A fire arrester fire resistance performance evaluation method based on multi-source data fusion

Abstract

The present application belongs to the technical field of fire arrester performance detection and evaluation, and discloses a fire arrester fire resistance performance evaluation method based on multi-source data fusion, comprising: constructing a fire arrester failure mode knowledge graph, constructing a worst operating environment interval set of a fire arrester to be evaluated, establishing a trigger risk probability sequence of different failure modes of the fire arrester to be evaluated, obtaining the failure mode of the fire arrester to be evaluated and the corresponding trigger risk probability prediction value, judging whether the failure mode of the fire arrester to be evaluated and the corresponding trigger risk probability prediction value meet the safety tolerance threshold of the application scene, if yes, determining that there is no abnormality, and if no, determining that there is an abnormality, the present application has the beneficial effect of realizing real-time evaluation and comprehensive risk prediction of the performance of the fire arrester, not only improving the accuracy and reliability of the evaluation result, but also discovering and warning potential safety hazards in time, guiding scientific maintenance, thereby improving the equipment utilization rate and maintenance efficiency and reducing the operation and maintenance cost.

Description

Technical Field

[0001] This invention relates to the field of flame arrester performance testing and evaluation technology, specifically to a method for evaluating the flame arresting performance of flame arresters based on multi-source data fusion. Background Technology

[0002] In industrial production and equipment operation, flame arresters are a key safety device widely used to prevent the spread of flames and protect equipment and personnel safety.

[0003] The performance of flame arresters directly affects the safety and stability of the entire system. However, with the increasing complexity of industrial environments and the ever-increasing safety requirements, traditional performance evaluation methods for flame arresters are no longer sufficient to meet the needs of modern industry.

[0004] Traditional flame arrester performance evaluation mainly relies on single test data and experience judgment, lacking the ability to comprehensively analyze and process multi-source data.

[0005] Specifically, traditional methods have the following shortcomings: Single data source: Traditional evaluation methods often rely solely on laboratory standard test data, ignoring the impact of various factors on the performance of flame arresters in the actual operating environment, resulting in discrepancies between the evaluation results and actual usage.

[0006] Insufficient failure mode analysis: Traditional methods do not analyze the failure modes of flame arresters in sufficient depth, making it difficult to fully identify the various factors that may lead to the failure of flame arresters and their interactions, thus failing to accurately predict the performance of flame arresters in actual operation.

[0007] Poor real-time performance: Traditional assessment methods are mostly offline analyses, which cannot reflect the performance changes of flame arresters in actual operation in real time, making it difficult to detect potential safety hazards in a timely manner.

[0008] Incomplete risk assessment: Traditional methods often only consider the trigger probability of a single failure mode when assessing risk, ignoring the correlation between different failure modes and the overall risk, resulting in incomplete and inaccurate assessment results. Summary of the Invention

[0009] To address the aforementioned problems, this invention provides a method for evaluating the flame arresting performance of flame arresters based on multi-source data fusion. This invention is achieved through the following technical solutions.

[0010] A method for evaluating the flame arresting performance of flame arresters based on multi-source data fusion includes the following steps: S1. Based on the standard test platform, acquire test operation environment data and failure mode reports of several flame arresters. Based on the Bayesian algorithm, determine the environmental impact factor range in the test operation environment data of interest for different failure modes of flame arresters, construct the failure mode association matrix, and construct the flame arrester failure mode knowledge graph based on the failure mode association matrix. S2. Based on heterogeneous sensors, acquire real-time operating environment data of the flame arrester to be evaluated; divide the real-time operating environment data based on time windows, compare the boundary distance of the environmental impact factor intervals in each unit time with the corresponding environmental impact factor reference intervals in the standard operating environment of the flame arrester, filter out environmental impact factors with actual deviation greater than the deviation threshold, and construct the worst operating environment interval set of the flame arrester to be evaluated. S3. The knowledge graph of the failure mode of the flame arrester is matched with the worst operating environment interval set of the flame arrester to be evaluated to determine several failure modes pointed to by the worst operating environment interval set of the flame arrester to be evaluated; the trigger risk probability of different failure modes pointed to by the worst operating environment interval set of the flame arrester to be evaluated is evaluated, and a trigger risk probability sequence of different failure modes of the flame arrester to be evaluated is established. S4. Based on the trigger risk probability sequence of different failure modes of the flame arrester to be evaluated, for each failure mode and its trigger risk probability in the trigger risk probability sequence within a unit time, convert it into a trigger risk probability vector of fixed length. For each component in the vector, if the corresponding failure mode occurs within the unit time, its component is the trigger risk probability of the failure mode corresponding to the vector. If the corresponding failure mode does not occur within the unit time, its component is zero. Construct the trigger risk probability vector sequence of the flame arrester to be evaluated. A trigger risk trend assessment model is constructed. The trigger risk probability vector sequence of the flame arrester to be evaluated at the current moment and several previous moments is used as the input of the model. The input of the model is mapped to a high-dimensional feature space and temporal features are extracted. The extracted temporal features are weighted and fused to obtain fused features. The fused features are mapped to the trigger risk probability prediction vector of the flame arrester to be evaluated at a certain future moment as the output of the model. For the trigger risk probability prediction vector, the failure mode and trigger risk probability corresponding to its non-zero components are extracted to obtain the failure mode of the flame arrester to be evaluated at a certain future moment and its corresponding trigger risk probability prediction value. S5. Determine whether the failure mode of the flame arrester to be evaluated and its corresponding predicted trigger risk probability value meet the safety tolerance threshold of the application scenario. If yes, determine that there is no abnormality; otherwise, determine that there is an abnormality.

[0011] As a further aspect of the present invention, the method for constructing the failure mode correlation matrix in step S1 is as follows: For each round of test operation environment data of the standard test platform, determine the range of all environmental impact factors in that round of test; For each round of failure mode reports from the standard test platform, identify the failure modes that occurred in that round of testing; Based on Bayesian networks, a sample is formed by the environmental impact factor range and failure mode during each round of testing to determine the environmental impact factor range in the test operation environment data of interest for different failure modes of the flame arrester. A failure mode correlation matrix is ​​constructed using failure modes as columns and the environmental impact factor intervals of interest to the failure modes as rows. The failure mode correlation matrix is ​​then filled with the probability of the failure mode occurring within its interest environmental factor interval.

[0012] As a further aspect of the present invention, the method for determining the environmental impact factor range in the test operating environment data of interest for different failure modes of the flame arrester is as follows: For each failure mode, calculate its prior probability of occurrence in all samples; For each interval of each environmental impact factor: Calculate the likelihood probability that the environmental impact factor falls within a certain range in a sample of failure modes. Calculate the marginal probability of the environmental impact factor falling within this interval across all samples; For each failure mode, calculate its posterior probability in each interval of each environmental impact factor; Calculate the lift of each environmental impact factor interval to the failure mode (Lift), and select environmental factor intervals with Lift values ​​greater than a preset risk threshold. These intervals are then identified as environmental impact factor intervals in the test operation environment data of interest to the failure mode.

[0013] As a further aspect of the present invention, the method for constructing the flame arrester failure mode knowledge graph is as follows: Using environmental impact factor ranges and failure modes as nodes, traverse the failure mode correlation matrix and execute the following for each non-zero element in the failure mode correlation matrix: Identify the environmental impact factor interval node and failure mode node corresponding to the element, create a directed edge from the corresponding environmental impact factor interval node to the failure mode node, and store the value of the element as the weight of the directed edge. For the text of each environmental impact factor interval node, it is divided into three semantic components according to the type of environmental impact factor, numerical range and unit. Each semantic component is independently encoded to obtain the corresponding vector. The upper limit, lower limit, range width and midpoint value of the numerical range are calculated as numerical feature vectors. The vectors corresponding to the three semantic components and the numerical feature vectors are fused to obtain the labeled semantic vector corresponding to the text of the environmental impact factor interval node, which is stored as the additional attribute of the node.

[0014] As a further aspect of the present invention, step S2 specifically comprises: The reference range and deviation threshold for each environmental impact factor under the standard operating environment of the flame arrester are preset. The real-time operating environment data of the flame arrester to be evaluated is divided based on the time window. For the real-time operating environment data of the flame arrester to be evaluated in each unit of time, the following processing is performed: Determine the intervals of all environmental impact factors within a unit of time; Based on the boundary distance method, the actual deviation between each environmental impact factor interval and the corresponding reference interval is calculated. The worst operating environment interval set is formed by screening environmental impact factors whose actual deviation is greater than the deviation threshold.

[0015] As a further aspect of the present invention, step S3 specifically includes: The worst operating environment interval set for each unit of time is processed as follows: Encode each environmental impact factor interval in the worst operating environment interval set to obtain a query semantic vector and input it into the flame arrester failure mode knowledge graph. Calculate the cosine similarity between each query semantic vector and the labeled semantic vector corresponding to the environmental impact factor interval node. Set a similarity threshold and filter out all environmental impact factor interval nodes with a cosine similarity greater than the similarity threshold to form a matching set of environmental impact factor interval nodes that match the query semantic vector. For all nodes in the matching set of environmental impact factor interval nodes, starting from their edges, determine all reachable failure mode nodes, which are the failure modes pointed to by the worst operating environment interval set, and constitute the candidate set of failure mode nodes. For each node in the candidate set of failure mode nodes, reverse search is performed to find the environmental impact factor interval nodes that can point to the failure mode node, forming a complete association set of environmental impact factor interval nodes pointing to the failure mode node. The intersection of the complete association set and the matching set is obtained to form a conditional matching set of environmental impact factor interval nodes pointing to the failure mode node. The trigger risk probability of each node in the failure mode candidate set is calculated by aggregation, specifically as follows:

[0016] in, For the first Each failure mode node is in The probability of triggering the risk at any given moment. The number of nodes in the set. for At this moment The set of environmental impact factor interval node condition matching for each failure mode node. for At this moment The environmental impact factor interval node condition matching set of the failure mode node. The node points to the first The weights of nodes with different failure modes. For the time now Complete association set of nodes within the environmental impact factor range of each failure mode node; in The unit of time interval represented by a moment is , The step size of the time window; The failure modes and their trigger risk probabilities for each unit of time are arranged in a time-order sequence.

[0017] As a further aspect of the present invention, step S4 specifically includes: For each failure mode and its trigger risk probability in the trigger risk probability sequence within a unit time, it is converted into a trigger risk probability vector of fixed length. For each component in the vector, if the corresponding failure mode occurs within the unit time, its component is the trigger risk probability of the failure mode corresponding to the vector. If the corresponding failure mode does not occur within the unit time, its component is zero. Construct the trigger risk probability vector sequence of the flame arrester to be evaluated. A trigger risk trend assessment model is constructed. The trigger risk probability vector sequence of the flame arrester to be evaluated at the current moment and several previous moments is used as the input of the model. The input of the model is mapped to a high-dimensional feature space and temporal features are extracted. The extracted temporal features are weighted and fused to obtain fused features. The fused features are mapped to the trigger risk probability prediction vector of the flame arrester to be evaluated at a certain future moment as the output of the model. For the trigger risk probability prediction vector, the failure mode and trigger risk probability corresponding to its non-zero components are extracted to obtain the failure mode of the flame arrester to be evaluated at a certain future moment and its corresponding trigger risk probability prediction value.

[0018] As a further aspect of the present invention, step S5 specifically involves: given a tolerance threshold for the trigger risk probability of different failure modes under the current application scenario of the flame arrester to be evaluated, determining the remaining design time limit of the flame arrester to be evaluated, predicting the failure modes and their corresponding trigger risk probability prediction values ​​when the flame arrester to be evaluated reaches the design time limit, comparing the trigger risk probability prediction values ​​with the tolerance threshold, and determining that there is no abnormality if the trigger risk probability prediction values ​​of all failure modes meet the tolerance threshold; otherwise, determining that there is an abnormality.

[0019] The beneficial effects of this invention are that the flame arrester performance evaluation method based on multi-source data fusion proposed in this invention, through comprehensive standard test and real-time operating environment data, deeply analyzes the failure modes of flame arresters and their relationship with environmental factors, realizing real-time evaluation and comprehensive risk prediction of flame arrester performance. This not only improves the accuracy and reliability of the evaluation results, but also timely detects and warns of potential safety hazards, guides scientific maintenance, thereby improving equipment utilization and maintenance efficiency, and reducing operation and maintenance costs. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 Flowchart of a method for evaluating the flame-arresting performance of flame arresters based on multi-source data fusion Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, a method for evaluating the flame arresting performance of flame arresters based on multi-source data fusion includes the following steps: S1. Based on the standard test platform, acquire test operation environment data and failure mode reports of several flame arresters. Based on the Bayesian algorithm, determine the environmental impact factor range in the test operation environment data of interest for different failure modes of flame arresters, construct a failure mode association matrix, and construct a knowledge graph of flame arrester failure modes based on the failure mode association matrix.

[0024] The failure mode correlation matrix is ​​constructed as follows: For each round of test environment data from the standard test platform, determine the range of all environmental impact factors in that round of testing; For each round of failure mode reports from the standard test platform, identify the failure modes that occurred in that round of testing; Based on Bayesian networks, a sample is formed by the environmental impact factor range and failure mode during each round of testing to determine the environmental impact factor range in the test operation environment data of interest for different failure modes of the flame arrester. A failure mode correlation matrix is ​​constructed using failure modes as columns and the environmental impact factor intervals of interest to the failure modes as rows. The failure mode correlation matrix is ​​then filled with the probability of the failure mode occurring within its interest environmental factor interval.

[0025] Specifically, environmental impact factors include, but are not limited to, temperature, pressure, particulate matter concentration, and corrosive substance concentration. For each environmental impact factor, its range includes the type of environmental impact factor, the numerical range, and the unit. For example, for temperature, the range is: .

[0026] Failure modes include, but are not limited to, blockage, corrosion, and breakdown.

[0027] The method for determining the environmental impact factor range in the test operation environment data for different failure modes of flame arresters is as follows: For each failure mode, calculate its prior probability of occurrence in all samples; For each interval of each environmental impact factor: Calculate the likelihood probability that the environmental impact factor falls within a certain range in a sample of failure modes. Calculate the marginal probability of the environmental impact factor falling within this interval across all samples; For each failure mode, calculate its posterior probability in each interval of each environmental impact factor; Calculate the lift of each environmental impact factor interval to the failure mode (Lift), and select environmental factor intervals with Lift values ​​greater than a preset risk threshold. These intervals are then identified as environmental impact factor intervals in the test operation environment data of interest to the failure mode.

[0028] Assuming there are 100 test samples, and the breakdown failure mode occurs 15 times, its prior probability is 0.15; assuming that the environmental influence factor range of 200-250℃ occurs 12 times during the 15 breakdown failure modes, the likelihood probability is 0.8; and if the temperature range of 200-250℃ occurs 20 times in all test samples, the marginal probability is 0.2, and the posterior probability is 0.6.

[0029] With a lift of 4 and a preset risk threshold of 2, since 4 is greater than 2, it indicates that the probability of this failure mode occurring at a temperature of 200-250℃ is 0.6, and the temperature range of 200-250℃ is the environmental impact factor range of concern for this failure mode.

[0030] The method for constructing a knowledge graph of flame arrester failure modes is as follows: Using environmental impact factor ranges and failure modes as nodes, traverse the failure mode correlation matrix and execute the following for each non-zero element in the failure mode correlation matrix: Identify the environmental impact factor interval node and failure mode node corresponding to the element, create a directed edge from the corresponding environmental impact factor interval node to the failure mode node, and store the value of the element as the weight of the directed edge. For the text of each environmental impact factor interval node, it is divided into three semantic components according to the type of environmental impact factor, numerical range and unit. Each semantic component is independently encoded to obtain the corresponding vector. The upper limit, lower limit, range width and midpoint value of the numerical range are calculated as numerical feature vectors. The vectors corresponding to the three semantic components and the numerical feature vectors are fused to obtain the labeled semantic vector corresponding to the text of the environmental impact factor interval node, which is stored as the additional attribute of the node.

[0031] Specifically, the failure mode association matrix is ​​transformed into a knowledge graph structure of "environmental impact factor interval nodes → failure mode nodes". It also innovatively proposes to perform multi-component (type, interval, unit) semantic encoding and numerical feature fusion on environmental impact factor interval text to generate labeled semantic vectors as attributes of environmental impact factor interval nodes.

[0032] Compared to existing technologies that use symbolic labels or simple encoding to represent nodes, multi-component semantic vector encoding enables nodes in the environmental impact factor interval to have a deep semantic understanding and numerical perception capability. The flame arrester failure mode knowledge graph shows the complex network relationships that may exist between failure modes in the environmental impact factor interval. This allows subsequent matching to be based on semantic similarity, supports fuzzy and generalized matching, and the graph supports complex causal and concurrent reasoning. It effectively solves the shortcomings of existing knowledge representation methods, such as rigidity, inability to handle semantic variations, and lack of relation modeling capabilities.

[0033] S2, based on heterogeneous sensors, acquires real-time operating environment data of the flame arrester to be evaluated, and constructs the worst operating environment range set of the flame arrester to be evaluated by comparing the real-time operating environment data of the flame arrester to be evaluated with the standard operating environment of the flame arrester.

[0034] Step S2 is as follows: The reference range and deviation threshold for each environmental impact factor under the standard operating environment of the flame arrester are preset. The real-time operating environment data of the flame arrester to be evaluated is divided based on the time window. For the real-time operating environment data of the flame arrester to be evaluated in each unit of time, the following processing is performed: Determine the intervals of all environmental impact factors within a unit of time; Based on the boundary distance method, the actual deviation between each environmental impact factor interval and the corresponding reference interval is calculated. The worst operating environment interval set is formed by screening environmental impact factors whose actual deviation is greater than the deviation threshold.

[0035] Specifically, real-time data is processed through a sliding time window, and the deviation between the real-time interval and the standard interval of each environmental impact factor is quantified using the boundary distance method. Based on the deviation threshold, the worst operating environment interval set is selected.

[0036] Compared to existing technologies that use fixed threshold alarms or simple statistical analysis for real-time data, the "boundary distance method" extracts the worst operating environment interval set for each unit of time in the real-time operating environment data of the flame arrester to be evaluated, and uses it as the core influencing factor for the generation of failure modes, ignoring the influence of the reference interval. This is beneficial for subsequent data processing and also makes the environmental impact factors that the failure modes focus on more accurate.

[0037] S3. The knowledge graph of the failure mode of the flame arrester is matched with the worst operating environment interval set of the flame arrester to be evaluated to determine several failure modes pointed to by the worst operating environment interval set of the flame arrester to be evaluated; the trigger risk probability of different failure modes pointed to by the worst operating environment interval set of the flame arrester to be evaluated is evaluated, and a trigger risk probability sequence of different failure modes of the flame arrester to be evaluated is established.

[0038] Step S3 specifically includes: The worst operating environment interval set for each unit of time is processed as follows: Encode each environmental impact factor interval in the worst operating environment interval set to obtain a query semantic vector and input it into the flame arrester failure mode knowledge graph. Calculate the cosine similarity between each query semantic vector and the labeled semantic vector corresponding to the environmental impact factor interval node. Set a similarity threshold and filter out all environmental impact factor interval nodes with a cosine similarity greater than the similarity threshold to form a matching set of environmental impact factor interval nodes that match the query semantic vector. For all nodes in the matching set of environmental impact factor interval nodes, starting from their edges, determine all reachable failure mode nodes, which are the failure modes pointed to by the worst operating environment interval set, and constitute the candidate set of failure mode nodes. For each node in the candidate set of failure mode nodes, reverse search is performed to find the environmental impact factor interval nodes that can point to the failure mode node, forming a complete association set of environmental impact factor interval nodes pointing to the failure mode node. The intersection of the complete association set and the matching set is obtained to form a conditional matching set of environmental impact factor interval nodes pointing to the failure mode node. The trigger risk probability of each node in the failure mode candidate set is calculated by aggregation, specifically as follows:

[0039] in, For the first Each failure mode node is in The probability of triggering the risk at any given moment. The number of nodes in the set. for At this moment The set of environmental impact factor interval node condition matching for each failure mode node. for At this moment The environmental impact factor interval node condition matching set of the failure mode node. The node points to the first The weights of nodes with different failure modes. For the time now Complete association set of nodes within the environmental impact factor range of each failure mode node; in The unit of time interval represented by a moment is , The step size of the time window; The failure modes and their trigger risk probabilities for each unit of time are arranged in a time-order sequence.

[0040] Specifically, the worst-case operating environment interval set is encoded into a query semantic vector for semantic matching. Candidate failure modes are obtained through graph traversal. A complete set of associations and a set of conditional matching are constructed by reverse lookup, and the probability of triggering risk is calculated.

[0041] It takes into account the failure modes directly caused by the current worst operating environment interval set, and also considers other environmental impact factor intervals that do not appear in the current worst operating environment interval set but will affect the failure modes by using the ratio of the condition matching set to the complete association set, making the probability calculation more reasonable.

[0042] S4. Based on the trigger risk probability sequence of different failure modes of the flame arrester to be evaluated, a trigger risk assessment model is constructed to obtain the failure modes of the flame arrester to be evaluated and their corresponding predicted trigger risk probabilities.

[0043] Step S4 specifically includes: For each failure mode and its trigger risk probability in the trigger risk probability sequence within a unit time, it is converted into a trigger risk probability vector of fixed length. For each component in the vector, if the corresponding failure mode occurs within the unit time, its component is the trigger risk probability of the failure mode corresponding to the vector. If the corresponding failure mode does not occur within the unit time, its component is zero. Construct the trigger risk probability vector sequence of the flame arrester to be evaluated. A trigger risk trend assessment model is constructed. The trigger risk probability vector sequence of the flame arrester to be evaluated at the current moment and several previous moments is used as the input of the model. The input of the model is mapped to a high-dimensional feature space and temporal features are extracted. The extracted temporal features are weighted and fused to obtain fused features. The fused features are mapped to the trigger risk probability prediction vector of the flame arrester to be evaluated at a certain future moment as the output of the model. For the trigger risk probability prediction vector, the failure mode and trigger risk probability corresponding to its non-zero components are extracted to obtain the failure mode of the flame arrester to be evaluated at a certain future moment and its corresponding trigger risk probability prediction value.

[0044] Specifically, by converting risk probability sequences into trigger risk probability vector sequences, a trigger risk trend assessment model is constructed. Through feature extraction and fusion, a risk probability prediction vector for future moments is output, and specific failure modes and their predicted probabilities are extracted. Compared to existing technologies that only focus on the current risk state or use simple trend extrapolation for prediction, its core advantage lies in transforming the risk sequence prediction problem into a time-series prediction task suitable for deep learning models. It can automatically capture and model complex nonlinear patterns, periodicities, and dependencies in risk evolution. This achieves a leap from "state assessment" to "trend prediction," providing a forward-looking basis for preventative decision-making and effectively addressing the shortcomings of existing technologies that lack effective risk prediction methods and cannot provide early warning time windows.

[0045] S5. Determine whether the failure mode of the flame arrester to be evaluated and its corresponding predicted trigger risk probability value meet the safety tolerance threshold of the application scenario. If yes, determine that there is no abnormality; otherwise, determine that there is an abnormality.

[0046] Step S5 specifically involves: given the tolerance threshold for the trigger risk probability of different failure modes under the current application scenario of the flame arrester to be evaluated, determining the remaining design time limit of the flame arrester to be evaluated, predicting the failure modes and their corresponding trigger risk probability prediction values ​​when the flame arrester to be evaluated reaches the design time limit, comparing the trigger risk probability prediction values ​​with the tolerance thresholds, and determining that there is no abnormality if the trigger risk probability prediction values ​​of all failure modes meet the tolerance thresholds; otherwise, determining that there is an abnormality.

[0047] Specifically, the tolerance threshold is determined based on expert experience, and the decision-making process integrates the differentiated safety requirements (tolerance threshold) of specific scenarios and the time dimension (remaining time limit), making maintenance decisions (such as replacement plans) more accurate, economical and persuasive, effectively solving the shortcomings of existing technologies that have a single decision-making dimension and cannot support differentiated and precise maintenance strategies based on life cycle.

[0048] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for evaluating the flame-arresting performance of flame arresters based on multi-source data fusion, characterized in that, Includes the following steps: S1. Based on the standard test platform, acquire test operation environment data and failure mode reports of several flame arresters. Based on the Bayesian algorithm, determine the environmental impact factor range in the test operation environment data of interest for different failure modes of flame arresters, construct the failure mode association matrix, and construct the flame arrester failure mode knowledge graph based on the failure mode association matrix. S2. Based on heterogeneous sensors, acquire real-time operating environment data of the flame arrester to be evaluated; divide the real-time operating environment data based on time windows, compare the boundary distance of the environmental impact factor intervals in each unit time with the corresponding environmental impact factor reference intervals in the standard operating environment of the flame arrester, filter out environmental impact factors with actual deviation greater than the deviation threshold, and construct the worst operating environment interval set of the flame arrester to be evaluated. S3. The knowledge graph of the failure mode of the flame arrester is matched with the worst operating environment interval set of the flame arrester to be evaluated to determine several failure modes pointed to by the worst operating environment interval set of the flame arrester to be evaluated; the trigger risk probability of different failure modes pointed to by the worst operating environment interval set of the flame arrester to be evaluated is evaluated, and a trigger risk probability sequence of different failure modes of the flame arrester to be evaluated is established. S4. Based on the trigger risk probability sequence of the flame arrester to be evaluated pointing to different failure modes, for each failure mode and its trigger risk probability within each unit time in the trigger risk probability sequence, convert it into a fixed-length trigger risk probability vector. For each component in the vector, if its corresponding failure mode occurs within that unit time, its component is the trigger risk probability of the failure mode corresponding to that vector; if the failure mode corresponding to that vector does not occur within that unit time, its component is zero. Construct the trigger risk probability vector sequence of the flame arrester to be evaluated, including: A trigger risk trend assessment model is constructed. The trigger risk probability vector sequence of the flame arrester to be evaluated at the current moment and several previous moments is used as the input of the model. The input of the model is mapped to a high-dimensional feature space and temporal features are extracted. The extracted temporal features are weighted and fused to obtain fused features. The fused features are mapped to the trigger risk probability prediction vector of the flame arrester to be evaluated at a certain future moment as the output of the model. For the trigger risk probability prediction vector, the failure mode and trigger risk probability corresponding to its non-zero components are extracted to obtain the failure mode of the flame arrester to be evaluated at a certain future moment and its corresponding trigger risk probability prediction value. S5. Determine whether the failure mode of the flame arrester to be evaluated and its corresponding predicted trigger risk probability value meet the safety tolerance threshold of the application scenario. If yes, determine that there is no abnormality; otherwise, determine that there is an abnormality.

2. The method for evaluating the flame-arresting performance of a flame arrester based on multi-source data fusion according to claim 1, characterized in that, In step S1, the failure mode correlation matrix is ​​constructed as follows: For each round of test operation environment data of the standard test platform, determine the range of all environmental impact factors in that round of test; For each round of failure mode reports from the standard test platform, identify the failure modes that occurred in that round of testing; Based on Bayesian networks, a sample is formed by the environmental impact factor range and failure mode during each round of testing to determine the environmental impact factor range in the test operation environment data of interest for different failure modes of the flame arrester. A failure mode correlation matrix is ​​constructed using failure modes as columns and the environmental impact factor intervals of interest to the failure modes as rows. The failure mode correlation matrix is ​​then filled with the probability of the failure mode occurring within its interest environmental factor interval.

3. The method for evaluating the flame-arresting performance of a flame arrester based on multi-source data fusion according to claim 2, characterized in that, The method for determining the environmental impact factor range in the test operation environment data for different failure modes of flame arresters is as follows: For each failure mode, calculate its prior probability of occurrence in all samples; For each interval of each environmental impact factor: Calculate the likelihood probability that the environmental impact factor falls within a certain range in a sample of failure modes. Calculate the marginal probability of the environmental impact factor falling within this interval across all samples; For each failure mode, calculate its posterior probability in each interval of each environmental impact factor; Calculate the lift of each environmental impact factor interval to the failure mode (Lift), and select environmental factor intervals with Lift values ​​greater than a preset risk threshold. These intervals are then identified as environmental impact factor intervals in the test operation environment data of interest to the failure mode.

4. The method for evaluating the flame-arresting performance of a flame arrester based on multi-source data fusion according to claim 2, characterized in that, The method for constructing the knowledge graph of the flame arrester failure modes is as follows: Using environmental impact factor ranges and failure modes as nodes, traverse the failure mode correlation matrix and execute the following for each non-zero element in the failure mode correlation matrix: Identify the environmental impact factor interval node and failure mode node corresponding to the element, create a directed edge from the corresponding environmental impact factor interval node to the failure mode node, and store the value of the element as the weight of the directed edge. For the text of each environmental impact factor interval node, it is divided into three semantic components according to the type of environmental impact factor, numerical range and unit. Each semantic component is independently encoded to obtain the corresponding vector. The upper limit, lower limit, range width and midpoint value of the numerical range are calculated as numerical feature vectors. The vectors corresponding to the three semantic components and the numerical feature vectors are fused to obtain the labeled semantic vector corresponding to the text of the environmental impact factor interval node, which is stored as the additional attribute of the node.

5. The method for evaluating the flame-arresting performance of a flame arrester based on multi-source data fusion according to claim 1, characterized in that, Step S2 specifically involves: The reference range and deviation threshold for each environmental impact factor under the standard operating environment of the flame arrester are preset. The real-time operating environment data of the flame arrester to be evaluated is divided based on the time window. For the real-time operating environment data of the flame arrester to be evaluated in each unit of time, the following processing is performed: Determine the intervals of all environmental impact factors within a unit of time; Based on the boundary distance method, the actual deviation between each environmental impact factor interval and the corresponding reference interval is calculated. The worst operating environment interval set is formed by screening environmental impact factors whose actual deviation is greater than the deviation threshold.

6. The method for evaluating the flame-arresting performance of a flame arrester based on multi-source data fusion according to claim 1, characterized in that, Step S3 specifically includes: The worst operating environment interval set for each unit of time is processed as follows: Encode each environmental impact factor interval in the worst operating environment interval set to obtain a query semantic vector and input it into the flame arrester failure mode knowledge graph. Calculate the cosine similarity between each query semantic vector and the labeled semantic vector corresponding to the environmental impact factor interval node. Set a similarity threshold and filter out all environmental impact factor interval nodes with a cosine similarity greater than the similarity threshold to form a matching set of environmental impact factor interval nodes that match the query semantic vector. For all nodes in the matching set of environmental impact factor interval nodes, starting from their edges, determine all reachable failure mode nodes, which are the failure modes pointed to by the worst operating environment interval set, and constitute the candidate set of failure mode nodes. For each node in the candidate set of failure mode nodes, reverse search is performed to find the environmental impact factor interval nodes that can point to the failure mode node, forming a complete association set of environmental impact factor interval nodes pointing to the failure mode node. The intersection of the complete association set and the matching set is obtained to form a conditional matching set of environmental impact factor interval nodes pointing to the failure mode node. The trigger risk probability of each node in the failure mode candidate set is calculated by aggregation, specifically as follows: ； in, For the first Each failure mode node is in The probability of triggering the risk at any given moment. The number of nodes in the set. for At this moment The set of environmental impact factor interval node condition matching for each failure mode node. for At this moment The environmental impact factor interval node condition matching set of the failure mode node. The node points to the first The weights of nodes with different failure modes. For the time now Complete association set of nodes within the environmental impact factor range of each failure mode node; in The unit of time interval represented by a moment is , The step size of the time window; The failure modes and their trigger risk probabilities for each unit of time are arranged in a time-order sequence.

7. The method for evaluating the flame-arresting performance of a flame arrester based on multi-source data fusion according to claim 1, characterized in that, Specifically, step S5 involves: given a tolerance threshold for the trigger risk probability of different failure modes under the current application scenario of the flame arrester to be evaluated, determining the remaining design time limit of the flame arrester to be evaluated, predicting the failure modes and their corresponding trigger risk probability prediction values ​​when the flame arrester to be evaluated reaches the design time limit, comparing the trigger risk probability prediction values ​​with the tolerance thresholds, and determining that there is no abnormality if the trigger risk probability prediction values ​​of all failure modes meet the tolerance thresholds; otherwise, determining that there is an abnormality.

Images