core hubs of natural gas deep processing, are showing a trend of continuous expansion in scale, increasingly complex structure, and increasingly precise functional coupling between equipment components. Core process units cover raw gas pretreatment, desulfurization, dehydration, cryogenic separation, sulfur recovery, acid water units, and flare units, etc., and each unit forms an organic whole through the close connection of materials, energy, and information. Due to the special nature of natural gas purification processes, the system involves hazardous substances such as H2S and methane, and often operates under high temperature and high pressure conditions, which causes the stability of the purification plant to drop sharply when facing risks. Local failure of any process unit or equipment may spread and diffuse to other related equipment and units through the tightly interwoven material flow (such as amine liquid circulation and raw gas transportation) and energy transfer (such as temperature and pressure conduction) network, forming a cascading failure.

[0003] The production process of a natural gas purification plant can be viewed as a topological network composed of a large number of heterogeneous nodes and edges. Nodes represent key equipment (absorption tower, regeneration tower, molecular sieve dehydrator, sulfur recovery reactor, etc.), process units (desulfurization unit, dehydration unit, etc.), and key process variables (H2S concentration, amine circulation volume, regeneration tower temperature, etc.) within the purification plant. Edges represent physical connections (material pipelines, energy transmission pipelines) or causal relationships between nodes (such as excessive H2S concentration leading to equipment corrosion, or amine degradation leading to decreased desulfurization efficiency).

[0004] However, traditional safety assessment methods, including fault trees and event trees, typically employ relatively simple topological network structures, describing only a finite set of objects within a single unit (such as a desulfurization unit) or single equipment within a purification plant. They lack the ability to model the entire natural gas purification plant process and all units as a whole, failing to capture the propagation process of cascading failures across units. Meanwhile, while Bayesian networks and their variants can perform network modeling of the overall purification plant structure and use Bayes' theorem for probabilistic reasoning to complete safety assessments, their construction heavily relies on expert knowledge in the natural gas purification field and historical accident records during plant operation. Due to the complexity of natural gas purification processes, the quality and processing load of raw gas vary among different purification plants, exacerbating the inherent subjectivity of expert knowledge. Furthermore, some purification plants have incomplete accident records and missing key failure data, often hindering the accurate modeling of natural gas purification plants using Bayesian networks and their variants. This impacts the accuracy of subsequent risk assessments, making it difficult to accurately pinpoint the root causes and propagation paths of cascading failures. Consequently, they fail to meet the purification plant's requirements for precise and real-time safety assessments and cannot effectively prevent cascading failure accidents across units.

[0005] Meanwhile, the analysis results provided by traditional security assessment methods are often independent of the closed-loop operation of the system. The corresponding security protection is mostly limited to the level of ex-post assessment, making it difficult to respond in a timely manner to the dynamic impact of system status or external environmental interference, which to some extent restricts the real-time and proactive nature of system security protection. Summary of the Invention

[0006] This invention discloses a quantitative assessment and decision-making method for the safety of natural gas purification production based on root cause analysis and cascade failure propagation. The specific method is as follows: Collect process variable data for the entire process unit of natural gas purification, desulfurization, dehydration, and sulfur recovery, and screen the main contributing variables; Granger causality analysis based on neural networks is used to identify causal relationships between variables and generate a causal adjacency matrix; Select initial risk nodes, propagation nodes, and termination nodes, and construct a directed topological network using the causal relationships in the causal adjacency matrix as the edge set; The node relationships in the directed topology network are classified into causal relationships, AND relationships, and OR relationships, and the cascade failure probability of successor nodes under each type of relationship is calculated. The safety events are sorted according to the calculated probability of occurrence. The safety event with the highest probability is selected as the target risk event to be dealt with first. The causal propagation path from the initial risk node to the terminal node corresponding to the target risk event is extracted. A comprehensive decision utility scoring function is constructed for each node on the propagation path. The candidate nodes are sorted according to the comprehensive decision utility score to generate a risk blocking sequence, which guides the operators to implement defensive measures. When the risk blocking sequence cannot effectively cut off the risk propagation chain, the correlation between the failed node and the process unit outside the risk propagation chain is calculated using the maximum mutual information coefficient and the propagation entropy. Potential risk nodes are identified and the causal direction is verified. The newly identified potential related nodes and their causal edges are incorporated into the existing topology network. The comprehensive decision utility score is recalculated and an updated risk blocking sequence is generated.

[0007] Furthermore, the main contributing variables are screened, and the specific methods are as follows: Kernel independent component analysis is used to screen the main contributing variables for a variety of given failures. The intersection of the screened set of main contributing variables with risk-related process variables is then used to determine if there is an intersection. If there is an intersection, it indicates that the failure is related to the risk event.

[0008] Furthermore, the initial risk nodes, propagation nodes, and termination nodes are screened in the following ways: Variables with zero in-degree and non-zero out-degree in the adjacency matrix obtained by Granger causality analysis based on neural networks are identified as root cause variables and designated as initial risk nodes. Use key variables related to security incidents as termination nodes; Variables in the adjacency matrix that have non-zero in-degree and non-zero out-degree and are not terminal nodes are used as propagation nodes.

[0009] Furthermore, the formal representation of the propagation node set is as follows: ; in, Represents the set of all failures; Represents belonging to a set A certain type of failure, Represents the elements in the adjacency matrix; Represents the order of the adjacency matrix; Represents the set of terminating nodes.

[0010] Furthermore, the node relationships in the directed topology network are classified into causal relationships, AND relationships, and OR relationships. The cascading failure probability of successor nodes under each type of relationship is calculated, as follows: For causal relationships, the formula for calculating the probability of cascading failure is: ; in, It means that by exist The cascading risks triggered by this location; Indicates in Independent risks arising from the premises; and Represents respectively in The probability of triggering cascading risks and independent risks; Then represents a node Cascade risk propagation The conditional probability at that location; For the "OR" relationship, the formula for calculating the cascade failure probability is: ; For the AND relationship, the formula for calculating the cascade failure probability is: .

[0011] Furthermore, a comprehensive decision-making utility scoring function is constructed, and the specific method is as follows: For each node in the propagation path Construct a comprehensive decision utility scoring function: ; in, Score the uniqueness of the nodes. Score the maintenance cost; and This represents the weighting parameters for uniqueness scoring and maintenance cost scoring, and satisfies... + =1.

[0012] Furthermore, the uniqueness score of a node is calculated using the following formula: ; in, Indicates the presence of nodes The total number of paths; when When, it indicates that it contains nodes. If there is exactly one, the specificity is highest, and the node's uniqueness score is [value missing]. ; The maintenance cost score is calculated using the following formula: ; in, It is a node The estimated maintenance time for the corresponding equipment, measured in seconds; A manually set sensitivity parameter is used to control the rate at which the maintenance cost score decays due to increased maintenance time. It is the average of the manually set estimated maintenance time.

[0013] Furthermore, the correlation between the risk and process units outside the risk propagation chain is calculated using the maximum mutual information coefficient and the propagation entropy to identify potential risk nodes and verify the causal direction. The specific method is as follows: Using the maximum mutual information coefficient Calculate the failure node With other process unit nodes The correlation between them is expressed by the formula: ; in, Representative node and Mutual information between them; and Representing nodes respectively and The number of rows and columns when performing network partitioning; An upper limit constraint on the total number of grid divisions; Transmission entropy is used to verify the causal direction of potential associations by comparing positive time delays. With reverse time delay The numerical difference determines the direction of risk propagation; if the positive time delay is less than the negative time delay, the direction of risk propagation is from the failure node to the potential associated node, forming a directional causal edge.

[0014] Furthermore, an updated risk blocking sequence is generated, as follows: After incorporating the newly identified potential associated nodes and their causal edges into the existing topology network, the comprehensive decision utility score involving the associated nodes is recalculated, and a brand-new risk blocking sequence is generated by sorting the updated scores in descending order.

[0015] This invention also discloses a quantitative assessment and decision-making system for the safety of natural gas purification production based on root cause analysis and cascade failure propagation, characterized by comprising: The data acquisition module is used to collect process variable data for the entire process unit of natural gas purification, desulfurization, dehydration, and sulfur recovery. The topology network construction module is used to screen the main contributing variables. It uses Granger causal analysis based on neural networks to identify causal relationships and generate a causal adjacency matrix. It screens initial risk nodes, propagation nodes and termination nodes, and constructs a directed topology network with causal relationships as the edge set. The risk assessment module is used to classify node relationships into causal relationships, AND relationships, and OR relationships, and calculate the cascading failure probability of subsequent nodes under each type of relationship. The decision-making module is used to select priority target risk events based on the probability of occurrence of each security incident, extract causal propagation paths, construct a comprehensive decision utility scoring function, and generate risk blocking sequences. The adaptive update module is used to identify potential risk nodes and verify causal direction by using the maximum mutual information coefficient and transmission entropy when the risk blocking sequence cannot effectively cut off the risk propagation chain, thereby realizing the adaptive adjustment of the local topology network and the dynamic update of the blocking sequence.

[0016] Due to the adoption of the above technical solutions, this application has the following beneficial effects: 1. This application, from the perspective of failure of natural gas purification system, identifies the root causes, propagation variables, termination variables and their causal relationships in the risk propagation process based on root cause analysis; 2. Based on the identified variables and causal relationships, a topological network covering all process units of clean production was constructed as an accident model. Based on the cascading failure model, the propagation path of failure was accurately captured, and the impact of the cumulative related failures on the safety of clean production was evaluated probabilistically. 3. It can accurately present different failure propagation paths and corresponding risk accumulation measures in natural gas purification and production, and complete the quantitative ranking of various potential safety events; 4. Deeply integrate the uniqueness of nodes in the propagation path of high-risk events with equipment maintenance costs, construct a comprehensive utility scoring function, identify key nodes, and formulate corresponding risk blocking sequences accordingly to guide front-line operators in implementing targeted defense measures and emergency responses to risks. 5. Based on decision feedback, the adaptive adjustment and dynamic updating of the topological causal network and decision sequence were further realized, and an integrated closed-loop risk management system of monitoring-evaluation-decision-feedback-optimization was constructed.

[0017] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0018] The accompanying drawings of this invention are described below.

[0019] Figure 1 This is a schematic diagram of the overall process of an embodiment.

[0020] Figure 2This is a visual illustration of the risk evolution and accumulation in an example. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] A quantitative assessment and decision-making method for safety in natural gas purification production based on root cause analysis and cascade failure propagation, such as Figure 1 As shown, the specific steps are as follows: S1. Construction of a topological network based on Granger causal analysis of neural networks Assume that the process variables in the natural gas purification and production process can be represented as: Among them, security incidents Related process variables The filtered node, and the terminal node, is represented as: .

[0023] For a given kernel-independent component analysis Type of failure, i.e. Expand the screening of major contributing variables. For specific faults... In other words, the set of main contributing variables obtained after screening can be represented as: To further ensure that failures are related to safety incidents during the purification process (such as absorption tower leaks, abnormal regeneration tower temperatures, excessive H2S concentrations, molecular sieve dehydration failures, and abnormal flare venting), further screening of failures is conducted. Main contributing variables Risk-related process variables The correlation. If the main contributing variable... There is overlap in the process variables related to risk. ,Right now This suggests that the failure is related to a risk event during the natural gas purification and production process. Further investigation is needed. Root cause analysis is performed on the failures contained therein.

[0024] For failure Root cause analysis was performed using Granger causality analysis with neural networks. The adjacency matrix obtained from the root cause analysis can be represented as: .in, This represents the total number of related variables corresponding to this failure. This represents an element in the adjacency matrix, usually 0 or 1.

[0025] Based on adjacency matrix Failure The root cause variable can be identified and represented as:

[0026] in, Representing variables Unaffected by any other variables, Then it means A variable that can influence at least one other variable through a causal relationship is ultimately considered a root cause variable.

[0027] By taking the union of the root cause variables corresponding to all failures, we obtain the complete set of root cause variables for all risk-related failures:

[0028] These variables can be represented as nodes as follows:

[0029] The obtained root cause variable is designated as the initial risk node. Key variables related to risk events will be used as termination points. The remaining variables in the adjacency matrix constitute the propagation nodes. The remaining variables in the adjacency matrix can be considered as having a non-zero in-degree in the adjacency matrix, i.e., This indicates that at least one predecessor node can have a cascading effect on it. Secondly, the propagating node also needs to have a non-zero out-degree, i.e., This indicates that it is able to further propagate cascading risks and ensure their connection to the termination node. Mutually exclusive. Therefore, propagation nodes... It can be formally represented as:

[0030] Ultimately, from the initial risk node Termination Node and propagation nodes Together they form a set of nodes. Building upon this, the causal relationships are further extracted from the adjacency matrix and used as edges in the topological network. For any two nodes... and In other words, if there exists in the adjacency matrix This means right A causal relationship exists. This causal relationship can be considered as an edge set in a topological network. This allows for the construction of a topology network for risk assessment. .

[0031] S2. Risk assessment based on topology networks In a directed topological network constructed based on Granger causality analysis of a neural network, the relationships between nodes determine the risk propagation pattern. Therefore, the node relationships in the topological network are first categorized into three different types: causal, AND, and OR relationships. For each of these three types of node relationships, the cascading failure probability is calculated. Specifically, if the node... and It is a causal relationship, and yes If the predecessor nodes are such that their corresponding cascade failure probabilities can be expressed as:

[0032] in, It means that by exist The cascading risks triggered by this location; Indicates in Independent risks arising from the premises; and Represents respectively in The probability of triggering cascading risks and independent risks; Then represents a node Cascade risk propagation The conditional probability at a given point. The conditional probability is determined by the node. and The maximum correlation coefficient between them is determined.

[0033] If node and Connect to nodes via an "OR" relationship. This means that as long as the node and The failure of any node can trigger a node. The cascading failure occurs at this point. According to the union formula in probability theory, the probability of the union of two events equals the sum of their individual probabilities minus the probability of their intersection: If the two events are independent, the union formula can be further written as: Given the nodes and Triggered separately The cascading failure process can be viewed as two independent events. Therefore, the node The cascading failure probability corresponding to a location can be written as:

[0034] At the same time, considering and They can be further represented as follows: and Therefore, node The cascading failure probability corresponding to the location is further written as:

[0035] Extending this logic, let's assume that there exists in the topological network. Each predecessor node triggers the node via an "OR" relationship. ,but The cascading failure probability can be expressed as:

[0036] If node and Connect to nodes via an "AND" relationship. This means that if and only if and Triggering is only possible when both failures occur simultaneously. The cascading failure occurs at the point. According to the intersection formula in probability theory, the probability of two events occurring simultaneously, under the assumption of independence, is equal to the product of their individual probabilities. Given the nodes and Triggered separately The cascading failure process can be viewed as two independent events. Therefore, the node The cascading failure probability arising from the AND relationship can be written as:

[0037] S3. Decision-making and implementation based on risk assessment Based on the constructed topology network and cascading failure model, when a node When an anomaly occurs, a search is performed within the network topology to identify all potential security events that could be triggered. , and The termination risk nodes corresponding to each potential security incident are as follows: , as well as .in, Using the cascading failure model described above, the probability of different safety events occurring is calculated as follows: , and .

[0038] Potential security incidents are ranked according to their probability of occurrence: Assuming Therefore, the security event corresponding to the highest probability is selected. As a priority target risk event, it is extracted from the topology network by the initial node. Causal propagation path to target risk .

[0039] For each node in the propagation path Construct a comprehensive decision utility scoring function This function uses a weighted fusion method to comprehensively consider the uniqueness score of nodes. Maintenance cost rating The uniqueness score of a node is used to quantify the degree to which a node exists only on the current high-risk path and is not coupled to other risk propagation paths.

[0040] in, Indicates the presence of nodes The total number of paths. When When, it indicates that it contains nodes. If there is exactly one, the specificity is highest, and the node's uniqueness score is [value missing]. And when it contains nodes As the number of paths increases, the uniqueness score of a node decreases.

[0041] Maintenance cost rating Then, based on the estimated maintenance time of the node device:

[0042] in, It is a node The estimated maintenance time for the corresponding equipment, measured in seconds; A manually set sensitivity parameter is used to control the rate at which the maintenance cost score decays due to increased maintenance time. It is the average of the manually set estimated maintenance time.

[0043] For nodes Comprehensive decision utility function It can be represented as:

[0044] in, and This represents the weighting parameters for uniqueness scoring and maintenance cost scoring, and satisfies... + =1.

[0045] Finally, based on the comprehensive decision utility scoring function Sort the candidate nodes according to their scores. For example, the obtained node scores are sorted as follows: This can be used to generate risk blocking sequences, guiding operators at critical junctures. and Precise defensive measures were implemented in sequence at various locations to cut off the risk transmission chain and ensure the safe and stable operation of natural purification production.

[0046] S4. Adaptive adjustment of local topology network based on decision feedback and dynamic updating of blocking sequences When risk-blocking sequences fail to effectively break the risk transmission chain, further measures will be taken at the nodes along the risk transmission chain. Centered on a specific node, the local topology of the relevant parts is adaptively updated. For example, taking a failed node as an example... For example, the maximum mutual information coefficient and transmission entropy are used as core quantitative indicators to calculate their correlation with other process units outside the risk propagation chain. For example, node points... With other potential process unit nodes Maximum mutual information coefficient between It can be written as:

[0047] in, Representative node and Mutual information between them; and Representing nodes respectively and The number of rows and columns when performing network partitioning; The upper limit of the total number of grid divisions.

[0048] When node With failure node When the maximum mutual information value between nodes exceeds 0.6, These can be considered potential risk nodes. To further ensure a substantial causal relationship between potential risk nodes and the risk propagation chain, the potential correlation was re-verified using transfer entropy, and the positive time delay was calculated by comparison. With reverse time delay The numerical differences between them are used to verify the directionality of potential causal relationships. If This indicates that the risk propagation direction is from the failure node. Pointing to potential related nodes And form directional causal edges. Conversely, the same applies. Subsequently, the newly identified potential associated nodes and their causal edges are integrated into the existing network in real time to achieve adaptive adjustment of the local topology, and the comprehensive decision utility decision function involving associated nodes is recalculated. Based on the updated scores, a new risk blocking sequence is generated, such as... This is to achieve closed-loop management of risks in natural gas purification and production.

[0049] In this embodiment, risk evolution and accumulation are as follows: Figure 2 As shown.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A natural gas purification production safety quantitative evaluation and decision method based on root cause analysis and cascade failure propagation, characterized in that, The specific method is as follows: Collect process variable data for the entire process unit of natural gas purification, desulfurization, dehydration, and sulfur recovery, and screen the main contributing variables; Granger causality analysis based on neural networks is used to identify causal relationships between variables and generate a causal adjacency matrix; Select initial risk nodes, propagation nodes, and termination nodes, and construct a directed topological network using the causal relationships in the causal adjacency matrix as the edge set; The node relationships in the directed topology network are classified into causal relationships, AND relationships, and OR relationships, and the cascade failure probability of successor nodes under each type of relationship is calculated. The safety events are sorted according to the calculated probability of occurrence. The safety event with the highest probability is selected as the target risk event to be dealt with first. The causal propagation path from the initial risk node to the terminal node corresponding to the target risk event is extracted. A comprehensive decision utility scoring function is constructed for each node on the propagation path. The candidate nodes are sorted according to the comprehensive decision utility score to generate a risk blocking sequence, which guides the operators to implement defensive measures. When the risk blocking sequence cannot effectively cut off the risk propagation chain, the correlation between the failed node and the process unit outside the risk propagation chain is calculated using the maximum mutual information coefficient and the propagation entropy. Potential risk nodes are identified and the causal direction is verified. The newly identified potential related nodes and their causal edges are incorporated into the existing topology network. The comprehensive decision utility score is recalculated and an updated risk blocking sequence is generated.

2. The method of claim 1, wherein, The specific method for screening major contributing variables is as follows: Kernel independent component analysis is used to screen the main contributing variables for a variety of given failures. The intersection of the screened set of main contributing variables with risk-related process variables is then used to determine if there is an intersection. If there is an intersection, it indicates that the failure is related to the risk event.

3. The method of claim 1, wherein, The initial risk node, propagation node, and termination node are selected in the following manner: Variables with zero in-degree and non-zero out-degree in the adjacency matrix obtained by Granger causality analysis based on neural networks are identified as root cause variables and designated as initial risk nodes. Use key variables related to security incidents as termination nodes; Variables in the adjacency matrix that have non-zero in-degree and non-zero out-degree and are not terminal nodes are used as propagation nodes.

4. The method of claim 3, wherein, The formal representation of the propagation node set is as follows: ； wherein, represents the set of all failures; represents the set of failures belonging to a certain specific failure, represents an element in the adjacency matrix; represents the order of the adjacency matrix; represents the set of terminal nodes.​ 5. The method according to claim 1, characterized in that, The node relationships in the directed topology network are classified into causal relationships, AND relationships, and OR relationships. The cascading failure probability of successor nodes under each type of relationship is calculated as follows: For causal relationships, the formula for calculating the probability of cascading failure is: ； in, It means that by exist The cascading risks triggered by this location; Indicates in Independent risks arising from the premises; and Represents respectively in The probability of triggering cascading risks and independent risks; Then represents a node Cascade risk propagation The conditional probability at that location; For the "OR" relationship, the formula for calculating the cascade failure probability is: ； For the "AND" relationship, the formula for calculating the cascade failure probability is: 。 6. The method according to claim 1, characterized in that, The comprehensive decision-making utility scoring function is constructed as follows: For each node in the propagation path Construct a comprehensive decision utility scoring function: ； in, Score the uniqueness of the nodes. Score the maintenance cost; and This represents the weighting parameters for uniqueness scoring and maintenance cost scoring, and satisfies... + =1.

7. The method according to claim 6, characterized in that, The uniqueness score of a node is calculated using the following formula: ； in, Indicates the presence of nodes The total number of paths; when When, it indicates that it contains nodes. If there is exactly one, the specificity is highest, and the uniqueness score of the node is [value missing]. ; The maintenance cost score is calculated using the following formula: ； in, It is a node The estimated maintenance time for the corresponding equipment, measured in seconds; A manually set sensitivity parameter is used to control the rate at which the maintenance cost score decays due to increased maintenance time. It is the average of the manually set estimated maintenance time.

8. The method according to claim 1, characterized in that, The correlation between the risk and process units outside the risk propagation chain is calculated using the maximum mutual information coefficient and the propagation entropy to identify potential risk nodes and verify the causal direction. The specific method is as follows: Using the maximum mutual information coefficient Calculate the failure node With other process unit nodes The correlation between them is expressed by the formula: ； in, Representative node and Mutual information between them; and Representing nodes respectively and The number of rows and columns when performing network partitioning; An upper limit constraint on the total number of grid divisions; Transmission entropy is used to verify the causal direction of potential associations by comparing positive time delays. With reverse time delay The numerical difference determines the direction of risk propagation; if the positive time delay is less than the negative time delay, the direction of risk propagation is from the failure node to the potential associated node, forming a directional causal edge.

9. The method according to claim 1, characterized in that, The updated risk blocking sequence is generated using the following method: After incorporating the newly identified potential associated nodes and their causal edges into the existing topology network, the comprehensive decision utility score involving the associated nodes is recalculated, and a brand-new risk blocking sequence is generated by sorting the updated scores in descending order.

10. A quantitative assessment and decision-making system for the safety of natural gas purification production based on root cause analysis and cascading failure propagation, characterized in that, include: The data acquisition module is used to collect process variable data for the entire process unit of natural gas purification, desulfurization, dehydration, and sulfur recovery. The topology network construction module is used to screen the main contributing variables. It uses Granger causal analysis based on neural networks to identify causal relationships and generate a causal adjacency matrix. It screens initial risk nodes, propagation nodes and termination nodes, and constructs a directed topology network with causal relationships as the edge set. The risk assessment module is used to classify node relationships into causal relationships, AND relationships, and OR relationships, and calculate the cascading failure probability of subsequent nodes under each type of relationship. The decision-making module is used to select priority target risk events based on the probability of occurrence of each security incident, extract causal propagation paths, construct a comprehensive decision utility scoring function, and generate risk blocking sequences. The adaptive update module is used to identify potential risk nodes and verify causal direction by using the maximum mutual information coefficient and transmission entropy when the risk blocking sequence cannot effectively cut off the risk propagation chain, thereby realizing the adaptive adjustment of the local topology network and the dynamic update of the blocking sequence.

Images