A method for intelligent analysis and positioning of faults in electric power ground adjustment automation equipment

Abstract

The present invention relates to a method for intelligent analysis and positioning of faults of electric power ground adjustment automation equipment, which belongs to the technical field of power grid fault analysis and solves the problem that the existing fault analysis method is complicated and the data to be analyzed is incomplete, resulting in inaccurate fault positioning. The method includes: after preprocessing the fault operation and maintenance records and the expert knowledge base, constructing a fault knowledge graph and dividing it into multiple communities; real-time monitoring of the data of the electric power ground adjustment automation equipment to obtain fault information; obtaining an expert diagnosis scheme based on the fault information and the fault knowledge graph; obtaining a fault vector to be analyzed based on the fault information, and extracting related faults from multiple communities based on the fault vector to be analyzed; fusing the fault information and the related faults and passing them into a trained large inference model to obtain an inference diagnosis scheme; merging the expert diagnosis scheme and the inference diagnosis scheme to obtain a comprehensive scheme, and locating the fault according to the comprehensive scheme. The accuracy of multi-modal fault analysis and positioning is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid fault analysis, and in particular to a method for intelligent analysis and positioning of faults in electric power ground adjustment automation equipment. Background Art

[0002] With the continuous expansion of the scale of power systems and the surge in electricity demand, power ground dispatching automation equipment is the key to the safe and stable operation of power systems. Failure of power equipment will not only cause unexpected power outages in the power supply system, resulting in reduced economic benefits for power companies, but may also cause significant economic losses to users. Therefore, intelligent analysis and positioning of power equipment faults are of great significance for timely discovering equipment fault hazards, taking corresponding measures, and realizing scientific equipment fault diagnosis and maintenance.

[0003] In the prior art, there are technologies that combine Petri nets and neural networks to model and analyze complex system behaviors. However, when the fault conditions are multi-state and complex, Petri net modeling is difficult, and the training and reasoning of neural networks require a lot of computing resources, and the performance depends on the quality and quantity of training data. If the data is insufficient or noisy, the accuracy of the model will be affected.

[0004] In addition, some solutions use computer vision technology, mainly using infrared and ultraviolet images to detect faults in power equipment. This method has a high cost for acquiring images, and the defects that the algorithm can detect are relatively limited. In addition, the use of visible light image processing makes defect detection more difficult due to image illumination changes and rotational distortion caused by shooting time and angle. Summary of the invention

[0005] In view of the above analysis, an embodiment of the present invention aims to provide a method for intelligent analysis and positioning of electric power ground adjustment automation equipment faults, so as to solve the problem that the existing fault analysis method is complicated and the data to be analyzed is incomplete, resulting in inaccurate fault positioning.

[0006] The embodiment of the present invention provides a method for intelligent analysis and location of faults of electric power ground adjustment automation equipment, comprising the following steps:

[0007] After preprocessing the fault operation and maintenance records and expert knowledge base, a fault knowledge graph is constructed and divided into multiple communities;

[0008] Real-time monitoring of the data of power ground adjustment automation equipment to obtain fault information; obtain expert diagnosis solutions based on fault information and fault knowledge graphs;

[0009] The fault vector to be analyzed is obtained based on the fault information. The shortest path with the lowest cost is searched through an adaptively adjusted heuristic function to extract related faults from multiple communities based on the fault vector to be analyzed. The fault information and related faults are fused and passed into the trained large inference model to obtain the inference diagnosis solution.

[0010] The expert diagnosis scheme and the reasoning diagnosis scheme are combined to obtain a comprehensive scheme, and the fault is located according to the comprehensive scheme.

[0011] Based on the further improvement of the above method, after preprocessing the fault operation and maintenance records and the expert knowledge base, a fault knowledge graph is constructed, including:

[0012] Extracting fault nodes and fault attributes from the fault operation and maintenance records, and establishing a first association relationship between the fault nodes according to the similarity between the fault attributes;

[0013] Extracting expert knowledge nodes and knowledge attributes from the expert knowledge base, and establishing a second association relationship between the expert knowledge node and the fault node according to the similarity between the knowledge attribute of the expert knowledge node and the fault attribute of the fault node;

[0014] According to the fault nodes and their fault attributes, as well as the expert knowledge nodes and their knowledge attributes, and taking the first association relationship and the second association relationship as edges, a fault knowledge graph is constructed.

[0015] Based on the further improvement of the above method, the fault knowledge graph is divided into multiple communities, including:

[0016] For each fault node in the fault knowledge graph, the fault attributes are concatenated and passed into the trained Bert model to obtain the feature vector of each fault node;

[0017] Calculate the similarity between the feature vectors of the faulty nodes that establish the first association relationship, obtain the corresponding edge weights, and divide the faulty nodes into multiple communities using a community discovery algorithm;

[0018] According to the similarity between the feature vectors of each faulty node in each community, a similarity matrix is ​​constructed; and the weight of each faulty node in each community is calculated according to the similarity matrix;

[0019] According to the weight of each faulty node in each community and its feature vector, the feature vector of each community center is calculated.

[0020] Based on the further improvement of the above method, the feature vector of each community center point is calculated by the following formula:

[0021] ,

[0022] in, The feature vector representing a community center point, represents the number of faulty nodes in a community, Indicates the first The feature vector of the faulty node, Indicates the first The weight of the faulty node, Represents the similarity matrix of a community The feature vector of the faulty node and The feature vector of the faulty node The cosine similarity between ; .

[0023] Based on the further improvement of the above method, related faults are extracted from multiple communities according to the fault vector to be analyzed, including:

[0024] According to the similarity between the fault vector to be analyzed and the feature vector of each community center point, a target community is obtained; according to the similarity between the fault vector to be analyzed and the feature vector of each fault node in the target community, multiple target nodes are obtained;

[0025] The fault vector to be analyzed is taken as the node to be analyzed, and several boundary nodes of the target community where each target node is located are taken as the starting nodes. The shortest path and path cost from the node to be analyzed through each starting node to the target node are obtained respectively. According to the shortest path with the minimum path cost, multiple associated fault nodes are obtained, that is, the associated faults are extracted.

[0026] Based on the further improvement of the above method, the shortest path and path cost from the node to be analyzed through each starting node to the target node are obtained respectively, including:

[0027] If the starting node is the target node, the shortest path includes the node to be analyzed and the target node. The cosine distance between the node to be analyzed and the target node is calculated to obtain the corresponding path cost.

[0028] If the starting node is not the target node, then the heuristic search algorithm is used to search for the faulty node with the smallest node cost in the current target community starting from the starting node until the target node is reached. The node to be analyzed, the starting node, the faulty node with the smallest node cost, and the target node are connected in sequence to form the shortest path. The sum of the cosine distances between adjacent nodes on the shortest path is taken as the path cost.

[0029] Based on the further improvement of the above method, when searching for the faulty node with the smallest node cost starting from the starting node based on the heuristic search algorithm, the node cost of the current faulty node is the sum of the actual cost from the node to be analyzed to the current faulty node and the heuristic estimated cost from the current faulty node to the target node; the actual cost is obtained by calculating the cosine distance; if the current faulty node is the target node, the heuristic estimated cost is 0, otherwise, the heuristic estimated cost is calculated by an adaptively adjusted heuristic function.

[0030] Based on the further improvement of the above method, the heuristic estimated cost is calculated by the following formula:

[0031] ,

[0032] in, Indicates the current fault node The heuristic estimate cost of Represents the current fault node calculated according to the initial heuristic function To the target node The initial estimated cost of Indicates the current fault node To the target node The actual cost The difference from the initial estimated cost; represents the learning rate; represents the dimension of the feature vector; and Respectively represent the current fault node The feature vector and target node The feature vector of dimensional vector value.

[0033] Based on the further improvement of the above method, multiple associated fault nodes are obtained according to the shortest path with the minimum path cost, including:

[0034] If the shortest path with the minimum path cost only contains the node to be analyzed and the target node, the target node and its adjacent faulty nodes in the current community are taken as multiple associated faulty nodes;

[0035] Otherwise, the faulty nodes other than the node to be analyzed in the shortest path with the minimum path cost are taken as multiple associated faulty nodes.

[0036] Based on the further improvement of the above method, the fault nodes and fault attributes are extracted from the fault operation and maintenance records, including:

[0037] After removing noise characters and stop words in the fault operation and maintenance records, the natural language recognition method is used to extract the fault attributes in the historical fault records. The fault attributes include: fault name, fault device, fault time, fault area, fault type, fault phenomenon, fault cause, inspection steps and solutions;

[0038] Each fault operation and maintenance record is regarded as a fault node, the fault name in the fault attribute is regarded as the fault node name, and other fault attributes are regarded as the fault attributes of the corresponding fault node.

[0039] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0040] 1. It provides a multimodal fault analysis and location method. It automatically identifies fault information through real-time monitoring of power equipment, integrates it with knowledge graphs and reasoning big models, and responds in a timely manner to output expert and reasoning diagnosis solutions. It can not only improve the efficiency of fault analysis, but also quickly locate the root cause of the problem, speed up fault recovery, realize comprehensive analysis and location of faults from multiple angles, and enhance the automation and intelligence level of fault analysis and location.

[0041] 2. Integrate fault operation and maintenance records and expert knowledge into the graph structure to form a fault knowledge graph, which not only makes full use of a large amount of historical data and retains the experience and wisdom of experts, but also can better transmit and apply this knowledge through the graph structure.

[0042] 3. By adaptively adjusting the heuristic function to search for the shortest path with the lowest cost, the historical faults with strong correlation are obtained from the knowledge graph, which enriches the fault context features, facilitates the capture of more complex fault modes, improves the comprehensiveness of fault analysis, and improves the accuracy of fault location.

[0043] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. In the entire drawings, the same reference symbols represent the same components;

[0045] Figure 1 A flow chart of a method for intelligent analysis and location of faults in electric power ground adjustment automation equipment according to an embodiment of the present invention;

[0046] Figure 2The present invention is a block diagram of a method for intelligent analysis and location of faults in electric power ground adjustment automation equipment in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0048] A specific embodiment of the present invention discloses a method for intelligent analysis and location of faults in electric power ground adjustment automation equipment, such as Figure 1 and Figure 2 As shown, the following steps are included:

[0049] S1. After preprocessing the fault operation and maintenance records and expert knowledge base, a fault knowledge graph is constructed and divided into multiple communities.

[0050] It should be noted that the electric power dispatching automation equipment is the automation equipment of the regional electric power dispatching center. This embodiment mainly monitors, analyzes and processes faults of switches, routers, intrusion detection equipment, etc. in the dual-plane and dual-access networks in the regional electric power dispatching data network.

[0051] This embodiment makes full use of the operation and maintenance records of historical faults and expert experience, and digitizes and continuously updates the previous fault handling experience by building a fault knowledge graph, so as to help operation and maintenance personnel quickly analyze and locate faults.

[0052] It should be noted that the fault knowledge graph is constructed through the following steps, including:

[0053] ① Extract fault nodes and fault attributes from the fault operation and maintenance records, and establish the first association relationship between fault nodes based on the similarity between the fault attributes.

[0054] Specifically, fault operation and maintenance records are usually recorded in document format. The fault operation and maintenance records are preprocessed, that is, the text information of the fault operation and maintenance records is obtained, and after removing the noise characters and stop words, the natural language recognition method is used to extract the fault attributes in the historical fault records. The fault attributes include: fault name, fault equipment, fault time, fault area, fault type, fault phenomenon, fault cause, inspection steps and solutions; each fault operation and maintenance record is taken as a fault node, the fault name in the fault attributes is taken as the fault node name, and the other fault attributes are taken as the fault attributes of the corresponding fault node.

[0055] It should be noted that natural language recognition methods include: using named entity recognition methods to identify entities with specific meanings in texts, such as proper nouns (fault names, faulty equipment), specific terms (fault types), place names (fault areas), organization names, and time (fault time); using deep learning networks, such as recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and transformers; and using pre-trained language models such as GPT. In addition, for data with obvious formats, such as fault time or fault type, data can also be extracted through regular expressions.

[0056] Preferably, entity fusion and entity disambiguation are performed on the identified fault names to avoid duplication of fault nodes and unclear expressions. Entity fusion refers to merging the same entity information from different sources or in different forms into a unified entity representation; entity disambiguation refers to resolving the ambiguity that may exist for the same entity in different contexts.

[0057] A first association relationship is established between fault nodes based on the similarity between fault attributes. The similarity here includes the same attributes and the attribute difference within a threshold range. Specifically, if the fault attributes have the same fault type, fault device, fault area, or the difference in fault time is within a time threshold range, a first association relationship is established.

[0058] ② Extract expert knowledge nodes and knowledge attributes from the expert knowledge base, and establish a second association relationship between the expert knowledge node and the fault node based on the similarity between the knowledge attribute of the expert knowledge node and the fault attribute of the fault node.

[0059] It should be noted that the expert knowledge base is a general term for the knowledge and experience of experts in the field. If it is recorded in a document format, the natural language recognition method is also used to identify the knowledge attributes, including: expert knowledge name, fault name, faulty equipment, fault phenomenon, fault cause, diagnostic steps and solutions; if an expert knowledge system has been built, the knowledge attributes are extracted from the expert knowledge system.

[0060] Each piece of expert knowledge is regarded as an expert knowledge node, the expert knowledge name in the knowledge attribute is regarded as the expert knowledge node name, and other knowledge attributes are regarded as the knowledge attributes of the corresponding expert knowledge node.

[0061] Since the knowledge attribute and the fault attribute have the same attributes, after splicing and converting multiple identical attributes of each into attribute vectors, when the similarity of the two attribute vectors exceeds the similarity threshold, a second association relationship between the expert knowledge node and the fault node is established.

[0062] Incorporating expert knowledge and experience as part of the knowledge graph not only improves the professionalism and accuracy of fault analysis, but also facilitates the provision of more comprehensive and practical solutions.

[0063] ③ According to the fault nodes and their fault attributes, as well as the expert knowledge nodes and their knowledge attributes, and taking the first association relationship and the second association relationship as edges, a fault knowledge graph is constructed.

[0064] The fault knowledge graph constructed in this embodiment is a comprehensive knowledge graph including fault operation and maintenance records and expert knowledge, and further divides the fault knowledge graph into multiple communities, thereby supporting global search and local search, including:

[0065] For each fault node in the fault knowledge graph, the fault attributes are concatenated and passed into the trained Bert model to obtain the feature vector of each fault node;

[0066] Calculate the similarity between the feature vectors of the faulty nodes that establish the first association relationship, obtain the corresponding edge weights, and divide the faulty nodes into multiple communities using a community discovery algorithm;

[0067] According to the similarity between the feature vectors of each faulty node in each community, a similarity matrix is ​​constructed; and the weight of each faulty node in each community is calculated according to the similarity matrix;

[0068] According to the weight of each faulty node in each community and its feature vector, the feature vector of each community center is calculated.

[0069] Preferably, the feature vectors of each fault node and each expert knowledge node in the knowledge graph are stored in an ArangoDB graph database to facilitate efficient graph query.

[0070] It should be noted that the similarity as the edge weight is calculated using the cosine similarity method; the community discovery algorithm adopts the Leiden algorithm, and is improved by dynamically adjusting parameters and multi-starting methods. The algorithm is run multiple times and the optimal solution is selected to improve the stability and reliability of the results.

[0071] Specifically, the resolution parameter is a key parameter in the Leiden algorithm that affects the number and size of communities. A range of values ​​is set for the parameter, and the parameter value is adjusted incrementally by iteration at a fixed step size. After each adjustment, the community partitioning algorithm is rerun, and the modularity is used as the objective function to evaluate the results each time. The parameter value that produces the highest modularity value is selected as the optimal parameter.

[0072] Multi-start is to run the algorithm from multiple random starting points to increase the chance of finding the global optimal solution. For each selected parameter combination, the community partitioning algorithm is run to reduce the deviation caused by chance and ensure the stability of the results.

[0073] Furthermore, a similarity matrix is ​​constructed for each community finally divided , the matrix is ​​a square matrix with a size of , represents the number of faulty nodes in a community; if The failed node and There is no edge between the faulty nodes, then the corresponding element in the similarity matrix is 0, otherwise, For the The feature vector of the faulty node and The feature vector of the faulty node The cosine similarity between .

[0074] The weight of each faulty node in each community is calculated based on the similarity matrix, which is obtained by calculating the sum of the element values ​​corresponding to each faulty node and other faulty nodes in the similarity matrix; further, based on the weight of each faulty node in each community and its eigenvector, the eigenvector of each community center is calculated by first multiplying the weight of each faulty node in the community by its own eigenvector, then summing the result and dividing it by the sum of the weights of all faulty nodes in the community. The formula is as follows:

[0075] Formula (1),

[0076] in, The feature vector representing a community center point, Indicates the first The feature vector of the faulty node, Indicates the first The weight of a failed node.

[0077] According to the above formula, this embodiment better reflects the importance of nodes in the community by introducing node weights, and improves the stability of the community center through weighted averaging.

[0078] S2. Real-time monitoring of the data of the electric power ground adjustment automation equipment to obtain fault information; obtain expert diagnosis solutions based on the fault information and fault knowledge graph.

[0079] This embodiment acquires real-time data of the electric power ground adjustment automation equipment by deploying probes or installing sensors. These data include but are not limited to: equipment status, such as current, voltage and power; environmental parameters, such as temperature and humidity; operating parameters, such as load and frequency; log files, such as equipment startup, shutdown, error information, etc.; port status, such as port connection or disconnection status, error counter, etc.; network traffic, such as the amount of data in and out of each network port.

[0080] Clean all kinds of data collected in real time, store them in time series data according to the set format, and obtain monitoring indicator values ​​from the real-time collected data according to the preset monitoring indicators, identify whether the monitoring indicator values ​​have faults according to the alarm rules, and obtain fault information. The fault information includes clear fault equipment, fault time, fault area, fault type and fault phenomenon.

[0081] It should be noted that the alarm rules are automatically obtained based on the fault causes of the expert knowledge nodes in the fault knowledge graph. Exemplarily, the alarm rules include: the error count of the port per minute exceeds the error count threshold, and the number of port connection interruptions per hour exceeds the interruption count threshold.

[0082] It is understandable that the fault information in a unified format can also be obtained by receiving the fault-related information input by the operation and maintenance personnel through the input interface.

[0083] Furthermore, an expert diagnosis solution is obtained based on the fault information and the fault knowledge graph, including:

[0084] According to the expert knowledge nodes and their attributes established in the fault knowledge graph, expert knowledge nodes with a similarity to the fault information higher than a threshold are obtained, and the fault information is filled in according to the diagnosis template, as well as the fault cause, diagnosis steps and solutions corresponding to the expert knowledge nodes. The expert diagnosis plan is obtained and output to the operation and maintenance personnel, so that the operation and maintenance personnel can diagnose and solve the fault according to the expert experience.

[0085] Preferably, a user-friendly interface is provided to display the status of each device and the detected faults based on the real-time collected data. For example, if a router fails, the specific router and other affected network devices are displayed in the form of a network topology diagram.

[0086] S3. Obtain the fault vector to be analyzed based on the fault information. Use the adaptively adjusted heuristic function to search for the shortest path with the lowest cost. Extract related faults from multiple communities based on the fault vector to be analyzed. After fusing the fault information and related faults, pass them into the trained large inference model to obtain the inference diagnosis solution.

[0087] It should be noted that, unlike the traditional method of relying only on a single fault description when reasoning about the source of the fault and the solution, this embodiment enriches the contextual features by obtaining related fault cases, which facilitates the capture of more complex fault modes, improves the comprehensiveness of the fault analysis, and improves the accuracy of fault location.

[0088] Specifically, related faults are extracted from multiple communities according to the fault vector to be analyzed, including:

[0089] ①According to the similarity between the fault vector to be analyzed and the feature vector of each community center point, the target community is obtained; according to the similarity between the fault vector to be analyzed and the feature vector of each fault node in the target community, multiple target nodes are obtained.

[0090] It should be noted that the target community is obtained through the following steps: the cosine similarity between the fault vector to be analyzed and the feature vector of each community center point is calculated; according to the preset maximum number of communities and the similarity threshold, multiple communities whose cosine similarity is greater than the similarity threshold and whose number is less than or equal to the maximum number of communities are obtained in descending order of cosine similarity are selected as target communities; if there is no community whose cosine similarity is greater than the similarity threshold, multiple communities of corresponding number are obtained in descending order of cosine similarity according to the maximum number of communities as target communities.

[0091] Exemplarily, the similarity threshold is set to 0.9.

[0092] According to the cosine similarity between the fault vector to be analyzed and the feature vectors of each fault node in the target community, the fault nodes with a similarity greater than a threshold are obtained as target nodes; if there is no fault node with a cosine similarity greater than the similarity threshold, the current target community is deleted.

[0093] ② The fault vector to be analyzed is taken as the node to be analyzed, and several boundary nodes of the target community where each target node is located are taken as the starting nodes. The shortest path and path cost from the node to be analyzed through each starting node to the target node are obtained respectively. According to the shortest path with the minimum path cost, multiple associated fault nodes are obtained, that is, the associated faults are extracted.

[0094] It should be noted that the boundary nodes are the two end nodes of the edge connecting adjacent communities. They are the ambassadors of information exchange between communities, and cross-community information propagation will inevitably pass through these nodes. This embodiment uses an improved graph traversal algorithm to find a series of historical cases with strong correlation by searching for the shortest path with the lowest path cost from the node to be analyzed through the boundary node to the target node.

[0095] Specifically, the shortest path and path cost from the node to be analyzed through each starting node to the target node are obtained respectively, including:

[0096] If the starting node is the target node, the shortest path includes the node to be analyzed and the target node. The cosine distance between the node to be analyzed and the target node is calculated to obtain the corresponding path cost.

[0097] If the starting node is not the target node, then the heuristic search algorithm is used to search for the faulty node with the smallest node cost in the current target community starting from the starting node until the target node is reached. The node to be analyzed, the starting node, the faulty node with the smallest node cost, and the target node are connected in sequence to form the shortest path. The sum of the cosine distances between adjacent nodes on the shortest path is taken as the path cost.

[0098] Among them, when searching for the faulty node with the smallest node cost starting from the starting node based on the heuristic search algorithm, the node cost of the current faulty node is the sum of the actual cost from the node to be analyzed to the current faulty node and the heuristic estimated cost from the current faulty node to the target node; the actual cost is obtained by calculating the cosine distance (1-cosine similarity); if the current faulty node is the target node, the heuristic estimated cost is 0, otherwise, the heuristic estimated cost is calculated by an adaptively adjusted heuristic function.

[0099] It should be noted that, in this embodiment, the difference between the actual cost from the current faulty node to the target node and the initial estimated cost is used as the feedback cost, and then the initial heuristic function is adaptively adjusted; the formula is as follows:

[0100] Formula (2),

[0101] in, Indicates the current fault node The heuristic estimate cost of Represents the current fault node calculated according to the initial heuristic function To the target node The initial estimated cost of Indicates the current fault node To the target node The actual cost The difference from the initial estimated cost is used as the feedback cost; represents the learning rate, that is, the magnitude of each update of the heuristic function; represents the dimension of the feature vector; and Respectively represent the current fault node The feature vector and target node The feature vector of dimensional vector value.

[0102] From the above formula, we can see that if the feedback cost is greater than 0, it means that the current heuristic function underestimates the actual cost and needs to be adjusted up. Conversely, it overestimates the actual cost and needs to be adjusted down. Through continuous adjustment, the heuristic function will gradually approach a more accurate cost estimate.

[0103] Furthermore, multiple associated fault nodes are obtained according to the shortest path with the minimum path cost, including:

[0104] If the shortest path with the minimum path cost only contains the node to be analyzed and the target node, the target node and its adjacent faulty nodes in the current community are taken as multiple associated faulty nodes;

[0105] Otherwise, the faulty nodes other than the node to be analyzed in the shortest path with the minimum path cost are taken as multiple associated faulty nodes.

[0106] The fault attributes of multiple related fault nodes are obtained from the fault knowledge graph, and the fault information is fused with the related faults, that is, the fault information is spliced ​​with the fault attributes of multiple related faults to form a more comprehensive context representation, and then passed into the trained inference big model. The inference big model analyzes the root cause and potential solutions of the fault, infers the cause of the fault based on the analysis results, provides solutions for the identified cause of the fault, and obtains the inference diagnosis plan. Among them, the trained inference big model is obtained by training and fine-tuning the pre-trained GPT series big model using historical fault information. The inference diagnosis plan includes: fault causes and solutions.

[0107] S4. Combine the expert diagnosis plan and the reasoning diagnosis plan to obtain a comprehensive plan, and locate the fault according to the comprehensive plan.

[0108] It should be noted that the expert diagnosis scheme and the reasoning diagnosis scheme are combined to obtain a comprehensive scheme, which is displayed through the user interface. If the expert diagnosis scheme and the reasoning diagnosis scheme have different fault causes, the expert diagnosis scheme is given priority.

[0109] During implementation, the operation and maintenance personnel can locate the cause of the fault according to the comprehensive plan and handle the fault according to the detection steps of the comprehensive plan; they can also first handle the steps according to the detection steps in the expert diagnosis plan, and if the fault cannot be solved, then handle it according to the reasoning diagnosis plan; if a new fault occurs during the processing, enter the new fault description, execute step S3 to obtain a new reasoning diagnosis plan, and diagnose and solve the fault here according to the new reasoning diagnosis plan.

[0110] Preferably, feedback from maintenance personnel on the final cause of the fault and the solution, or supplementary fault information provided by maintenance personnel, is received through an interactive interface and updated to the fault knowledge graph to update and expand knowledge. This also eliminates the need to retrain the large inference model, thus reducing the cost of knowledge update. This interactive feedback and learning maintains the timeliness and accuracy of the fault knowledge graph and large model, and continuously optimizes the accuracy of fault analysis and location.

[0111] Preferably, detailed fault reports are generated regularly, and statistical analysis is performed from the dimensions of fault type, fault time, faulty equipment, etc., to help the operation and maintenance team to prevent faults and optimize expert knowledge.

[0112] Compared with the prior art, the method for intelligent analysis and location of faults of electric ground adjustment automation equipment provided in this embodiment provides a multimodal fault analysis and location method. It automatically identifies fault information through real-time monitoring of power equipment, integrates with knowledge graph and reasoning large model, and responds to output expert and reasoning diagnosis schemes in time, which not only improves the efficiency of fault analysis, but also quickly locates the root cause of the problem, accelerates fault recovery, realizes comprehensive analysis and location of faults from multiple angles, and enhances the automation and intelligence level of fault analysis and location. Integrating fault operation and maintenance records and expert knowledge into the graph structure to form a fault knowledge graph not only makes full use of a large amount of historical data, retains the experience and wisdom of experts, but also can better transmit and apply this knowledge through the graph structure. Through the adaptively adjusted heuristic function to search for the shortest and least costly path, historical faults with strong correlation are obtained from the knowledge graph, enriching the fault context features, facilitating the capture of more complex fault modes, and improving the comprehensiveness of fault analysis and the accuracy of fault location.

[0113] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0114] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for intelligent analysis and location of faults in electric power ground adjustment automation equipment, characterized in that: The following steps are involved: After preprocessing the fault operation and maintenance records and expert knowledge base, a fault knowledge graph is constructed and divided into multiple communities; Real-time monitoring of the data of the electric power ground adjustment automation equipment to obtain fault information; obtaining an expert diagnosis solution based on the fault information and the fault knowledge graph; Obtain a fault vector to be analyzed according to the fault information, extract associated faults from the multiple communities according to the fault vector to be analyzed by searching for the shortest path with the lowest cost through an adaptively adjusted heuristic function; merge the fault information and the associated faults and pass them into a trained large inference model to obtain an inference diagnosis solution; Combining the expert diagnosis scheme and the reasoning diagnosis scheme to obtain a comprehensive scheme, and locating the fault according to the comprehensive scheme; The extracting associated faults from the multiple communities according to the fault vector to be analyzed includes: obtaining a target community according to the similarity between the fault vector to be analyzed and the feature vector of each community center point; obtaining multiple target nodes according to the similarity between the fault vector to be analyzed and the feature vector of each fault node in the target community; taking the fault vector to be analyzed as the node to be analyzed, taking several boundary nodes of the target community where each target node is located as the starting node, respectively obtaining the shortest path and path cost from the node to be analyzed through each starting node to the target node, and obtaining multiple associated fault nodes according to the shortest path with the minimum path cost, that is, extracting the associated fault; The method of respectively obtaining the shortest path and path cost from the node to be analyzed to the target node via each starting node includes: if the starting node is the target node, the node to be analyzed and the target node are included on the shortest path, and the cosine distance between the node to be analyzed and the target node is calculated to obtain the corresponding path cost; if the starting node is not the target node, the fault node with the smallest node cost is searched in sequence from the starting node in the current target community based on a heuristic search algorithm until the target node is reached, and the node to be analyzed, the starting node, the fault nodes with the smallest node cost, and the target node are sequentially connected to form the shortest path, and the sum of the cosine distances between adjacent nodes on the shortest path is used as the path cost.

2. The method for intelligent analysis and location of faults of electric power ground adjustment automation equipment according to claim 1, characterized in that: After preprocessing the fault operation and maintenance records and the expert knowledge base, a fault knowledge graph is constructed, including: Extracting fault nodes and fault attributes from the fault operation and maintenance records, and establishing a first association relationship between the fault nodes according to the similarity between the fault attributes; Extracting expert knowledge nodes and knowledge attributes from the expert knowledge base, and establishing a second association relationship between the expert knowledge node and the fault node according to the similarity between the knowledge attribute of the expert knowledge node and the fault attribute of the fault node; According to the fault nodes and their fault attributes, as well as the expert knowledge nodes and their knowledge attributes, and taking the first association relationship and the second association relationship as edges, a fault knowledge graph is constructed.

3. The method for intelligent analysis and location of faults of electric power ground adjustment automation equipment according to claim 2, characterized in that: The fault knowledge graph is divided into multiple communities, including: For each fault node in the fault knowledge graph, the fault attributes are concatenated and passed into the trained Bert model to obtain the feature vector of each fault node; Calculate the similarity between the feature vectors of the faulty nodes that establish the first association relationship, obtain the corresponding edge weights, and divide the faulty nodes into multiple communities using a community discovery algorithm; Constructing a similarity matrix according to the similarity between the feature vectors of each faulty node in each community; calculating the weight of each faulty node in each community according to the similarity matrix; According to the weight of each faulty node in each community and its feature vector, the feature vector of each community center is calculated.

4. The method for intelligent analysis and location of faults of electric power ground adjustment automation equipment according to claim 3 is characterized in that: The characteristic vector of each community center point is calculated by the following formula: ， in, The feature vector representing a community center point, represents the number of faulty nodes in a community, Indicates the first The feature vector of the faulty node, Indicates the first The weight of the faulty node, Represents the similarity matrix of a community The feature vector of the faulty node and The feature vector of the faulty node The cosine similarity between ; .

5. The method for intelligent analysis and location of faults of electric power ground adjustment automation equipment according to claim 3, characterized in that: When searching for the faulty node with the smallest node cost in sequence from the starting node based on the heuristic search algorithm, the node cost of the current faulty node is the sum of the actual cost from the node to be analyzed to the current faulty node and the heuristic estimated cost from the current faulty node to the target node; The actual cost is obtained by calculating the cosine distance; If the current faulty node is the target node, the heuristic estimated cost is 0; otherwise, the heuristic estimated cost is calculated by an adaptively adjusted heuristic function.

6. The method for intelligent analysis and location of faults of electric power ground adjustment automation equipment according to claim 5, characterized in that: The heuristic estimated cost is calculated by the following formula: ， in, Indicates the current fault node The heuristic estimate cost of Represents the current fault node calculated according to the initial heuristic function To the target node The initial estimated cost of Indicates the current fault node To the target node The actual cost The difference from the initial estimated cost; represents the learning rate; represents the dimension of the feature vector; and Respectively represent the current fault node The feature vector and target node The feature vector of dimensional vector value.

7. The method for intelligent analysis and location of faults of electric power ground adjustment automation equipment according to claim 6, characterized in that: The method of obtaining a plurality of associated fault nodes according to the shortest path with the minimum path cost includes: If the shortest path with the minimum path cost only contains the node to be analyzed and the target node, the target node and its adjacent faulty nodes in the current community are taken as multiple associated faulty nodes; Otherwise, the faulty nodes other than the node to be analyzed in the shortest path with the minimum path cost are taken as multiple associated faulty nodes.

8. The method for intelligent analysis and location of faults of electric power ground adjustment automation equipment according to claim 2, characterized in that: The extracting of the fault node and fault attributes from the fault operation and maintenance record includes: After removing noise characters and stop words in the fault operation and maintenance records, the natural language recognition method is used to extract the fault attributes in the historical fault records. The fault attributes include: fault name, fault device, fault time, fault area, fault type, fault phenomenon, fault cause, inspection steps and solutions; Each fault operation and maintenance record is regarded as a fault node, the fault name in the fault attribute is regarded as the fault node name, and other fault attributes are regarded as the fault attributes of the corresponding fault node.

Images