Transformer fault hybrid diagnosis method, system and equipment based on knowledge graph, and medium

Abstract

The invention provides a transformer fault hybrid diagnosis method, system and device based on a knowledge graph and a medium, and belongs to the technical field of transformer fault diagnos.The method comprises the steps that historical fault data and real-time monitoring data of a transformer in a region are collected, and the historical fault data are preprocessed; creating a dynamically layered independent sub-graph for each operation age limit interval, and calculating an association weight between adjacent independent sub-graphs; a diagnosis logic chain is generated based on the fault tree, an independent sub-graph of a corresponding age limit interval is embedded as a graph rule, real-time monitoring data is adopted to train a Bayesian network to establish a data model, a graph rule threshold is dynamically corrected, trained parameters are written back to the graph rule, and a local knowledge graph is obtained; and deploying a federal map learning framework, and formulating a directional migration strategy for migrating the aging rule from the high-age-limit area to the low-age-limit area. Cross-regional knowledge safety sharing is realized, the transformer fault is accurately diagnosed, and the reliability of the transformer and the stability of a power grid are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of transformer fault diagnosis, and particularly relates to a transformer fault hybrid diagnosis method, system, device and medium based on a knowledge graph. BACKGROUND

[0002] As the core equipment of the power system, the accuracy and timeliness of the transformer fault diagnosis directly relate to the safe and economic operation of the power grid. The traditional diagnosis method mainly relies on threshold judgment (such as dissolved gas analysis in oil) or a single data-driven model (such as deep learning). However, such methods have significant limitations: first, the transformer fault features are strongly correlated with the service life. The fault modes caused by insulation degradation and mechanical fatigue dynamically evolve with the service time of the transformer, so it is difficult for a static model to adapt to the differentiated laws in different service life intervals. Second, although the existing hybrid diagnosis framework attempts to integrate mechanism rules and data models, it is mostly independent or simply concatenated, lacks deep interaction, and has poor generalization ability and is prone to logical chain breakage in a small sample scenario.

[0003] In addition, although the knowledge graph technology is introduced to incorporate expert experience and historical cases, it generally uses a static topology and fails to explicitly express the dynamic influence of the service life on the fault features. Meanwhile, the data barriers across regions further limit the efficiency of knowledge reuse.

[0004] Therefore, how to construct a dynamic diagnosis technology with layered service life, considering both fault mechanism and data model, and realizing cross-regional knowledge sharing to guide the transformer maintenance and improve the reliability of the transformer and the economy of the power grid is a technical problem to be solved at present. SUMMARY

[0005] The purpose of the present application is to provide a transformer fault hybrid diagnosis method, system, device and medium based on a knowledge graph to solve the above problems: In a first aspect, the present application provides a transformer fault hybrid diagnosis method based on a knowledge graph, comprising the following steps: S1. Collecting historical fault data and real-time monitoring data of transformers in the region, dividing the historical fault data into intervals according to the service life, and labeling the fault type and the service life association label; S2. Creating a dynamic layered independent sub-graph for each service life interval in the Neo4j graph database, and storing the fault modes specific to the interval in the independent sub-graph corresponding to each service life interval, and calculating the association weight between the adjacent independent sub-graphs through a graph attention network; S3. Based on the fault tree, a diagnostic logic chain is generated, and each diagnostic logic chain is embedded in the corresponding independent sub-graph of the running age interval as a graph rule. A data model is established by training a Bayesian network using real-time monitoring data, the graph rule threshold is dynamically corrected, and the trained parameters are written back to the graph rule to obtain a local knowledge graph that introduces a forward reasoning mechanism and backward data-driven optimization. When the data model result conflicts with the graph rule, update the rule threshold or reduce the weight of the conflicting rule, and trigger the manual review process at the same time; S4. Deploy a federated graph learning framework, use graph embedding algorithms to extract features from the local knowledge graph, obtain encrypted parameters through homomorphic encryption, and generate a global knowledge graph model by aggregating the encrypted parameters and returning to each region. Formulate a directional migration strategy for migrating aging rules from high-age areas to low-age areas.

[0006] Further, in step S1, the historical fault data of each age interval is statistically analyzed to identify the frequently occurring fault types and fault characteristics in the age interval. The fault determination rules in the industry standard are structurally analyzed to extract the relationship between fault characteristic thresholds and fault types.

[0007] Further, the specific steps of step S2 are: S21. In the Neo4j graph database, a dynamic hierarchical independent sub-graph is created for each running age interval. The independent sub-graph corresponding to each age interval stores the interval-specific fault mode, including fault type nodes, fault characteristic nodes, and age rule nodes. S22. Set the dynamic fault characteristic threshold rule related to the age. The dynamic fault characteristic threshold rule function is: CO content fault characteristic threshold = basic content ppm + running age x coefficient ppm; Furfural content fault characteristic threshold = basic content mg / g x (1 + running age / 20); S23. Input the node features of each independent sub-graph into the graph attention network, output the correlation weight matrix between adjacent independent sub-graphs, and establish the age migration edge between adjacent age independent sub-graphs through a graph structure learning algorithm.

[0008] Further, the specific steps of step S3 are: S31. Analyze the transformer fault tree from the industry standard, establish a diagnostic logic chain corresponding to each running age interval independent sub-graph, and embed the diagnostic logic chain into each running age interval independent sub-graph as an executable rule. Through Cypher syntax, the rule query is realized, and a candidate fault list is generated. S32. Construct a Bayesian network that includes fault type, fault characteristics, and years of operation, model the conditional probability relationship between fault characteristic parameters and fault probability, as well as the relationship between years of operation and fault characteristic threshold, train the probability model using real-time monitoring data, and write it back to the graph rules. S33. The diagnostic accuracy after the statistical graph rule is triggered is determined by setting a confidence threshold based on the actual situation, and judging whether the confidence level is lower than the confidence threshold. If so, automatically freeze the map rule and proceed to manual review; If not, proceed to step S34; S34. Compare the real-time fault diagnosis results with the manual maintenance records, mark conflict cases, and trigger the reverse optimization process to update the fault feature threshold or reduce the weight of the conflict graph rules.

[0009] Furthermore, the specific steps of step S31 are as follows: S311. Analyze the fault tree of the transformer from the industry standard and decompose the top-level faults into the logical relationship of the bottom-level characteristics; S312. Transform the fault tree logic into a diagnostic logic chain and embed it into the independent sub-graph corresponding to each time interval as an executable rule; S313. Use Cypher syntax to query the rules embedded in the independent sub-graph. By inputting the running years, trigger the corresponding query operation and generate a candidate fault list containing possible faults.

[0010] Furthermore, the specific steps of step S4 are as follows: S41. Deploy the local knowledge graph in each region, containing only data on transformers in that region, and label the age distribution characteristics of local transformers. S42. Use the graph embedding algorithm Node2Vec to convert the independent subgraph nodes into node feature vectors; S43. Use the Microsoft SEAL library to generate public and private keys, and encrypt the node feature vectors and the year-related weights using binary ciphertext. S44. The central server receives the key parameters encrypted by each region, performs a weighted average in the ciphertext state, aggregates them to generate a global knowledge graph model, and then encrypts the global knowledge graph model and returns it to each region. S45. Formulate a knowledge transfer strategy based on the number of years of operation, add the aforementioned targeted transfer rules to the global knowledge graph, and mark the source of the rules. Then, encrypt and transfer the aging rules of the transformer operation high-year area to the transformer operation low-year area. S46. Based on requirements and historical migration data, set an accuracy threshold. If the accuracy of the targeted migration rule in local validation is lower than the accuracy threshold, automatically reduce the weight of the targeted migration.

[0011] Furthermore, it also includes the following steps: The suitability of the time frame is verified through offline testing, and misjudged cases are marked by comparing the results of online testing with manual detection. Incremental learning is triggered based on new data. The specific steps are as follows: Cases with labeled transformer operating years were selected from publicly available datasets, grouped at certain time intervals, and noise data was injected to simulate actual monitoring errors; age matching tests, fault characteristic threshold dynamic tests, and performance tests were conducted. By accessing real-time power grid data via REST API or OPC UA protocol, each transformer is tagged with its operating years, and the transformer fault type and probability are output in real time as a diagnostic basis, which is compared with the manual maintenance report. If the diagnostic results do not match the manual maintenance report, it is determined that the system has misjudged. If a fault type not covered by the global knowledge graph appears, it is determined that the diagnostic basis is insufficient. When the number of new faults in a certain time interval exceeds the set value, incremental learning is triggered to update the graph; when the monthly accuracy of an independent subgraph decreases by more than the specified percentage, node weight adjustment, merging and splitting of independent subgraphs and iteration of the fault threshold formula are triggered. Quarterly statistics are compiled on performance indicators and knowledge base change records. The performance indicators include: accuracy, false alarm rate, and rule triggering frequency for each year interval. A quarterly evaluation report is prepared to verify the rationality of the node weight adjustment, independent subgraph merging and splitting, and fault threshold formula iteration, and to make corresponding optimizations.

[0012] Secondly, embodiments of the present invention also provide a knowledge graph-based hybrid fault diagnosis system for transformers, comprising: The data collection and preprocessing module is used to collect historical fault data and real-time monitoring data of transformers in this area, divide the historical fault data into intervals according to the years of operation, and label the fault type with the associated label of the years of operation. The dynamic hierarchical knowledge graph construction module is used to construct a knowledge graph. It creates a dynamically hierarchical independent subgraph in the Neo4j graph database for each operating time interval. The independent subgraph corresponding to each time interval stores the fault modes unique to that interval, and calculates the association weight between adjacent independent subgraphs through a graph attention network. The bidirectional reasoning and data optimization module is used to generate a diagnostic logic chain based on the fault tree, embed the diagnostic logic chain into the knowledge graph as graph rules, train a Bayesian network using the real-time monitoring data to build a data model, dynamically correct the graph rule thresholds, and write the trained parameters back to the graph rules to obtain a local knowledge graph with forward reasoning mechanism and reverse data-driven optimization. When the above data model results conflict with the above graph rules, the rule thresholds are updated or the weight of the conflicting rules is reduced, and a manual review process is triggered. The federated graph learning and sharing module is used to deploy the federated graph learning framework, use graph embedding algorithms to extract features from the local knowledge graph, obtain encryption parameters through homomorphic encryption, and the central server aggregates the encryption parameters to generate a global knowledge graph model and returns it to each region. It also formulates a targeted migration strategy for aging rules from high-age regions to low-age regions. The system verification and iterative optimization module is used to verify the age adaptability through offline testing, mark misjudgment cases by comparing the results of manual detection through online testing, and trigger incremental learning based on new data. Thirdly, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the knowledge graph-based hybrid diagnosis method for transformer faults as described in the first aspect.

[0013] Fourthly, embodiments of the present invention also provide a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the knowledge graph-based hybrid diagnosis method for transformer faults as described in the first aspect.

[0014] As can be seen from the above technical solutions, the present invention has the following advantages: The dynamic graph hierarchical architecture based on service life intervals, by dividing the graph into multi-level subgraphs and using graph attention networks to quantify the weight transfer between adjacent subgraphs, enables fault features to be adaptively modeled as the service life of the equipment increases. This solves the inherent defect of traditional static models that cannot adapt in a timely and accurate manner when facing dynamic changes in equipment aging patterns, and significantly improves the accuracy and timeliness of transformer fault diagnosis.

[0015] The bidirectional reasoning mechanism of the operating mechanism and data model achieves deep integration and interaction between the two. The forward approach generates diagnostic hypotheses through fault trees and constructs a diagnostic logic chain; the reverse approach trains a probabilistic model with real-time data and dynamically adjusts rule thresholds. This not only meets the requirements of engineering practice for transparency in diagnostic logic, but also achieves high-precision fault diagnosis through a data-driven approach. Especially in small sample scenarios, it effectively avoids problems such as broken logic chains or high false positive rates, enhancing the reliability and credibility of diagnostic results.

[0016] By employing advanced homomorphic encryption technology, it ensures that each region only shares encrypted sub-map features, protecting data privacy while enabling secure knowledge transfer through the aggregation of global models via a central server. This not only meets relevant security regulations but also solves the problem of low knowledge reuse efficiency caused by the uneven distribution of transformer ages across regions, providing technical support for the large-scale promotion and intelligent operation and maintenance of the power industry. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description 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.

[0018] Figure 1 This is a flowchart illustrating the knowledge graph-based hybrid diagnosis method for transformer faults according to the present invention.

[0019] Figure 2 This is a schematic diagram of the knowledge graph-based hybrid diagnostic system for transformer faults according to the present invention. Detailed Implementation

[0020] The various embodiments of the present invention will be described more fully in the detailed steps of the knowledge graph-based hybrid diagnostic method for transformer faults described below. The present invention may have various embodiments, and adjustments and modifications may be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present invention to the specific embodiments disclosed herein, but rather the present invention should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of the present invention.

[0021] To make the objectives, features, and advantages of this invention more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solutions protected by this invention. Obviously, the embodiments described below are only a part of the embodiments of this invention, and not all of them. Based on the embodiments in this patent, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] Please see Figure 1 The diagram shows a flowchart of a knowledge graph-based hybrid fault diagnosis method for transformers in a specific embodiment. The method includes the following steps: S1. Collect historical fault data and real-time monitoring data of transformers in this area, divide the historical fault data into intervals according to the years of operation, and label the fault type with the associated label of the years of operation; It should be noted that by collecting historical fault data and real-time monitoring data of transformers, and by grouping and labeling them in detail according to their years of operation, the changing patterns of fault characteristics as equipment ages can be accurately captured, providing a rich data foundation for the subsequent construction of dynamic hierarchical independent sub-maps. S2. Create a dynamically hierarchical independent subgraph in the Neo4j graph database for each operating period interval. The independent subgraph corresponding to each operating period interval stores the fault modes unique to that interval, and calculate the association weight between adjacent independent subgraphs through a graph attention network. It should be noted that by creating dynamically hierarchical independent subgraphs for each service life interval in the Neo4j graph database, the unique fault modes of each service life interval can be stored and managed in a precise manner. By using graph attention networks to calculate the correlation weights between adjacent subgraphs, the correlation of fault features in different service life intervals can be quantified, enabling adaptive modeling of fault features as the equipment's service life increases. This dynamic hierarchical architecture not only improves the accuracy and adaptability of the fault diagnosis model, but also effectively captures the evolution patterns of fault modes. S3. Generate diagnostic logic chains based on fault trees, and embed each diagnostic logic chain into an independent sub-graph of the corresponding operating year interval as a graph rule. Use the real-time monitoring data to train a Bayesian network to establish a data model, dynamically correct the graph rule threshold, and write the trained parameters back to the graph rule to obtain a local knowledge graph that incorporates forward inference mechanism and reverse data-driven optimization. When the above data model results conflict with the above graph rules, update the rule threshold or reduce the weight of the conflicting rule, and trigger the manual review process. It should be noted that by using a two-way reasoning mechanism of operating mechanism and data model, the real-time monitoring data can be accurately optimized. When the data model results conflict with the operating mechanism rules, the system can automatically reduce the weight of the conflicting rules and trigger the manual review process in a timely manner. This mechanism not only improves the accuracy and reliability of fault diagnosis, but also effectively integrates the advantages of mechanism knowledge and data-driven approach, and enhances the model's adaptability and generalization ability under complex working conditions. S4. Deploy a federated graph learning framework, use graph embedding algorithms to extract features from the local knowledge graph, obtain encryption parameters through homomorphic encryption, and have the central server aggregate the encryption parameters to generate a global knowledge graph model and return it to each region. Formulate a targeted migration strategy for aging rules from high-age regions to low-age regions. It should be noted that by deploying the federated graph learning framework, cross-regional knowledge sharing and collaborative optimization have been achieved. With the help of homomorphic encryption technology, under the premise of protecting data privacy and security, encrypted parameters are aggregated to generate a global knowledge graph model and distributed to various regions. This not only breaks down data silos in the power industry and improves the efficiency of knowledge reuse, but also enhances the generalization ability and adaptability of the model, providing richer and more accurate knowledge support for transformer fault diagnosis. This embodiment improves diagnostic accuracy and enhances model generalization ability by collecting and refining grouped labeled data, constructing dynamic hierarchical independent subgraphs, employing a bidirectional reasoning mechanism, and deploying a federated graph learning framework, thereby more accurately diagnosing transformer faults.

[0023] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, in order to fully illustrate the specific implementation process of this embodiment, another knowledge graph-based hybrid diagnosis method for transformer faults is provided, which includes the following steps: S1. Collect historical fault data and real-time monitoring data of transformers in this area, divide the historical fault data into intervals according to the years of operation, and label the fault type with the associated label of the years of operation; In step S1, the historical fault data of each time limit is statistically analyzed to identify the high-frequency fault types and fault characteristics within that time limit. The fault judgment rules in the industry standard are structured and parsed to extract the relationship between fault feature thresholds and fault types. For example, historical fault and real-time monitoring data of 500 transformers in a power grid area over the past 15 years were collected and divided into three groups according to their operating years: 0-5 years, 5-10 years, and 10-15 years. Fault types were labeled with tags associated with their operating years. Taking the 10-15 year range as an example, from 180 fault records, insulation aging (62%) and winding deformation (28%) were identified as high-frequency faults, with corresponding characteristics including CO content exceeding 350 ppm (91%) and vibration spectrum 100Hz component >0.5g (87%). The IEC 60599 standard was analyzed to extract rules such as "CO threshold increases by 5 ppm annually after 10 years of operation." It should be noted that by dividing the time intervals by years and statistically analyzing fault data, it is possible to dynamically adapt to the aging patterns of transformers, generate more accurate fault characteristic thresholds, provide structured data for constructing a knowledge graph that reflects fault modes at different ages, improve the fit between diagnostic rules and the actual state of the equipment, and enhance the accuracy and pertinence of fault diagnosis. S2. Create a dynamically hierarchical independent subgraph in the Neo4j graph database for each operating period interval. The independent subgraph corresponding to each operating period interval stores the fault modes unique to that interval, and calculate the association weight between adjacent independent subgraphs through a graph attention network. S21. Create a dynamically hierarchical independent subgraph for each running time interval in the Neo4j graph database. The independent subgraph corresponding to each time interval stores the fault modes unique to that interval, including fault type nodes, fault feature nodes, and time rule nodes. S22. Set dynamic fault characteristic threshold rules related to the number of years, wherein the dynamic fault characteristic threshold rule function is: CO content fault characteristic threshold = base content ppm + years of operation × coefficient ppm; Furfural content fault characteristic threshold = base content mg / g × (1 + years of operation / 20); S23. Input the features of each independent subgraph node into the graph attention network, output the association weight matrix between adjacent independent subgraphs, and establish the age migration edge between adjacent age independent subgraphs through the graph structure learning algorithm; For example, define fault type nodes such as "insulation aging" and "winding deformation" and label them with the attribute "10-15 years high-frequency fault"; establish feature nodes such as "CO>350ppm" and "furfural>2.0mg / g" and associate them with dynamic threshold formulas, such as CO threshold = 300 + 5 × years (12-year equipment threshold 360ppm), furfural threshold = 1.5 × (1 + 12 / 20) = 2.4mg / g; add age rule nodes such as "operation > 10 years → insulation detection priority increase"; input the sub-graph node features (fault type, threshold parameters) into the graph attention network, calculate the association weight of "insulation aging" nodes of 10-15 years and 15-20 years sub-graphs as 0.7, and generate age migration edges through graph structure learning algorithm to realize cross-interval association of aging features; It should be noted that by constructing a dynamic hierarchical knowledge graph, the unique fault modes of each age range can be accurately stored, the dynamic threshold rules are in line with the aging patterns of equipment, and the graph attention network establishes cross-age fault associations, thereby improving the graph's adaptive modeling capability for the transformer aging process and providing more dynamic and accurate support for fault diagnosis. S3. Generate diagnostic logic chains based on fault trees, and embed each diagnostic logic chain into an independent sub-graph of the corresponding operating year interval as a graph rule. Use the real-time monitoring data to train a Bayesian network to establish a data model, dynamically correct the graph rule threshold, and write the trained parameters back to the graph rule to obtain a local knowledge graph that incorporates forward inference mechanism and reverse data-driven optimization. When the above data model results conflict with the above graph rules, update the rule threshold or reduce the weight of the conflicting rule, and trigger the manual review process. S31. Analyze the transformer fault tree from industry standards, establish a diagnostic logic chain corresponding to an independent sub-graph for each operating life interval, embed the diagnostic logic chain into the independent sub-graph for each operating life interval as an executable rule, implement rule query through Cypher syntax, and generate a candidate fault list; S311. Analyze the fault tree of the transformer from the industry standard and decompose the top-level faults into the logical relationship of the bottom-level characteristics; S312. Transform the fault tree logic into a diagnostic logic chain and embed it into the independent sub-graph corresponding to each time interval as an executable rule; S313. Use Cypher syntax to query the rules embedded in the independent sub-graph. By inputting the running years, the corresponding query operation is triggered, and a candidate fault list containing possible faults is generated. S32. Construct a Bayesian network that includes fault type, fault characteristics, and years of operation, model the conditional probability relationship between fault characteristic parameters and fault probability, as well as the relationship between years of operation and fault characteristic threshold, train the probability model using real-time monitoring data, and write it back to the graph rules. S33. The diagnostic accuracy after the statistical graph rule is triggered is determined by setting a confidence threshold based on the actual situation, and judging whether the confidence level is lower than the confidence threshold. If so, automatically freeze the map rule and proceed to manual review; If not, proceed to step S34; S34. Compare the real-time fault diagnosis results with the manual maintenance records, mark conflict cases, and trigger the reverse optimization process to update the fault feature threshold or reduce the weight of the conflict graph rules. For example, taking the IEC 60599 standard as an example, the top-level fault "insulation aging" is analyzed from the fault tree and decomposed into the logical AND relationship between "CO content exceeding the standard" and "abnormal furfural concentration". Rule nodes are then created in Neo4j using Cypher. CREATE(r:Rule{id:"R001",logic:"AND(CO>300+years×5,furfural>1.5×(1+years / 20))",years_condition:"years>10", fault_type:"insulation aging"}); Input 12 years of equipment operation data, automatically call the 10-15 year sub-map, and generate a list of candidate faults for "insulation aging"; When constructing the Bayesian network, the CO content and furfural concentration data of the equipment in the region over 12 years were grouped and trained according to the number of years. The CO threshold formula was updated to 300+ years × 5.5 and written back to the graph rules. The accuracy of the rule after being triggered 100 times was 75%. The confidence threshold was set to 70%. Since it was higher than the threshold, it continued to be used. When a 14-year-old device is diagnosed as having "insulation aging (probability 82%)", but is manually inspected and confirmed as having "winding deformation", the system marks the conflict case, reduces the rule weight from 0.8 to 0.6, and triggers the manual review process. Finally, based on expert opinions, the furfural threshold formula is adjusted to 1.5×(1+years / 18) to improve the accuracy of subsequent diagnoses. It should be noted that forward mechanistic reasoning is achieved through fault tree embedding, combined with dynamic threshold correction via Bayesian networks, forming a bidirectional reasoning mechanism. In case of conflicts, rule weights are automatically adjusted and manual review is triggered, improving diagnostic accuracy and enhancing model adaptability. This provides a dynamically optimized solution for transformer fault diagnosis that combines interpretability and data-driven approaches.

[0024] S4. Deploy a federated graph learning framework, use graph embedding algorithms to extract features from the local knowledge graph, obtain encryption parameters through homomorphic encryption, and have the central server aggregate the encryption parameters to generate a global knowledge graph model and return it to each region. Develop a targeted migration strategy to migrate aging rules from high-age regions to low-age regions. The specific steps of step S4 are as follows: S41. Deploy the local knowledge graph in each region, containing only data on transformers in that region, and label the age distribution characteristics of local transformers. S42. Use the graph embedding algorithm Node2Vec to convert the independent subgraph nodes into node feature vectors; S43. Use the Microsoft SEAL library to generate public and private keys, and encrypt the node feature vectors and the year-related weights using binary ciphertext. S44. The central server receives the key parameters encrypted by each region, performs a weighted average in the ciphertext state, aggregates them to generate a global knowledge graph model, and then encrypts the global knowledge graph model and returns it to each region. S45. Formulate a knowledge transfer strategy based on the number of years of operation, add the aforementioned targeted transfer rules to the global knowledge graph, and mark the source of the rules. Then, encrypt and transfer the aging rules of the transformer operation high-year area to the transformer operation low-year area. S46. Based on requirements and historical migration data, set an accuracy threshold. If the accuracy of the targeted migration rule in local validation is lower than the accuracy threshold, automatically reduce the weight of the targeted migration.

[0025] For example, region A is set to mainly consist of transformers with a lifespan of 15-20 years, and region B is set to mainly consist of equipment with a lifespan of 5-10 years. Local knowledge graphs are deployed in both regions. Region A extracts a "insulation aging" sub-graph from data of 120 high-life equipment units, and region B constructs a "winding deformation" sub-graph from data of 180 low-life equipment units, marking the local lifespan distribution characteristics. The Node2Vec algorithm converts the "insulation aging" nodes in region A (e.g., CO threshold = 300 + 5 × years) into 128-dimensional feature vectors, encrypts them using the Microsoft SEAL library, and transmits them to the central server. The server then performs a weighted average of the parameters from regions A and B in encrypted form, generating a global model containing "high-age aging rules" and returning it. For example, the "CO threshold = 300 + 5 × years" rule from region A is encrypted and migrated to region B, labeled "Rule source: Region A - 15-20 year sub-map". When verifying this rule in region B, it is found that the false alarm rate for 10-year-old devices exceeds 15%, which is lower than the 8% accuracy threshold. The system automatically reduces the migration rule weight from 0.7 to 0.4, and simultaneously triggers incremental learning to refit the threshold formula to "300 + 4 × years", ultimately improving the early warning accuracy of new devices in region B to 89%.

[0026] It should be noted that the federated graph learning framework enables secure cross-regional knowledge sharing, breaking down data silos. Homomorphic encryption protects privacy, and aging rules from high-age regions are migrated to low-age regions, improving early warning capabilities. Dynamically adjusting migration weights ensures rule applicability, increasing the diagnostic accuracy of new equipment in low-age regions and enhancing power grid operation and maintenance efficiency. This embodiment also includes the following steps: The suitability of the time frame is verified through offline testing, and misjudged cases are marked by comparing the results of online testing with manual detection. Incremental learning is triggered based on new data. The specific steps are as follows: Cases with labeled transformer operating years were selected from publicly available datasets, grouped at certain time intervals, and noise data was injected to simulate actual monitoring errors; age matching tests, fault characteristic threshold dynamic tests, and performance tests were conducted. By accessing real-time power grid data via REST API or OPC UA protocol, each transformer is tagged with its operating years, and the transformer fault type and probability are output in real time as a diagnostic basis, which is compared with the manual maintenance report. If the diagnostic results do not match the manual maintenance report, it is determined that the system has misjudged. If a fault type not covered by the global knowledge graph appears, it is determined that the diagnostic basis is insufficient. When the number of new faults in a certain time interval exceeds the set value, incremental learning is triggered to update the graph; when the monthly accuracy of an independent subgraph decreases by more than the specified percentage, node weight adjustment, merging and splitting of independent subgraphs and iteration of the fault threshold formula are triggered. Quarterly statistics are compiled on performance indicators and knowledge base change records. The performance indicators include: accuracy, false alarm rate, and rule triggering frequency for each year interval. A quarterly evaluation report is prepared to examine whether the node weight adjustment, independent subgraph merging and splitting, and fault threshold formula iteration are reasonable and to make corresponding optimizations. For example, transformer cases were selected from the IEEE PHM Challenge dataset, grouped by 5-year intervals, and a ±10% noise simulation error was injected. For 18-year-old equipment, the system called the 15-20 year sub-patterns, and the CO threshold was generated as 300 + 5 × 18 = 400 ppm, achieving an accuracy of 86% and a false alarm rate of 7%. When online access was provided to SCADA data, a 12-year-old equipment was diagnosed with "insulation aging," which did not match the manually diagnosed "overheating," and was marked as a misjudgment. When a new fault type, "multi-point grounding of the iron core," was encountered, insufficient coverage was identified. When 60 new fault cases were added to a certain interval, incremental learning was triggered, merging sub-patterns with less data and updating the threshold formula using linear regression. Quarterly reports showed a significant improvement in accuracy after optimization. It should be noted that offline testing verifies the adaptability of the dynamic map, while online testing identifies misjudged and under-covered cases. Incremental learning is then used to dynamically optimize the map. This process improves diagnostic accuracy, reduces false alarm rates, allows the model to continuously adapt to equipment aging patterns, and enables closed-loop optimization through quarterly evaluations, ensuring the reliability and timeliness of the diagnostic system. It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0027] like Figure 2 As shown, the following are embodiments of the knowledge graph-based hybrid diagnosis system for transformer faults provided by the present invention. This system and the knowledge graph-based hybrid diagnosis method for transformer faults described in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the knowledge graph-based hybrid diagnosis system for transformer faults, please refer to the embodiments of the knowledge graph-based hybrid diagnosis method for transformer faults described above.

[0028] The system includes: The data collection and preprocessing module is used to collect historical fault data and real-time monitoring data of transformers in this area, divide the historical fault data into intervals according to the years of operation, and label the fault type with the associated label of the years of operation. The dynamic hierarchical knowledge graph construction module is used to construct a knowledge graph. It creates a dynamically hierarchical independent subgraph in the Neo4j graph database for each operating time interval. The independent subgraph corresponding to each time interval stores the fault modes unique to that interval, and calculates the association weight between adjacent independent subgraphs through a graph attention network. The bidirectional reasoning and data optimization module is used to generate a diagnostic logic chain based on the fault tree, embed the diagnostic logic chain into the knowledge graph as graph rules, train a Bayesian network using the real-time monitoring data to build a data model, dynamically correct the graph rule thresholds, and write the trained parameters back to the graph rules to obtain a local knowledge graph with forward reasoning mechanism and reverse data-driven optimization. When the above data model results conflict with the above graph rules, the rule thresholds are updated or the weight of the conflicting rules is reduced, and a manual review process is triggered. The federated graph learning and sharing module is used to deploy the federated graph learning framework, use graph embedding algorithms to extract features from the local knowledge graph, obtain encryption parameters through homomorphic encryption, and the central server aggregates the encryption parameters to generate a global knowledge graph model and returns it to each region. It also formulates a targeted migration strategy for aging rules from high-age regions to low-age regions. The system verification and iterative optimization module is used to verify the service life adaptability through offline testing, mark misjudgment cases by comparing the results of manual inspection through online testing, and trigger incremental learning based on new data. This embodiment achieves hybrid diagnosis of transformer faults through the interactive collaboration of the data collection and preprocessing module, the dynamic hierarchical knowledge graph construction module, the bidirectional reasoning and data optimization module, the federated graph learning and sharing module, and the system verification and iterative optimization module.

[0029] The knowledge graph-based hybrid diagnosis method for transformer faults provided in this application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structures involved in the embodiments of this invention do not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0030] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0031] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0032] A processor may include one or more processing units, such as a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0033] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0034] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0035] The aforementioned electronic device implements the knowledge graph-based hybrid diagnosis method for transformer faults of this application, which collects historical fault data and real-time monitoring data of transformers in the local area. The historical fault data is divided into intervals according to the service life, and fault type and service life association labels are labeled. A dynamically hierarchical independent subgraph is created in the Neo4j graph database for each service life interval. Each independent subgraph corresponding to a service life interval stores the fault patterns unique to that interval, and the association weights between adjacent independent subgraphs are calculated using a graph attention network. A diagnostic logic chain is generated based on the fault tree, and each diagnostic logic chain is embedded into the independent subgraph of the corresponding service life interval as a graph rule. A Bayesian network is trained using the real-time monitoring data to establish a data model, dynamically... The process involves revising the threshold values ​​of the graph rules and writing the trained parameters back to the graph rules to obtain a local knowledge graph with forward inference and reverse data-driven optimization. When the results of the data model conflict with the graph rules, the rule thresholds are updated or the weights of the conflicting rules are reduced, triggering a manual review process. A federated graph learning framework is deployed, and graph embedding algorithms are used to extract features from the local knowledge graph. Encryption parameters are obtained through homomorphic encryption. The central server aggregates the encryption parameters to generate a global knowledge graph model and returns it to each region. A targeted migration strategy for migrating aging rules from high-age regions to low-age regions is formulated. These steps achieve the beneficial effects of secure cross-regional knowledge sharing, accurate transformer fault diagnosis, and improved transformer reliability and power grid stability.

[0036] The storage medium provided in this application stores a program product capable of implementing a knowledge graph-based hybrid diagnostic method for transformer faults.

[0037] The knowledge graph-based hybrid fault diagnosis method for transformers includes: collecting historical fault data and real-time monitoring data of transformers in the region; dividing the historical fault data into intervals according to their operating years and labeling them with fault type and operating year association tags; creating dynamically hierarchical independent subgraphs in the Neo4j graph database for each operating year interval, where each subgraph stores fault patterns unique to that interval, and calculating the association weights between adjacent independent subgraphs using a graph attention network; generating diagnostic logic chains based on fault trees, and embedding each diagnostic logic chain into the corresponding independent subgraph of the operating year interval as graph rules; and using the real-time monitoring... A Bayesian network is trained using data to build a data model, dynamically adjusting the threshold of the graph rules, and writing the trained parameters back to the graph rules to obtain a local knowledge graph with forward inference mechanism and reverse data-driven optimization. When the results of the above data model conflict with the above graph rules, the rule threshold is updated or the weight of the conflicting rule is reduced, and a manual review process is triggered. A federated graph learning framework is deployed, and graph embedding algorithms are used to extract features from the local knowledge graph. Encryption parameters are obtained through homomorphic encryption. The central server aggregates the encryption parameters to generate a global knowledge graph model and returns it to each region, and a targeted migration strategy is formulated to migrate aging rules from high-age regions to low-age regions.

[0038] In some possible implementations, the knowledge graph-based hybrid diagnosis method for transformer faults of this disclosure can be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0039] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0040] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A hybrid diagnostic method for transformer faults based on knowledge graphs, characterized in that, Includes the following steps: S1. Collect historical fault data and real-time monitoring data of transformers in this area, divide the historical fault data into intervals according to the years of operation, and label the fault type with the associated label of the years of operation; S2. Create a dynamically hierarchical independent subgraph in the Neo4j graph database for each operating period interval. The independent subgraph corresponding to each operating period interval stores the fault modes unique to that interval, and calculate the association weight between adjacent independent subgraphs through a graph attention network. S3. Generate diagnostic logic chains based on fault trees, and embed each diagnostic logic chain into an independent sub-graph of the corresponding operating year interval as a graph rule. Use the real-time monitoring data to train a Bayesian network to establish a data model, dynamically correct the graph rule threshold, and write the trained parameters back to the graph rule to obtain a local knowledge graph that incorporates forward inference mechanism and reverse data-driven optimization. When the above data model results conflict with the above graph rules, update the rule threshold or reduce the weight of the conflicting rule, and trigger the manual review process. S4. Deploy a federated graph learning framework, use graph embedding algorithms to extract features from the local knowledge graph, obtain encryption parameters through homomorphic encryption, and have the central server aggregate the encryption parameters to generate a global knowledge graph model and return it to each region. Formulate a targeted migration strategy for aging rules from high-age regions to low-age regions.

2. The knowledge graph-based hybrid diagnosis method for transformer faults according to claim 1, characterized in that, In step S1, the historical fault data of each time limit is statistically analyzed to identify the high-frequency fault types and fault characteristics within that time limit. The fault judgment rules in the industry standard are then structured and parsed to extract the relationship between fault feature thresholds and fault types.

3. The knowledge graph-based hybrid diagnosis method for transformer faults according to claim 1, characterized in that, The specific steps of step S2 are as follows: S21. Create a dynamically hierarchical independent subgraph for each running time interval in the Neo4j graph database. The independent subgraph corresponding to each time interval stores the fault modes unique to that interval, including fault type nodes, fault feature nodes, and time rule nodes. S22. Set dynamic fault characteristic threshold rules related to the number of years, wherein the dynamic fault characteristic threshold rule function is: CO content fault characteristic threshold = base content ppm + years of operation × coefficient ppm; Furfural content fault characteristic threshold = base content mg / g × (1 + years of operation / 20); S23. Input the features of each independent subgraph node into the graph attention network, output the association weight matrix between adjacent independent subgraphs, and establish the age migration edge between adjacent age independent subgraphs through the graph structure learning algorithm.

4. The knowledge graph-based hybrid diagnosis method for transformer faults according to claim 1, characterized in that, The specific steps of step S3 are as follows: S31. Analyze the transformer fault tree from industry standards, establish a diagnostic logic chain corresponding to an independent sub-graph for each operating life interval, embed the diagnostic logic chain into the independent sub-graph for each operating life interval as an executable rule, implement rule query through Cypher syntax, and generate a candidate fault list; S32. Construct a Bayesian network that includes fault type, fault characteristics and years of operation, model the conditional probability relationship between fault characteristic parameters and fault probability, and the relationship between years of operation and fault characteristic threshold, train the probability model using real-time monitoring data and write it back to the graph rules. S33. The diagnostic accuracy after the statistical graph rule is triggered is determined by setting a confidence threshold based on the actual situation, and judging whether the confidence level is lower than the confidence threshold. If so, automatically freeze the map rule and proceed to manual review; If not, proceed to step S34; S34. Compare the real-time fault diagnosis results with the manual maintenance records, mark conflict cases, and trigger the reverse optimization process to update the fault feature threshold or reduce the weight of the conflict graph rules.

5. The knowledge graph-based hybrid diagnosis method for transformer faults according to claim 4, characterized in that, The specific steps of step S31 are as follows: S311. Analyze the fault tree of the transformer from the industry standard and decompose the top-level faults into the logical relationship of the bottom-level characteristics; S312. Transform the fault tree logic into a diagnostic logic chain and embed it into the independent sub-graph corresponding to each time interval as an executable rule; S313. Use Cypher syntax to query the rules embedded in the independent sub-graph. By inputting the running years, trigger the corresponding query operation and generate a candidate fault list containing possible faults.

6. The knowledge graph-based hybrid diagnosis method for transformer faults according to claim 1, characterized in that, The specific steps of step S4 are as follows: S41. Deploy the local knowledge graph in each region, containing only data on transformers in that region, and label the age distribution characteristics of local transformers. S42. Use the graph embedding algorithm Node2Vec to convert the independent subgraph nodes into node feature vectors; S43. Use the Microsoft SEAL library to generate public and private keys, and encrypt the node feature vectors and the year-related weights using binary ciphertext. S44. The central server receives the key parameters encrypted by each region, performs a weighted average in the ciphertext state, aggregates them to generate a global knowledge graph model, and then encrypts the global knowledge graph model and returns it to each region. S45. Formulate a knowledge transfer strategy based on the number of years of operation, add the aforementioned targeted transfer rules to the global knowledge graph, and mark the source of the rules. Then, encrypt and transfer the aging rules of the transformer operation high-year area to the transformer operation low-year area. S46. Based on requirements and historical migration data, set an accuracy threshold. If the accuracy of the targeted migration rule in local validation is lower than the accuracy threshold, automatically reduce the weight of the targeted migration.

7. The knowledge graph-based hybrid diagnosis method for transformer faults according to claim 1, characterized in that, It also includes the following steps: Offline testing verifies the suitability of the time frame, online testing compares the results of manual detection to mark misjudgments, and incremental learning is triggered based on new data. The specific steps are as follows: Cases with labeled transformer operating years were selected from publicly available datasets, grouped at certain time intervals, and noise data was injected to simulate actual monitoring errors; age matching tests, fault characteristic threshold dynamic tests, and performance tests were conducted. By accessing real-time power grid data through REST API or OPC UA protocol, each transformer is labeled with its operating years, and the transformer fault type and fault probability are output in real time as a diagnostic basis, which is then compared with manual maintenance reports. If the diagnostic results do not match the manual inspection report, it is determined to be a system misjudgment; if a fault type not covered by the global knowledge graph appears, it is determined to be due to insufficient coverage of diagnostic basis. When the number of new faults in a certain time interval exceeds the set value, incremental learning is triggered to update the graph; when the monthly accuracy of an independent subgraph decreases by more than the specified percentage, node weight adjustment, merging and splitting of independent subgraphs and iteration of the fault threshold formula are triggered. Quarterly statistics are compiled on performance indicators and knowledge base change records. The performance indicators include: accuracy, false alarm rate, and rule triggering frequency for each year interval. A quarterly evaluation report is prepared to verify the rationality of the node weight adjustment, independent subgraph merging and splitting, and fault threshold formula iteration, and to make corresponding optimizations.

8. A knowledge graph-based hybrid fault diagnosis system for transformers, characterized in that, include: The data collection and preprocessing module is used to collect historical fault data and real-time monitoring data of transformers in this area, divide the historical fault data into intervals according to the years of operation, and label the fault type with the associated label of the years of operation. The dynamic hierarchical knowledge graph construction module is used to construct a knowledge graph. It creates a dynamically hierarchical independent subgraph in the Neo4j graph database for each operating time interval. The independent subgraph corresponding to each time interval stores the fault modes unique to that interval, and calculates the association weight between adjacent independent subgraphs through a graph attention network. The bidirectional reasoning and data optimization module is used to generate a diagnostic logic chain based on the fault tree, embed the diagnostic logic chain into the knowledge graph as graph rules, train a Bayesian network using the real-time monitoring data to build a data model, dynamically correct the graph rule thresholds, and write the trained parameters back to the graph rules to obtain a local knowledge graph with forward reasoning mechanism and reverse data-driven optimization. When the above data model results conflict with the above graph rules, the rule thresholds are updated or the weight of the conflicting rules is reduced, and a manual review process is triggered. The federated graph learning and sharing module is used to deploy the federated graph learning framework, use graph embedding algorithms to extract features from the local knowledge graph, obtain encryption parameters through homomorphic encryption, and the central server aggregates the encryption parameters to generate a global knowledge graph model and returns it to each region. It also formulates a targeted migration strategy for aging rules from high-age regions to low-age regions. The system verification and iterative optimization module is used to verify the age adaptability through offline testing, mark misjudgment cases by comparing the results of manual detection through online testing, and trigger incremental learning based on new data.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the steps of the knowledge graph-based hybrid diagnosis method for transformer faults as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the knowledge graph-based hybrid diagnosis method for transformer faults as described in any one of claims 1 to 7.

Images