Electric power material detection system based on improved association rule and knowledge graph

Abstract

The invention provides an electric power material detection system based on an improved association rule and a knowledge graph, and belongs to the field of electric power material detection. According to the system, an association analysis model for equipment detection is established by using an association rule algorithm, a hidden association relationship between unqualified detection items is mined, and a judgment basis for the disqualification of the related items is provided. In addition, in combination with information such as a material and equipment detection report and expert knowledge, knowledge such as entities, attributes and relationships is extracted from semi-structured data by adopting a natural language processing technology, a material and equipment detection management knowledge graph is constructed in a graph database Neo4j, a systematic material detection knowledge base is provided, network visualization is carried out, and the material and equipment detection management knowledge graph is established. And information query and management are facilitated. According to the invention, the detection efficiency of the electric power material equipment is improved, and increasing detection requirements are met.

Description

Technical Field

[0001] The present invention belongs to the field of electric power materials, and in particular relates to an electric power material detection system based on improved association rules and knowledge graphs. Background Art

[0002] High-quality power grid equipment and materials are paramount to ensuring grid security. Grid companies attach great importance to the testing of grid-connected products, and the quality inspection of grid materials is a top priority in material quality supervision. As a crucial control measure for power material management, grid material quality inspection connects material production sites with construction sites, impacting the safe and stable operation of power grid projects and holding crucial importance for power construction, grid security, and operation and maintenance. Grid material quality inspection involves methods such as witnessed testing, spot checks, and third-party testing. With the continuous increase in electricity demand, the expansion of the power grid, and the increasing supply of materials, the corresponding material inspection workload is also increasing year by year. Therefore, there is an urgent need to improve inspection efficiency. Existing research has primarily focused on condition-based maintenance, condition evaluation, and fault diagnosis during the equipment operation phase, with little attention paid to the analysis and research of equipment quality during the grid connection phase. As a result, while large amounts of test data can effectively characterize equipment performance, their use is limited to determining equipment compliance, lacking effective analysis and application. As power systems increasingly rely on information technology, data mining techniques, which extract implicit, unknown, non-trivial, and potentially applicable information or patterns from big data, are increasingly being used in power systems, particularly in typical big data scenarios.

[0003] Association rules are typically used to reflect the interdependencies and relationships between one thing and other things. Association rules can be applied in a variety of scenarios. In commercial sales, association rules can be used for cross-selling to maximize revenue; in healthcare, they can identify possible treatment combinations. These are all examples of association rule mining problems. Association rule mining objectively analyzes the hidden relationships between influencing factors and behaviors, making it more relevant to the realities of power system operation and production. Compared to other data mining methods, it has a wider range of applications and is often used for causal analysis, such as transformer fault diagnosis based on association rules. For example, by summarizing and analyzing the variables influencing transformer status, methods for objectively evaluating transformer status are being studied based on association rule mining. Furthermore, association rule mining is also used to analyze the operating status of equipment within power plants.

[0004] A knowledge graph is a technical approach that uses a graphical model to describe the relationships between knowledge and objects. A knowledge graph consists of nodes and edges. A node can be an entity, such as a device or a person, or an abstract concept, such as a knowledge graph or artificial intelligence. Edges typically represent relationships between entities, such as between a teacher and a student, or between family members, or entity attributes, such as device name or device type. Knowledge graphs are generally represented as triples, namely (entity A, relationship, entity B) or (entity, relationship, attribute). Entities and attributes are nodes, and relationships are directed edges connecting two nodes. Triples connect entities or attributes and the relationships between them, effectively organizing large-scale network information with minimal effort. Knowledge graphs enable the visual storage, organization, and management of entities, attributes, and relationships, forming a structured knowledge representation. This graph data structure is easily understood and accepted by humans, and easily recognized and processed by computers. A knowledge graph is essentially a knowledge base, a knowledge network that connects and organizes entities and attributes through relationships. It identifies, filters, and infers complex relationships between entities from unstructured knowledge, efficiently performing relational operations and analysis such as storage and querying. It has widespread application in search engines, social networks, medical education, and the power sector. In the electrical sector, knowledge graphs can integrate dispersed knowledge within the power grid and effectively mine useful underlying rules from large amounts of textual information within the power system. Furthermore, the graph data structure of knowledge graphs greatly facilitates human understanding. Currently, research on knowledge graphs in the power sector is still in its infancy, with relevant literature primarily focusing on application exploration and macro-framework design. Summary of the Invention

[0005] In response to the above-mentioned deficiencies in the prior art, the present invention provides an electric power material detection system based on improved association rules and knowledge graphs, which effectively solves the problems of insufficient efficiency, in-depth analysis and other aspects of electric power material detection, and improves the advancement and rationality of material detection.

[0006] To achieve the above objectives, the present invention adopts a technical solution: a power material detection system based on improved association rules and knowledge graph, comprising: Material and equipment detection module, used to obtain and process power material and equipment detection data; The association rule mining module is used to construct a frequent pattern tree for unqualified test items based on the processing results using the frequent pattern growth algorithm, and to mine strong association rules based on the frequent pattern tree. In particular, a time decay factor is introduced into the association rule mining; The knowledge retrieval module is used to define the ontology layer and extract knowledge in the data layer. At the same time, it repeatedly updates the ontology layer, adds the newly acquired knowledge representation to the ontology layer, and obtains the power material equipment inspection and management knowledge graph based on the strong association rules obtained by mining. Among them, the power material equipment inspection and management knowledge graph is used to display the power material inspection results.

[0007] The beneficial effects of the present invention are as follows: the present invention analyzes a large amount of power material equipment detection data, mines the correlation between unqualified detection items, and constructs a material equipment detection management knowledge graph based on knowledge graph technology, comprehensively improves the carrying capacity of power grid material detection, effectively solves the problems of insufficient capabilities in work efficiency, in-depth analysis, etc., and improves the advancement and rationality of material detection.

[0008] Furthermore, the processing of the power material equipment detection data is specifically as follows: Based on the acquired power material equipment inspection data, the inspection items that do not meet the requirements are sorted into transaction data sets D , complete the processing of power material equipment inspection data, where the transaction data set includes the material equipment name, equipment serial number and unqualified inspection items.

[0009] Furthermore, the mining of strong association rules is specifically as follows: Use the frequent pattern growth algorithm to scan the processed transaction data set to obtain frequent item sets and create the header table H; Use frequent item sets to filter infrequent items in transaction data sets and generate frequent pattern trees; Based on the header table H, recursive operations are performed on the frequent pattern tree to mine strong association rules, wherein a time decay factor is introduced into the association rule mining.

[0010] Furthermore, the recursive operation is performed on the frequent pattern tree, which is specifically: On the frequent pattern tree, start from the last item d in the header table H and process each item in turn, and add the current processed item to the base item set base-itemset={d} with an initial value of empty. Each base item set generated is a new frequent item set, and all nodes of item d on the frequent pattern tree can be obtained from the header table H; Scan the path from each node d to the root node on the frequent pattern tree, count all the items that appear and the number of supports for each item, and save the statistical results to the sub-header table subH with the same structure as the header table H. d In which, the support number of all items on the same path is the support number of node d on the path, and all items in the header table H are sorted from small to large according to the support number. In the sorting process, the time decay factor is introduced; Scan all the paths from node d to the root node again, and put all the items on the path into the subheader table subH d Sort in the order of subheader table subH d , and add the ordered item sets and support numbers to the subtree subH d superior; If the sub-header table subH d If it is not empty, construct a conditional subtree subT for the base item set d ; Recursively process the subheader table subH d and conditional subtree subT d ; Based on the recursive processing results, the currently processed item is deleted from the current base-itemset, strong association rules are mined, and the next item in the currently processed header table H is processed.

[0011] Furthermore, the support degree of the association rule is expressed as follows:

[0012] in, Represents the support of association rules, A and B are item sets A is a true subset of , and A∩B=Ø, A is called the premise, B is called the sequel of the association rule, Indicates the first k elements, I Represents a transaction dataset D The set of all distinct items in , Indicates that the transaction set also contains items A and B The probability of Indicates that the transaction set also contains items A and B The number of samples.

[0013] Furthermore, the confidence expression of the association rule is as follows:

[0014] in, represents the confidence of the association rule, express A Appears in the transaction B The conditional probability of Indicates that the transaction set also contains items A and B The number of samples, Indicates that the transaction set contains items A The number of samples.

[0015] Furthermore, the construction of the knowledge graph for power material equipment inspection and management is specifically as follows: A top-down and bottom-up hybrid construction approach is used to define the ontology layer and extract knowledge in the data layer to obtain knowledge representation; The knowledge representation is corrected, and the ontology layer is repeatedly updated in the process of constructing the knowledge graph of power material equipment inspection and management. The newly acquired knowledge representation is added to the ontology layer to obtain the knowledge graph of power material equipment inspection and management.

[0016] Furthermore, the correction includes: Data processing: modify and / or delete data that is not processed in the correct format during the knowledge extraction process; Knowledge fusion: Processing knowledge from different sources to form globally unified knowledge identification and association; Manual correction: Eliminate erroneous data and correct data that does not meet the description requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the system structure of the present invention.

[0018] Figure 2 It is a structural framework diagram of the present invention.

[0019] Figure 3 This is a schematic diagram of the FP-Tree structure generated by the FP-Growth-based association rule mining module in the present invention.

[0020] Figure 4 This is an example of a transaction data set input into the FP-Growth-based association rule mining module of the present invention.

[0021] Figure 5 This is a schematic diagram of an FP-Tree instance constructed by the FP-Growth-based association rule mining module in the present invention.

[0022] Figure 6 This is an illustration of the subtrees and subheads in the FP-Tree instance constructed by the FP-Growth-based association rule mining module in the present invention.

[0023] Figure 7 A schematic diagram of the process of constructing a knowledge graph for power material equipment inspection and management based on the knowledge graph in the present invention.

[0024] Figure 8 Schematic diagram of the knowledge graph for detection and management of electric power materials and equipment constructed by the knowledge retrieval module based on the knowledge graph in the present invention. DETAILED DESCRIPTION

[0025] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0026] Example like Figure 1-Figure 2 As shown, the present invention provides an electric power material detection system based on improved association rules and knowledge graph, including: The material and equipment detection module is used to obtain and process the power material and equipment detection data, specifically: Based on the acquired power material equipment inspection data, the inspection items that do not meet the requirements are sorted into transaction data sets D , complete the processing of power material equipment inspection data, where the transaction data set includes the material equipment name, equipment serial number and unqualified inspection items.

[0027] In this embodiment, the material equipment detection module will be used to obtain detailed power material equipment detection data and organize the data format, that is, to organize unqualified detection items into a transaction data set, which includes the material equipment name, equipment serial number and unqualified detection items. N transaction data set D ={ T 1, T 2,…, T N},in, T j ( j = 1, 2, ..., N) is the transaction data set D No. j Transaction sets. T j The elements of a transaction set are called items I k ( k =1, 2, …, p), I ={ I 1, I 2,…, I p} is the data set D The set of all distinct items in a transaction. I k is the name of the test item. All unqualified test items of the same equipment form a transaction set. T j So that subsequent modules can use the data.

[0028] The association rule mining module is used to construct a frequent pattern tree (Frequent Pattern Tree, FP-Tree, such as Figure 3 As shown in Figure 2), and based on the frequent pattern tree, strong association rules are mined. In the association rule mining, a time decay factor is introduced, which is specifically: Using the frequent pattern growth algorithm, the processed transaction data set is scanned to obtain frequent item sets and create a header table H. The frequent item sets are used to filter the non-frequent items in the transaction data set to generate a frequent pattern tree. Based on the header table H, recursive operations are performed on the frequent pattern tree to mine strong association rules, wherein a time decay factor is introduced into the association rule mining.

[0029] Perform recursive operations on the frequent pattern tree, which are as follows: on the frequent pattern tree, start from the last item d in the header table H and process each item in turn, and add the currently processed item to the base item set base-itemset={d} with an initial value of empty. Each base item set generated is a new frequent item set, and all nodes of item d on the frequent pattern tree can be obtained from the header table H; scan the path from each node d to the root node on the frequent pattern tree, count all the items that appear and the number of supports for the items, and save the statistical results to the sub-header table subH with the same structure as the header table H d In which, the support number of all items on the same path is the support number of node d on the path, and all items in the header table H are sorted from small to large according to the support number. In the sorting process, the time decay factor is introduced; scan all the paths from node d to the root node again, and sort all the items on the path according to the sub-header table subH d Sort in the order of subheader table subH d , and add the ordered item sets and support numbers to the subtree subH d On; If the subheader table subH d If it is not empty, construct a conditional subtree subT for the base item set d ; Recursively process the subheader table subH d and conditional subtree subT d ; Based on the recursive processing results, delete the currently processed item from the current base-itemset, mine the strong association rules, and process the next item in the currently processed header table H. In this embodiment, the association rule mining module based on FP-Growth constructs an FP-Tree based on the unqualified power material detection item set, mines strong association rules based on the tree, and finally outputs the rule predecessor item set and the rule successor item set.

[0030] In this embodiment, Figure 2-Figure 5 As shown in the figure, the association rule mining module based on FP-Growth will construct an FP-Tree based on the set of unqualified power material inspection items, and introduce a time decay factor in this process to make the recent inspection data have a greater impact on the association rules.

[0031] In this embodiment, the association rule mining module based on FP-Growth constructs an FP-Tree based on the set of unqualified power material inspection items, mines strong association rules based on the tree, and finally outputs the rule leader item set and rule follower item set. The smallest indivisible information unit (i.e., record) in the database is called an item (or project), which is represented by the symbol i The set of items is called an itemset. I ={ i 1, i 2,…, i k} is an item set, I The number of middle terms is k , then the set I called k -item set. Each transaction is an item set. I ={ i 1, i 2,…, i k} is the complete set of all items in the database, then each transaction t i The corresponding item sets are all subsets of I. Transaction database T={ t 1, t 2,…, t n} is a set of transactions with unique identifiers. Association rules are of the form A The expression for item set B is: A and B, respectively, are proper subsets of item set I, and A∩B=Ø. A is called the antecedent or leading condition, and B is called the consequence or sequel of the association rule. The association rule reflects the pattern that when items in A appear, items in B also appear. Here, A and B refer to item sets. For example, {A, B}→{C} means that if a piece of equipment (such as a transformer) has problems in both test items A and B, then test item C will also have problems.

[0032] The support of an association rule is the ratio of the number of transactions in a transaction set that contain both items A and B to the total number of transactions in the transaction set. It reflects the probability that the items contained in A and B appear simultaneously in the transaction set. The calculation formula is:

[0033] in, Represents the support of association rules, A and B are item sets A is a true subset of , and A∩B=Ø, A is called the premise, B is called the sequel of the association rule, Indicates the first k elements, I Represents a transaction dataset D The set of all distinct items in , Indicates that the transaction set also contains items A and B The probability of Indicates that the transaction set also contains items A and B The number of samples.

[0034] The support reflects the effectiveness of the association rule, indicating the importance or probability of the association rule in the transaction database. The higher the support, the higher the association degree.

[0035] The confidence of an association rule is the ratio of the number of transactions that contain both A and B to the number of transactions that contain A. It reflects the conditional probability P(B|A) that B appears in a transaction that contains A. The calculation formula is:

[0036] in, represents the confidence of the association rule, express A Appears in the transaction B The conditional probability of Indicates that the transaction set also contains items A and B The number of samples, Indicates that the transaction set contains items A The number of samples.

[0037] Confidence reflects the certainty of the association rule and indicates the credibility of the association rule. The higher the confidence, the higher the credibility of its inference. Usually the support and confidence must be greater than or equal to the threshold set by humans to indicate that there is an association between items. If the support of a certain item set is greater than or equal to the minimum support, then the item set is called a frequent item set. Finding frequent item sets is the first step in finding strong association rules. Frequent item sets mean that the items always appear together or in pairs. If an association rule If the support is greater than or equal to the minimum support and the confidence is greater than or equal to the minimum confidence, then it is called an association rule. The present invention mines the strong association rule item set as the final output.

[0038] The FP-Growth algorithm used is an algorithm that does not need to generate candidate item sets. The algorithm only needs to scan the data set twice: the first scan of the data set obtains frequent 1-item sets; the second scan of the data set uses the frequent 1-item sets to filter the infrequent items in the data set and generates an FP-Tree (such as Figure 3 As shown in the figure, a recursive algorithm is then executed on the FP-Tree to mine all frequent item sets. Among them, those that meet the strong association rule conditions are the final strong association rule item sets required. In addition, a time decay factor is introduced in the association rule mining process to make the recent detection data have a greater impact on the association rules, avoiding the interference of historical data on the identification of current key parameters. Figure 4 The transaction dataset is used as an example to illustrate the mining steps of the FP-Growth algorithm (here the user-defined minimum support number is set to 3).

[0039] 1) Scan the data set, count the support of each item, and save the items whose support is greater than or equal to the minimum support into a header table H in descending order of support. Figure 5 (a) is the header table H of the data statistics in the example (the link field in the header table stores all node pointers for each item in the tree; it is also possible to store only the first node of each item in the tree in the link, and the pointer to the next node with the same name is stored in the previous node in sequence).

[0040] 2) Scan the data set again and add all transaction item sets in the data set to a tree. First, delete the infrequent items in the transaction item set (items not in the header table are infrequent). The remaining items in the transaction item set are added to the tree in the order of the header table, and the support count of each corresponding node is increased by 1. Figure 5 (b) Figure 3 The result after the first transaction item set is added to the tree, and the newly created node pointer is also saved in the link of the corresponding item in the header table. The number on the node indicates the support count of the node; Figure 5(c) is the result after the first two transaction item sets are added to the tree. When the third transaction item set {a, b, c, d} is added to the tree, since it can share nodes a and c with the path root-acd, it only needs to increase the support count of the two existing nodes by 1, and then add the remaining items to the descendants of node c in sequence; Figure 5 (d) is the result after the first three transaction item sets are added to the tree; Figure 5 (e) is the result after all transaction item sets are added to the tree (to make the FP-Tree structure clearer, Figure 5 (e) The link information in the header table is not identified).

[0041] 3) Start from the last item in the header table H and process each item in turn. For example, for the last item d in the header table H, first add the item to the base-itemset whose initial value is empty, that is, base-itemset={d}; from the link information in the header table H, we can find all the nodes of item d on the FP-Tree, scan the path from each node d to the root node, count all the items that appear and the support number of the items, where the support number of all items on the same path is the support number of node d on the path, and save the statistical results to a new header table subH with the same structure as the header table H d (This header table is also called a sub-header table). Similarly, this sub-header table only stores items whose support number is greater than or equal to the minimum support number 3. At the same time, all items in the header table are sorted from small to large according to the support number. In the sorting process, the introduction of the time decay factor will affect its sorting order. The newly created header table subH d like Figure 6 (a) shows that; scan all the paths from node d to the root node again, and put all the items on the path into the subheader table subH d Sort in the order of , delete the items that are not in the subheader table, and then add the ordered item set and support number (the support number of all items is equal to the support number of node d) to a subtree subH d (This tree is also called the conditional subtree of the current base item set {d}, or subtree for short), such as Figure 6 (b) As shown; recursively process the subheader table subH according to the above processing method d and subtree subT d , we can get two new base item sets {dc}, {dca}, {db}, {dba} and {da} respectively. Figure 6 (c) is the subtree and child of the item set {dc}.

[0042] 4) Delete the currently processed item from the current base-itemset.

[0043] 5) Continue processing the next item in the currently processed header table.

[0044] The above steps can be briefly described as follows: 1) Scan the data set and create the header table H, such as Figure 5 (a) Middle head table.

[0045] 2) Scan the data set and create FP-Tree, such as Figure 5 FP-Tree in (e).

[0046] 3) Process all items in the header table H in sequence (starting from the last item, assuming the currently processed item is Q): 3.1) Add the current processing item Q to a base item set with an initial value of empty (each new base item set is a new frequent item set); 3.2) Scan all nodes Q to the root node path on the FP-Tree and build a subheader table subH Q The specific method of constructing the subheader table is as shown in the above example. d The construction ( Figure 6 (a)); 3.3) If the subheader table subH Q If it is not empty, a conditional subtree subT is constructed for the base item set. Q The specific construction method is the same as the conditional subtree subT in the above example d Otherwise, execute step 4); 3.4) Recursively process the sub-header table subH according to the method of processing the header table H Q and subtree subT d .

[0047] 4) Delete the currently processed item from the base item set.

[0048] 5) Process the next item in the current processing header table.

[0049] The knowledge retrieval module is used to define the ontology layer and extract knowledge in the data layer. At the same time, the ontology layer is repeatedly updated, and the newly acquired knowledge representation is added to the ontology layer. Based on the strong association rules obtained by mining, the knowledge graph of power material equipment inspection and management is obtained. The knowledge graph of power material equipment inspection and management is used to display the results of power material inspection. The knowledge graph of power material equipment inspection and management is constructed as follows: A hybrid construction method of top-down and bottom-up is adopted to define the ontology layer, and knowledge extraction is performed in the data layer to obtain the knowledge representation. The knowledge representation is corrected, and the ontology layer is repeatedly updated in the process of constructing the knowledge graph of power material equipment inspection and management. The newly obtained knowledge representation is added to the ontology layer to obtain the knowledge graph of power material equipment inspection and management.

[0050] Correction is carried out, which includes: data processing: modifying and / or deleting data that is not processed in the correct format during the knowledge extraction process; knowledge fusion: processing knowledge from different sources to form a globally unified knowledge identification and association; manual correction: eliminating erroneous data and correcting data that does not meet the description requirements.

[0051] In this embodiment, Figure 7 As shown, the knowledge retrieval module based on the knowledge graph adopts a hybrid construction approach combining top-down and bottom-up approaches. Specifically, the ontology layer is first defined, then knowledge is extracted at the data layer, while the ontology layer is repeatedly updated. Finally, the newly acquired knowledge is added to the ontology to complete the graph construction. The specific operation is as follows: After clarifying the design requirements of the material and equipment inspection management knowledge graph, the present invention studies and analyzes relevant knowledge in the field of material and equipment inspection, summarizes important terms and concepts, and organizes unstructured and semi-structured data such as material inspection reports. Under the premise of clarifying the rationality of concept division, the material inspection knowledge system and the relationships between knowledge points are accurately described, ambiguous and easily confused terms are corrected, and a precise and complete set of core concepts for material inspection is formed. Secondly, entities and their attributes need to be extracted from the data. Entity extraction aims to extract words representing entities, as well as their attributes and attribute values, from the material inspection analysis corpus and perform part-of-speech tagging. Through the above process, the triple word table <entity, attribute, attribute value> can be used to describe the attributes and attribute values ​​of entities. Attributes play the role of describing entities in the knowledge graph. Next, it's necessary to extract semantic relationships between entities from the material inspection analysis corpus and describe the connections between them. This is an essential step for providing query functionality and visualization. This process is often referred to as relationship extraction. Relationships are also represented as triples: <entity 1, relationship, entity 2> or <entity, relationship, attribute>. Entities and attributes are nodes, and relationships are directed edges connecting two nodes. Triples are linked by shared entities or attributes and their relationships, forming a network-like knowledge graph. Finally, based on the Neo4j graph database, the material and equipment inspection knowledge graph is constructed and managed using graph nodes and relationship edges. Neo4j stores structured data in a topological graph, enabling centralized management of large amounts of data. The graph model consists of nodes and relationships. Nodes store entity information, and nodes are linked by relationships. Properties and labels of nodes and relationships are stored as key-value pairs.

[0052] In this embodiment, knowledge extraction is as follows: Figure 7As shown, the extraction process is categorized into entity extraction, attribute extraction, and relationship extraction. Entity extraction aims to extract words representing entities, their attributes, and their values ​​from the material inspection and analysis corpus, and perform part-of-speech tagging. The triplet vocabulary <entity, attribute, attribute value> can be used to describe an entity's attributes and their values. Attributes serve to describe entities in the knowledge graph. Relationship extraction aims to extract semantic relationships between entities from the material inspection and analysis corpus, elaborating on the connections between entities. This is essential for providing query functionality and visualization. Relationships are generally represented as triples: <entity 1, relationship, entity 2> or <entity, relationship, attribute>. Entities and attributes are nodes, and relationships are directed edges connecting two nodes. Triples are linked by shared entities or attributes and their relationships, forming a network-like knowledge graph.

[0053] In this embodiment, knowledge correction involves data processing, knowledge fusion, and manual correction. Data processing corrects or deletes data that was not formatted correctly during the knowledge extraction process. Knowledge fusion aligns and merges knowledge from different sources, removing redundant and ambiguous information to form globally unified knowledge identifiers and associations, providing a more comprehensive, high-quality knowledge representation. This specifically involves data model layer fusion and data layer fusion. Data model layer fusion includes concept merging, concept-hyperlink merging, and concept attribute definition merging, mapped from structured data. During the mapping process, fusion rules are set to ensure data consistency. Data layer fusion includes entity merging, entity attribute merging, and conflict detection and resolution. Specifically, it refers to the fusion of entity and relationship (including attribute) tuples, primarily for entity matching or alignment. Because some entities in the knowledge base have the same meaning but different identifiers, these entities need to be merged. Finally, manual correction involves data clarification and screening, eliminating erroneous data and correcting inaccurate data. During the graph construction process, the system repeatedly updates the ontology layer and adds new knowledge to the ontology, ultimately completing the construction of the knowledge graph for power material equipment inspection and management.

[0054] In this embodiment, the knowledge graph is constructed based on the Neo4j graph database, and the material equipment inspection knowledge graph is constructed and managed with graph nodes and relationship edges. The graph model consists of nodes and relationships. Nodes store entity information, and nodes are linked by their relationships. The properties and labels of nodes and relationships are stored in the form of key-value pairs. This example uses the inspection information related to distribution transformers as the data source, and provides an example of graph visualization of distribution transformers. Figure 8As shown, the nodes in the diagram are composed of distribution transformer components, inspection items, inspection indicators, problem causes, and problem types. Different types of nodes are distinguished by different colors when visualized. The Neo4j graph database can clearly display distribution transformer inspection-related entities and their relationships through query language, which is helpful for knowledge query and problem diagnosis.

[0055] The present invention mines the correlation between unqualified inspection items by analyzing massive amounts of material inspection data; constructs a material and equipment inspection management knowledge graph based on knowledge graph technology; and comprehensively improves the carrying capacity of power grid material inspection by performing association rule analysis and equipment inspection management knowledge graph construction on each unqualified inspection item, effectively solving the problems of insufficient capabilities in work efficiency, in-depth analysis, etc.; the system can also significantly improve the advancement and rationality of material inspection work, and give full play to the role of power grid material quality inspection in safeguarding power construction and operation and maintenance.

[0056] In summary, the present invention uses an association rule algorithm to establish an association analysis model for equipment inspection, mines the hidden associations between unqualified inspection items, and provides a basis for judging the unqualifiedness of related items. In addition, the present invention also combines information such as material equipment inspection reports and expert knowledge, uses natural language processing technology from semi-structured data, extracts knowledge such as entities, attributes and relationships, and constructs a material equipment inspection management knowledge graph in the graph database Neo4j, provides a systematic material inspection knowledge base, and performs network visualization to facilitate information query and management. By mining the associations between unqualified distribution transformer inspection items, the system sorts out key inspection items and provides a basis for judging the unqualifiedness of related items. To a certain extent, it can clarify the internal connections between various inspection items, improve the pertinence and effectiveness of material inspection work, and improve work efficiency. The constructed material equipment inspection management knowledge graph can provide a systematic material inspection knowledge base, which is convenient for knowledge content query and management.

Claims

1. A power material detection system based on improved association rules and knowledge graph, characterized in that: include: Material and equipment detection module, used to obtain and process power material and equipment detection data; The association rule mining module is used to construct a frequent pattern tree for unqualified test items based on the processing results using the frequent pattern growth algorithm, and to mine strong association rules based on the frequent pattern tree. In particular, a time decay factor is introduced into the association rule mining; The knowledge retrieval module is used to define the ontology layer and extract knowledge in the data layer. At the same time, it repeatedly updates the ontology layer, adds the newly acquired knowledge representation to the ontology layer, and obtains the power material equipment inspection and management knowledge graph based on the strong association rules obtained by mining. Among them, the power material equipment inspection and management knowledge graph is used to display the power material inspection results.

2. The electric power material detection system based on improved association rules and knowledge graph according to claim 1 is characterized in that: The processing of the power material equipment detection data is specifically as follows: Based on the acquired power material equipment inspection data, the inspection items that do not meet the requirements are sorted into transaction data sets D , complete the processing of power material equipment inspection data, where the transaction data set includes the material equipment name, equipment serial number and unqualified inspection items.

3. The power material detection system based on improved association rules and knowledge graph according to claim 2 is characterized in that: The mining of strong association rules is specifically as follows: Use the frequent pattern growth algorithm to scan the processed transaction data set to obtain frequent item sets and create the header table H; Use frequent item sets to filter infrequent items in transaction data sets and generate frequent pattern trees; Based on the header table H, recursive operations are performed on the frequent pattern tree to mine strong association rules, wherein a time decay factor is introduced into the association rule mining.

4. The electric power material detection system based on improved association rules and knowledge graph according to claim 3 is characterized in that: The recursive operation is performed on the frequent pattern tree, which is specifically: On the frequent pattern tree, start from the last item d in the header table H and process each item in turn, and add the current processed item to the base item set base-itemset={d} with an initial value of empty. Each base item set generated is a new frequent item set, and all nodes of item d on the frequent pattern tree can be obtained from the header table H; Scan the path from each node d to the root node on the frequent pattern tree, count all the items that appear and the number of supports for each item, and save the statistical results to the sub-header table subH with the same structure as the header table H. d In which, the support number of all items on the same path is the support number of node d on the path, and all items in the header table H are sorted from small to large according to the support number. In the sorting process, the time decay factor is introduced; Scan all the paths from node d to the root node again, and put all the items on the path into the subheader table subH d Sort in the order of subheader table subH d , and add the ordered item sets and support numbers to the subtree subH d superior; If the sub-header table subH d If it is not empty, construct a conditional subtree subT for the base item set d ; Recursively process the subheader table subH d and conditional subtree subT d ; Based on the recursive processing results, the currently processed item is deleted from the current base-itemset, strong association rules are mined, and the next item in the currently processed header table H is processed.

5. The electric power material detection system based on improved association rules and knowledge graph according to claim 4 is characterized in that: The expression of the support of the association rule is as follows: in, Represents the support of association rules, A and B are item sets A is a true subset of , and A∩B=Ø, A is called the premise, B is called the sequel of the association rule, Indicates the first k elements, I Represents a transaction dataset D The set of all distinct items in , Indicates that the transaction set also contains items A and B The probability of Indicates that the transaction set also contains items A and B The number of samples.

6. The electric power material detection system based on improved association rules and knowledge graph according to claim 5 is characterized in that: The confidence expression of the association rule is as follows: in, represents the confidence of the association rule, express A Appears in the transaction B The conditional probability of Indicates that the transaction set also contains items A and B The number of samples, Indicates that the transaction set contains items A The number of samples.

7. The electric power material detection system based on improved association rules and knowledge graph according to claim 1 is characterized in that: The construction of the knowledge graph for power material equipment inspection and management is specifically as follows: A top-down and bottom-up hybrid construction approach is used to define the ontology layer and extract knowledge in the data layer to obtain knowledge representation; The knowledge representation is corrected, and the ontology layer is repeatedly updated in the process of constructing the knowledge graph of power material equipment inspection and management. The newly acquired knowledge representation is added to the ontology layer to obtain the knowledge graph of power material equipment inspection and management.

8. The electric power material detection system based on improved association rules and knowledge graph according to claim 7 is characterized in that: The correction comprises: Data processing: modify and / or delete data that is not processed in the correct format during the knowledge extraction process; Knowledge fusion: Processing knowledge from different sources to form globally unified knowledge identification and association; Manual correction: Eliminate erroneous data and correct data that does not meet the description requirements.

Images