Multi-source data fused power equipment fault root cause analysis method and system

Abstract

The invention provides a multi-source data fused power equipment fault root cause analysis method and system. The method comprises the steps of performing credibility evaluation on multi-source power equipment data; the power equipment comprises sensing monitoring data, operation and maintenance record data and power grid topology data; calculating a fusion weight according to a credibility evaluation result; fusing the power equipment data by using the fusion weight to obtain an equipment fault feature vector; performing correlation calculation on the fault root causes and the key features to obtain a correlation matrix; performing root cause matching degree calculation and screening according to the incidence matrix and the equipment fault feature vector to obtain candidate root causes; constructing a Bayesian network according to the candidate root causes and the corresponding key features, dynamically initializing the conditional probability of the Bayesian network by using real-time data update factors according to an incidence matrix, and adjusting the Bayesian network by using fault reasoning knowledge of a fault knowledge graph; and performing root cause screening on the equipment fault feature vector by adopting a Bayesian network to obtain a power equipment matching fault root cause and a corresponding confidence coefficient. Through the scheme, the accuracy and the disposal efficiency of fault root cause analysis in a complex scene can be improved, and the operation and maintenance cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment operation and maintenance, in particular to a power equipment fault root cause analysis method and system fusing multi-source data. BACKGROUND

[0002] The current power equipment fault root cause analysis technology has three major defects, including the following contents: Insufficient multi-source data fusion: existing technologies rely on single sensing source data including images, sounds and videos, such as using image, sound and video data monitored by sensors to judge the appearance or internal fault of the equipment, but without integrating the operation and maintenance records in the power system and database and the power grid topology data. Single data dimension analysis leads to high fault misjudgment rate and cannot cover complex scenarios such as "data missing, cross fault".

[0003] Fault feature quantization is missing: existing technologies mostly use qualitative description, such as patent number CN118211125A, an artificial intelligence-based power equipment condition-based maintenance method and system, which uses an artificial intelligence model to only identify fault types and does not establish a quantitative correlation model between fault features and root causes. No method is given for calculating the importance weight of features and the matching degree of root causes, resulting in ambiguous root cause positioning, such as "local discharge exceeding the standard", which may correspond to multiple root causes such as insulation aging and component loosening, and cannot be accurately distinguished.

[0004] Low efficiency of root cause reasoning: existing root cause reasoning mostly uses traditional expert systems, such as patent number CN104091290A, an intelligent diagnosis and analysis method for substation monitoring information, which uses case library and rule library to analyze related fault sources. The existing rule base matching system requires manual updating of the rule base, and the reasoning fails when facing new faults based only on sensor monitoring data. Meanwhile, patent number CN110110905A, a power equipment fault judgment and early warning method based on CNN, uses a supervised convolutional neural network algorithm that requires a large number of training samples. The trained algorithm can only represent the correlation between historical root causes and features of the training samples. For fault features other than training samples, the model needs to be retrained for new faults, which has deficiencies in fault recognition migration adaptability and root cause recognition accuracy. Meanwhile, the training phase of the convolutional neural network algorithm requires a large computing power deployment environment to perform effectively. Finally, patent number CN114298188A, an intelligent analysis method and system for power equipment faults, directly traverses large-scale historical fault data for root cause reasoning without deeply mining the correlation between root causes and features in real-time data. Moreover, the reasoning process does not dynamically adjust in combination with real-time data, which cannot meet the "minute-level disposal" demand of the power grid.

[0005] In summary, the prior art cannot meet the three-dimensional requirements of "effective integration of multi-source data, quantitative positioning, and efficient reasoning" for power equipment fault analysis, and there is an urgent need for a root cause analysis scheme that integrates multi-source data, quantifies feature correlation, and dynamically optimizes reasoning. SUMMARY

[0006] In view of this, the present application proposes a power equipment fault root cause analysis method and system that integrates multi-source data to solve the above-mentioned problems existing in the prior art.

[0007] To achieve the above-mentioned purpose, the present application proposes a power equipment fault root cause analysis method and system that integrates multi-source data, comprising: Obtain multi-source power equipment data, and perform credibility evaluation on the multi-source power equipment data; Calculate the fusion weight of different types of power equipment data according to the credibility evaluation result; According to the fusion weight, the standardized power equipment data is weighted and calculated to obtain the equipment fault feature vector; Obtain the fault root cause and key features of the power equipment, calculate the correlation degree of the fault root cause and the key features, and obtain the correlation matrix; According to the correlation matrix and the equipment fault feature vector, the root cause matching degree is calculated and screened to obtain the candidate root cause of the equipment fault feature vector; According to the candidate root cause and the corresponding key feature, a Bayesian network is constructed, and the network parameters of the Bayesian network are initialized according to the correlation matrix. In the initialization process, a real-time data update factor adjusted according to the current real-time data and historical feature deviation is introduced; at the same time, the fault reasoning knowledge of the fault knowledge graph is used to adjust the Bayesian network; The equipment fault feature vector is calculated through the Bayesian network to obtain the posterior probability of each candidate root cause, and the posterior probability is screened to obtain the matched fault root cause of the power equipment and the corresponding confidence.

[0008] Optionally, the calculation process of the fusion weight comprises:

[0009] Among them, indicates the fusion weight coefficient, is the weight coefficient, indicates the data credibility, indicates the fault correlation; the fault correlation is obtained by training the historical fault data.

[0010] Optionally, the process of calculating the correlation degree of the fault root cause and the key features comprises: Calculate the correlation strength between the fault root cause and the key features, wherein the correlation strength is quantified by mutual information entropy:

[0011] in, Represents mutual information entropy. For the first Root cause, For the first One key feature Key features The probability of occurrence root cause The probability of occurrence Key features and root cause The probability of co-occurrence; The mutual information entropy is normalized and integrated to obtain the correlation matrix.

[0012] Optionally, the process of obtaining the candidate root cause includes: The matching degree of the correlation matrix and the equipment fault feature vector is calculated as follows: For the equipment fault feature vector to be analyzed Calculate the matching degree for each type of root cause.

[0013]

[0014] Indicates the degree of matching of the root cause of the failure. Indicates the first Each device fault feature vector, Represents the first in the matrix Line number Column elements, Indicates the feature label, Indicates the total number of features; Construct interaction features, adjust the matching degree based on the interaction features to obtain the final matching degree, judge the final matching degree, and obtain candidate root causes based on the judgment result.

[0015] Optionally, in the Bayesian network, the candidate root is the parent node and the corresponding key feature is the child node, and the conditional probability of the Bayesian network is adjusted according to the elements in the association matrix; The Bayesian network incorporates reasoning knowledge between root causes and key features in a knowledge graph. Specifically, the knowledge correlation degree between candidate root causes and key features is calculated through the association paths between candidate root causes and key features in the fault knowledge graph. The conditional probabilities in the Bayesian network are then readjusted based on the knowledge correlation degree to assist the computational reasoning of the Bayesian network.

[0016] Optionally, during the initialization of the Bayesian network, the conditional probabilities of the Bayesian network are adjusted using a real-time data update factor:

[0017] in, This represents the co-occurrence probability of features and root causes in real-time data. To introduce conditional probabilities for association with fault knowledge graphs. For real-time data update factors, This is the adjusted conditional probability.

[0018] Optionally, the process of obtaining the posterior probability of each candidate root cause includes:

[0019] in, root cause The prior probability is obtained based on historical failure frequency statistics. This represents the probability of occurrence of the equipment failure feature vector. It represents the likelihood probability.

[0020] On the other hand, the present invention provides a power equipment fault root cause analysis system that integrates multi-source data for performing the above-described method, including: The data acquisition layer is used to acquire power equipment data from multiple sources. The data fusion layer is used to assess the credibility of multi-source power equipment data; calculate the fusion weight of different types of power equipment data based on the credibility assessment results; and perform weighted calculation on the standardized power equipment data based on the fusion weight to obtain the equipment fault feature vector. The root cause analysis layer is used to obtain the root causes and key features of power equipment faults. It calculates the correlation between the root causes and key features to obtain a correlation matrix. Based on the correlation matrix and equipment fault feature vectors, it calculates and filters root cause matching, obtaining candidate root causes of the equipment fault feature vectors. A Bayesian network is constructed based on the candidate root causes and corresponding key features, and the conditional probability of the Bayesian network is initialized according to the correlation matrix. During initialization, a real-time data update factor adjusted based on current real-time data and historical feature deviations is introduced. Simultaneously, fault reasoning knowledge from a fault knowledge graph is used to adjust the Bayesian network. The Bayesian network is used to calculate the posterior probability of each candidate root cause from the equipment fault feature vectors. Based on the posterior probability, the network is filtered to obtain the matching fault root cause of the power equipment and its corresponding confidence level.

[0021] The results output layer is used to provide root cause analysis reports and handling guidance to operations and maintenance personnel. Compared with the prior art, the beneficial effects of the present invention are as follows: Improved analytical accuracy: Multi-source data fusion and quantitative correlation models reduce the false fault rate, improve the root cause location accuracy, and avoid blind repairs caused by false faults. Optimization of handling efficiency: The dynamic reasoning mechanism shortens the time required for root cause analysis, meets the grid's minute-level handling needs, and reduces equipment downtime; Adaptation to complex scenarios: It adopts dynamic reasoning based on fault knowledge fusion, eliminating the need for training samples and reducing the computational dependence of deployed equipment. By leveraging the correlation between root causes and features and the inference relationship of fault knowledge, it improves the identification rate of new faults, covering new fault scenarios such as new energy grid connection and extreme weather, thereby enhancing the power grid's resilience. Reduced maintenance costs: Precise root cause analysis reduces unnecessary repairs, thus lowering the maintenance cost per device. Attached Figure Description

[0022] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. In the drawings: Figure 1 This is a timing diagram of the entire process of root cause analysis in this embodiment of the invention; Figure 2 This is a system architecture diagram in an embodiment of the present invention; Figure 3 This is a partial fault knowledge architecture diagram in the fault knowledge graph of this invention embodiment; Figure 4 This is the reasoning logic diagram of the fused fault knowledge graph in this embodiment of the invention. Detailed Implementation

[0023] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] This invention relates to the field of power equipment operation and maintenance technology, specifically to a method and system for root cause analysis of power equipment faults that integrates sensor monitoring, operation and maintenance records, and power grid topology data. It is applicable to key power equipment such as transformers, circuit breakers, and gas-insulated switchgear (GIS), and can solve the technical problems of existing fault analysis data being of limited dimensions, unclear root cause location, and low reasoning efficiency, thereby achieving "precise location, rapid source tracing, and efficient handling" of power equipment faults. This invention proposes a method for root cause analysis of power equipment faults that integrates multi-source data, such as... Figure 1 As shown, it includes: Collect multi-source power equipment data and corresponding credibility, train and generate fault correlation based on historical fault data, determine the hierarchical fusion weight of different types of power equipment data based on credibility and fault correlation, perform weighted fusion of the standardized multi-source power equipment data feature values, and generate the equipment fault feature vector to be analyzed. Extract the root causes of power equipment failures and their corresponding key features to form a key feature set for each root cause; calculate the correlation strength between the root causes and the key feature sets to generate a correlation matrix; use the correlation matrix to calculate and filter the root cause matching degree of the equipment failure feature vectors to be analyzed, and generate candidate root causes of the equipment failure feature vectors to be analyzed. Using candidate root causes as parent nodes and key features from the key feature set as child nodes, the network parameters of the Bayesian network are initialized through the correlation matrix, and the conditional probability, i.e. the network parameters, is dynamically adjusted by introducing a real-time data update factor. The posterior probability of each candidate root cause is calculated by using the fault feature vector of the equipment to be analyzed as input through the Bayesian network, and one or more candidate root causes with the highest posterior probability are selected as the matching fault root causes and corresponding confidence levels of the power equipment.

[0025] As a preferred approach, key features with low relevance are removed from the key feature set to reduce feature dimensionality. Root cause analysis is then performed using a hierarchical reasoning method, moving from high-matching root causes to low-matching root causes, to improve the efficiency of equipment failure root cause analysis.

[0026] In light of the existing technical problems, the above technical solution is described in detail: To address the issue of insufficient multi-source data fusion, a multi-source data hierarchical fusion model is constructed, and a hierarchical fusion mechanism based on data credibility is designed. The core logic is as follows: Data Classification and Credibility Definition: Multi-source power equipment data are classified into three categories, and credibility coefficients are defined. The credibility coefficient The value ranges from 0 to 1. The credibility coefficient is based on the evaluation of data collection accuracy and completeness, which can be achieved by directly weighting the data collection accuracy and completeness or by directly using either the data collection accuracy or completeness as the credibility value. Multi-source power equipment data can be divided into: The sensor monitoring data, including temperature, partial discharge, and dissolved gases in oil, has a reliability coefficient of [value missing]. Sensors deployed on power equipment collect data in real time; Maintenance and repair records, including repair time, spare parts replacement, and defect records, have a reliability coefficient of [value missing]. There are a few omissions due to manual data entry in the database; Power grid topology data, including equipment relationships, load factor, voltage level, and the reliability coefficient of the power grid topology data. The power system is generated automatically.

[0027] Layered fusion weight calculation: For different types of power equipment data, a fusion weight coefficient is introduced. ,in Indicates data category, and integrates weight coefficients. It takes into account the credibility of the data. Correlation with faults Fault correlation The value ranges from 0 to 1 and is obtained by fitting historical fault data, such as the correlation between partial discharge data and insulation faults. Fusion weighting coefficients The calculation formula is as follows:

[0028] in, These are the weighting coefficients. The aforementioned weighting coefficients are generated by fitting historical data or selected based on empirical values. These values ​​enable the fused data to improve the accuracy of fault characterization.

[0029] Fusion Data Output: After spatiotemporal alignment of the three types of multi-source power equipment data, the data is output according to the fusion weighting coefficient. Fusion generation of equipment fault feature vectors in, Standardized feature values ​​for various types of power equipment data are used, with the timestamp of sensor monitoring data as the benchmark in spatiotemporal alignment, and the time deviation is ≤1s, providing full-dimensional data support for subsequent root cause analysis.

[0030] To address the issue of "missing fault feature quantification," a fault feature-root cause quantification correlation model is designed, constructing a feature-root cause correlation matrix based on mutual information entropy. The core design is as follows: Root Cause Classification and Feature Extraction: Common root causes of power equipment failures are classified into five categories: insulation aging, component loosening, oil deterioration, overload, and external interference. Key features corresponding to each type of root cause are extracted. For example, the key fault features corresponding to insulation aging are excessive partial discharge and increased dielectric loss, forming a key feature set. , where n=12, covering key characteristic dimensions such as partial discharge quantity, temperature, discharge, and oil quality.

[0031] Feature-root cause correlation calculation: using mutual information entropy The formula for quantifying the correlation strength between key features and root causes is as follows:

[0032] in, For the first Root cause, For the first One key feature Key features The probability of occurrence root cause The probability of occurrence Key features and root cause The probability of co-occurrence is calculated based on historical fault data. Normalized to [0,1], forming Correlation matrix ,like Indicates "partial discharge quantity" With insulation aging The correlation coefficient was 0.92.

[0033] Root cause matching degree calculation: For the equipment fault feature vector of the fault to be analyzed Calculate the matching degree for each type of root cause.

[0034]

[0035] in, Represents the equipment fault feature vector The i-th fault feature in the fault feature vector only represents the direct matching degree between the root cause and the real-time data. However, among the features in the fault feature vector, the feature values ​​that represent the same root cause may also show correlation. Based on the direct matching degree, the correlation between feature values ​​should be integrated. Therefore, interactive features are introduced on the basis of the matching degree: ; in, The scaling factor represents the initial matching degree and the interaction features, ensuring... Within the range of 0-1, the weights of the interaction features are controlled simultaneously. Representing the interaction coefficient:

[0036] in, It is the interaction term adjustment coefficient. It is the geometric mean of the correlation strength. It is an exponentially decaying term of the difference in fault characteristics. This represents the data in the p-th row and j-th column of the correlation matrix. This represents the p-th fault feature in the fault feature vector. and This represents the mean of the historical data corresponding to the p-th and q-th fault features in the fault feature vector. The scaling parameter for the differences in fault characteristics is the mean of the standard deviation of fault characteristics under historical data.

[0037] A higher value indicates a higher degree of match between the root cause and the current fault, thus enabling initial screening. The root cause was used as a candidate root cause.

[0038] like Figure 4 As shown, to address the problem of "low efficiency in root cause reasoning," a dynamic root cause reasoning optimization mechanism is designed, and a dynamic reasoning method based on improved Bayesian networks is proposed. The core steps are as follows: Bayesian network initialization: starting with candidate root causes As the parent node, the key characteristics of the fault For child nodes, initialize network parameters based on the association matrix M, i.e., the conditional probabilities in the Bayesian network. .

[0039] In the root cause reasoning process, the initialization of the Bayesian network depends on the association between candidate root causes and key features. The association matrix only represents the direct relationship between root causes and key features. However, in addition to the connection between root causes and key features, it should also include the guiding or reasoning relationship between the occurrence of root cause phenomena and the representation of key features. However, the above association matrix cannot represent this relationship. In order to incorporate the reasoning relationship between fault root causes and fault features into the Bayesian reasoning process, in this invention, knowledge-driven methods using fault knowledge graphs are used to enhance the node association of the Bayesian network. Specifically, by starting with candidate root causes in the knowledge graph and searching along multi-hop paths, and combining the importance weights of various relationships along the path, the correlation between root causes and features is calculated. This correlation is used as the deductive relationship between fault root causes and fault features, which serves as a further supplement to the association matrix. This allows for knowledge-guided adjustments to the conditional probabilities in the Bayesian network, thereby structurally integrating the domain knowledge implicit in the knowledge graph into the probabilistic reasoning model. This enables the network parameters to use not only the association matrix representing direct relationships but also the reasoning relationships representing knowledge reasoning relationships for subsequent reasoning calculations, making it more consistent with actual fault logic and equipment operation experience.

[0040] To address the above, integrating fault knowledge graphs into the Bayesian network reasoning process specifically includes the following: introducing fault knowledge graphs to assist in adjusting the conditional probabilities in the Bayesian network, such as... Figure 3As shown, for the fault knowledge graph, at the entity definition level, core entities covering the entire process of equipment operation faults are extracted. These include equipment classes representing the equipment itself, component classes constituting the equipment (such as core and auxiliary components), root causes describing abnormal states, and parameter values ​​characterizing key fault features. These entities collectively form the foundation of the fault knowledge graph. Furthermore, by establishing relationships between these entities and assigning corresponding importance to different relationships, knowledge connections are established between these entities.

[0041] Based on the aforementioned root cause categories and corresponding key features, starting with the query entity of the root cause category under the same type of equipment, the corresponding key features are retrieved in a multi-hop manner. The associated path is the same equipment class, root cause of failure, faulty component, fault phenomenon, key features of failure, and cause of failure. The retrieval directly retrieves the faulty component from the root cause of failure for the same equipment class, and indirectly retrieves the associated entity through the faulty component.

[0042] Based on the comprehensive relationship weight of the paths, paths from root cause categories to key features are filtered, which may also include the relationship between device type and root cause category. The paths with the highest relationship weights are then sorted and filtered. The calculation method is as follows:

[0043] in, This indicates the importance of different types of relationships in the path from candidate root causes to key features. The query path length is the number of relation edges in the path from the candidate root cause to the key feature.

[0044] Faced with the aforementioned related paths, the relevance scores of entities and relational fragments in the paths from candidate root causes to key features are calculated based on factors such as relational weights, and the most relevant key information fragments are selected as the final retrieval results.

[0045] Specifically, to characterize the degree of fault knowledge correlation between candidate root causes and key features, a correlation degree is introduced. The correlation degree is calculated as follows:

[0046] in: The value ranges from 0 to 1, representing the relevance between the candidate root causes in the query and the corresponding entities in the fault knowledge graph. This is the length of the query path, i.e., the number of relation edges in the path; , , In this embodiment, the preset weight values ​​are used. ;

[0047] in, It is the feature vector representation of the retrieved entity corresponding to the candidate root cause. It is a vector representation of the candidate root causes of the query.

[0048] Finally press Output the top few paths in descending order, such as the Top-5, and calculate the mean of the above correlations as the adjustment amount S for the conditional probability. Z By adjusting the correlation strength of the mutual information entropy representation in the fault knowledge graph by considering the correlation degree of different knowledge in the fault knowledge graph, the inference efficiency and accuracy of the data Bayesian network are improved. The initial probability of the adjusted Bayesian network is:

[0049] in, and The weights representing the correlation matrix and the adjustment amount are 0.7 and 0.3, respectively. This represents the adjusted initial conditional probability.

[0050] Introducing a real-time data update factor Real-time data update factor The value ranges from 0 to 1. When the deviation between the real-time data and the mean value of the corresponding entity's node in the fault knowledge graph is greater than 15%, Dynamically adjust conditional probabilities:

[0051] in, This represents the co-occurrence probability of features and root causes in real-time data.

[0052] Root Cause Reasoning and Ranking: Input the fault feature vector of the equipment to be analyzed The posterior probability of each candidate root cause is calculated using a Bayesian network. :

[0053] in, root cause The prior probability is obtained based on historical failure frequency statistics. The likelihood probability represents the probability of occurrence of a device fault feature vector. According to the posterior probability of each candidate root cause. Sort in descending order and output. Root causes and confidence levels, such as "insulation aging, confidence level 95%; loose components, confidence level 8%; oil deterioration, confidence level 2%".

[0054] After outputting the corresponding root cause and confidence level in the Bayesian network, the root cause of the fault is used as the query entity. Using the fault knowledge graph, the fault cause or other entity nodes are used as the query endpoints for relevant searches through the multi-hop retrieval method. The most relevant search results are output to clearly describe the association chain between the corresponding root cause and key features, and to assist in the generation of subsequent handling solutions.

[0055] Inference efficiency optimization: The "feature screening-hierarchical inference" strategy is adopted: first, key features are screened to reduce feature dimensions, and then inference is performed hierarchically according to "high matching degree root cause → low matching degree root cause". The time consumption of single equipment failure root cause analysis is shortened, meeting the power grid's minute-level handling requirements.

[0056] During the key feature selection process, the normal distribution of the key feature's normal data is statistically analyzed, and its mean and standard deviation are calculated; corresponding dynamic thresholds are set.

[0057] Indicates a dynamic threshold. and These represent the corresponding mean and standard deviation, and k represents the dynamically adjusted weighting coefficient.

[0058] in, This represents the base value, which is 1.5. and This indicates the severity score of the fault and the equipment reliability score. and This represents the weighting coefficient, set to 0.2 or 0.3. When the key feature exceeds the above dynamic threshold, the key feature is retained. Simultaneously, the lower bound dynamic threshold can be calculated using the method described above; that is, the difference between the mean and the adjusted standard deviation is used as the dynamic threshold in the calculation.

[0059] like Figure 2 As shown, on the other hand, this invention also proposes a power equipment fault root cause analysis system that integrates multi-source data. This system corresponds to the above-mentioned method flow and includes a four-layer architecture. Each layer collaboratively realizes multi-source data fusion and fault root cause analysis, and the functions of each module do not overlap with existing technologies. 1) Data acquisition layer: Deploy sensing devices, operation and maintenance terminals and power grid topology database to realize real-time acquisition and storage of data from three types of power equipment; The deployed sensing equipment includes temperature sensors, partial discharge detectors, and oil quality analyzers; the maintenance terminals include PADs for maintenance personnel.

[0060] 2) Data fusion layer: includes a spatiotemporal alignment module, a credibility assessment module, and a hierarchical fusion calculation module, which outputs standardized equipment fault feature vectors; The spatiotemporal alignment module is used to perform spatiotemporal alignment and standardized feature extraction on power equipment data.

[0061] The credibility assessment module is used to assess the credibility of power equipment data.

[0062] The hierarchical fusion calculation module is used to calculate the corresponding fusion weight coefficient based on the credibility assessment results. The fusion weight coefficient is used to perform weighted fusion of the standardized feature values ​​of power equipment data to output a standardized equipment fault feature vector.

[0063] 3) Root cause analysis layer: including the association matrix construction module, the candidate root cause generation module, the Bayesian network inference module, and the root cause ranking module, to complete the candidate root cause screening and confidence calculation; The correlation matrix construction module is used to set the root cause type of the fault and extract the key features corresponding to the root cause, and calculate the correlation matrix between different root causes and key features based on the key features.

[0064] The candidate root cause generation module is used to calculate the matching degree between the root cause of the fault and the feature vector of the equipment fault based on the correlation matrix. Based on the matching degree, it is adjusted through the interaction features between key features. Based on the adjusted matching degree, the root cause of the fault is screened to generate the corresponding candidate root cause.

[0065] The Bayesian network inference module is used to construct a Bayesian network. It initializes the Bayesian network based on candidate root causes, key features, and association matrices, using the equipment fault feature vector as input data. The Bayesian network calculates the posterior probability of each candidate root cause and uses the posterior probability as an indicator to filter the candidate root causes, generate power equipment matching fault root causes, and output their corresponding confidence scores.

[0066] The fault knowledge module is used to introduce fault reasoning knowledge from the fault knowledge graph into the Bayesian network. The introduction of fault reasoning knowledge from the fault knowledge graph adjusts the prior conditional probabilities in the Bayesian network, so as to introduce fault reasoning knowledge into the reasoning computation.

[0067] 4) Results output layer: This includes a visualization module and a handling suggestion generation module, which outputs root cause analysis reports and handling guidance to operation and maintenance personnel.

[0068] The visualization module displays the results through root cause confidence bar charts and feature correlation heatmaps, and provides corresponding root cause analysis reports to operations and maintenance personnel. The action suggestion generation module provides corresponding action suggestions to operations and maintenance personnel based on root cause matching and historical action plans.

[0069] In the results output layer, the visualization module also integrates a human-computer interaction mechanism to build an intelligent analysis interface with real-time dynamic presentation capabilities. This module develops an interactive dashboard based on Web front-end technology, integrating multi-dimensional visualization components such as root cause confidence bar charts, feature correlation heatmaps, device topology views, and key feature time-series curves. It also uses WebSocket technology to achieve real-time data synchronization with the backend inference module. When the Bayesian network outputs new root cause analysis results, the visualization module automatically pushes updates to the front-end interface, achieving real-time refresh of the confidence chart and heatmap correlation matrix, and supporting pop-up alarms and multi-channel notifications for abnormal root causes. Maintenance personnel can click on the bar chart or heatmap cells to query and view the detailed inference path, associated feature values, and historical similar cases for the root cause; a timeline slider control is provided to support tracing back the fault status and feature trends of any time period; the visualization module also provides a "one-click report generation" function, automatically integrating the current analysis results, visualization charts, and handling suggestions to output a structured PDF / Word report. The interface supports a responsive layout, adapts to tablets and mobile devices, and can integrate a voice interaction interface to enable natural interactive operations such as "voice query for root cause" and "voice export of report," thereby improving the efficiency of maintenance personnel in real-time perception, interactive investigation, and decision-making regarding the root cause of faults.

[0070] This invention overcomes three major bottlenecks in current power equipment fault analysis technology: insufficient multi-source data fusion, lack of fault feature quantification, and low efficiency of root cause reasoning. It achieves a systematic, dynamic, and quantifiable deep root cause analysis paradigm. Compared to existing technologies, this invention, for the first time, performs hierarchical fusion of three types of heterogeneous information—sensor monitoring data, maintenance record data, and power grid topology data—based on reliability and fault correlation, forming a full-dimensional, highly reliable equipment fault feature vector, laying a solid data foundation for fault analysis in complex scenarios. In feature association modeling, this invention abandons traditional qualitative descriptions or simple rule matching, innovatively introducing a mutual information entropy model to achieve precise quantification of the correlation strength between key features and potential root causes, constructing a computable correlation matrix, fundamentally solving the problem of ambiguous root cause localization. In terms of the core reasoning mechanism, this invention designs a dynamic reasoning framework based on a Bayesian network that integrates fault knowledge. It constructs a Bayesian network model using current, relevant real-time data, scientifically initializes the network conditional probabilities using the aforementioned quantized correlation matrix, and introduces a real-time data update factor. This framework integrates knowledge associations from the fault knowledge graph, providing supplementary support to the Bayesian network and serving as an aid and explanation for Bayesian reasoning, thus enhancing the interpretability and context awareness of the reasoning process. Employing dynamic reasoning based on fault knowledge fusion eliminates the need for training samples and reduces the computational dependence of deployed equipment. By leveraging the correlations between root causes and features and the derivation relationships of fault knowledge, it can improve the identification rate of novel faults in complex scenarios, covering new energy grid connection, extreme weather, and other novel fault scenarios, thereby enhancing the resilience of power equipment.

[0071] In summary, compared with the prior art, the main advantages of the present invention are as follows: Improved analytical accuracy: Multi-source data fusion and quantitative correlation models reduce the false fault rate and improve the root cause location accuracy, avoiding blind repairs caused by false faults; Optimization of handling efficiency: The dynamic reasoning mechanism shortens the time required for root cause analysis and reduces equipment downtime; Adaptation to complex scenarios: It can improve the identification rate of new faults in complex scenarios, and can cover new fault scenarios such as new energy grid connection and extreme weather, thereby improving the power grid's ability to withstand risks; Reduced maintenance costs: Precise root cause identification reduces the number of unnecessary repairs, lowering the maintenance cost per device and demonstrating significant economic value.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for root cause analysis of power equipment faults by integrating multi-source data, characterized in that, include: Acquire multi-source power equipment data and conduct a reliability assessment of the multi-source power equipment data; The fusion weights of different types of power equipment data are calculated based on the credibility assessment results; The standardized power equipment data is weighted according to the fusion weight to obtain the equipment fault feature vector; Obtain the root causes and key features of power equipment failures, calculate the correlation degree of the root causes and key features, and obtain the correlation matrix. The root cause matching degree is calculated and filtered based on the correlation matrix and equipment fault feature vectors to obtain candidate root causes of equipment fault feature vectors; A Bayesian network is constructed based on candidate root causes and corresponding key features. The conditional probabilities of the Bayesian network are initialized based on the correlation matrix. During the initialization process, a real-time data update factor, adjusted based on the current real-time data and historical feature deviations, is introduced to dynamically adjust the conditional probabilities. At the same time, fault reasoning knowledge from the fault knowledge graph is used to adjust the Bayesian network. The fault feature vector of the equipment is calculated by using a Bayesian network to obtain the posterior probability of each candidate root cause. The root causes of power equipment faults and their corresponding confidence levels are then obtained by filtering based on the posterior probabilities.

2. The method according to claim 1, characterized in that, The calculation process of the fusion weight includes: ， in, Indicates the fusion weight coefficient. These are the weighting coefficients. Indicates data credibility. This indicates the fault correlation; the fault correlation is obtained by training based on historical fault data.

3. The method according to claim 1, characterized in that, The process of calculating the correlation between the root causes and key features of the fault includes: Calculate the correlation strength between the root cause of the failure and the key features, wherein the correlation strength is quantified by mutual information entropy: ， in, Represents mutual information entropy. For the first Root cause, For the first One key feature Key features The probability of occurrence root cause The probability of occurrence Key features and root cause The probability of co-occurrence; The mutual information entropy is normalized and integrated to obtain the correlation matrix.

4. The method according to claim 1, characterized in that, The process of obtaining the candidate root causes includes: The matching degree of the correlation matrix and the equipment fault feature vector is calculated as follows: For the equipment fault feature vector to be analyzed Calculate the matching degree for each type of root cause. ， Indicates the degree of matching of the root cause of the failure. Indicates the first Each device fault feature vector, Represents the first in the matrix Line 1 Column elements, Indicates the feature label, Indicates the total number of features; Construct interaction features, adjust the matching degree based on the interaction features to obtain the final matching degree, judge the final matching degree, and obtain candidate root causes based on the judgment result.

5. The method according to claim 1, characterized in that, In the Bayesian network, the candidate root is the parent node and the corresponding key feature is the child node. The conditional probability of the Bayesian network is adjusted according to the elements in the association matrix. The Bayesian network incorporates reasoning knowledge between root causes and key features in a knowledge graph. Specifically, the knowledge correlation degree between candidate root causes and key features is calculated through the association paths between candidate root causes and key features in the fault knowledge graph. The conditional probabilities in the Bayesian network are then readjusted based on the knowledge correlation degree to assist the computational reasoning of the Bayesian network.

6. The method according to claim 1, characterized in that, During the initialization of the Bayesian network, the conditional probability of the Bayesian network is adjusted using a real-time data update factor: ， in, This represents the co-occurrence probability of features and root causes in real-time data. To introduce conditional probabilities for association with fault knowledge graphs. For real-time data update factors, This is the adjusted conditional probability.

7. The method according to claim 1, characterized in that, The process of obtaining the posterior probability of each candidate root cause includes: ， in, root cause The prior probability is obtained based on historical failure frequency statistics. This represents the probability of occurrence of the equipment failure feature vector. It represents the likelihood probability.

8. A power equipment fault root cause analysis system that integrates multi-source data, characterized in that, Used to perform the method described in any one of claims 1-7.

Images