Method and system for constructing multi-source information carrier of CLCC converter based on knowledge graph

Abstract

The present invention discloses a method and system for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph, obtains relevant data of a CLCC converter, constructs and stores triple information in the form of entity-attribute-attribute value based on the data; constructs a knowledge hierarchy of the converter based on the converter device entity; constructs a knowledge graph of the converter multi-source information based on knowledge fusion, establishes knowledge relationship connection and knowledge fusion; realizes the dynamic flow of transparent information of the CLCC converter based on the knowledge hierarchy of the CLCC converter multi-source information; uses a graph database to perform network topology modeling on the knowledge graph of the CLCC converter, and constructs a transparent information carrier of the CLCC converter. The present invention solves the problems of low efficiency in multi-source operation and maintenance information processing of CLCC converters, serious fragmentation problems, inconsistency in data structure, technical architecture, construction standards and construction content, and difficulty in forming an effective data organization framework.

Description

Technical Field

[0001] The present invention relates to the field of information organization technology, and in particular to a method and system for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph. Background Art

[0002] As a new type of converter, the controllable commutation converter, or CLCC converter, is not only a huge technological breakthrough and innovative application, but also a practical guide for the future development of conventional DC transmission. The CLCC converter combines the advantages of traditional commutation converter technology and flexible DC transmission technology, such as large capacity, low loss, low cost, and no commutation failure. However, it also faces the problem of insufficient operating experience brought by the first set. There is an urgent need for digital equipment, intelligent management, and lean operation and maintenance of CLCC converters on site.

[0003] CLCC converter equipment is numerous, technologically advanced, and of high value. However, due to the lack of systematic top-level design and overall guidance, the construction standards, construction content, and technical architecture of the operation and maintenance management system are not unified. At present, the operation and maintenance work of each converter station mainly relies on digital systems such as control and protection, online monitoring, and operation and maintenance assistance. There are problems such as low information processing efficiency, serious fragmentation problems, inconsistent data structure, and difficulty in forming an effective data organization framework. There are also large differences in technical data architecture and functions. In terms of information organization, the information and data organization and interactivity between different systems such as electric field, temperature field, and airflow field are poor. The operation and maintenance status information such as electricity, heat, and sound are independent and interactively distributed. The degree of interrelationship is high, but there is a lack of a unified mapping paradigm for the operation and maintenance status. There is no effective operation and maintenance information support, which affects the state control capability of the CLCC converter; in terms of information management, there is a lack of information docking methods for the internal operation and maintenance states such as loop state association and control behavior logic, and it is impossible to achieve the management and coordination of the decentralized operation and maintenance information of the CLCC converter. Therefore, a technical solution that can realize the transparency of CLCC converter information is needed to improve the efficiency of information organization and processing.

[0004] The existing technologies for constructing information carriers for knowledge graphs include:

[0005] Chinese patent application publication number CN117093718A provides a method for integrating massive unstructured knowledge graphs based on cloud computing power and big data technology, the method comprising: screening out texts containing lyrics or comments, and using a sentiment analysis module to perform sentiment scoring; identifying the acquired data on a cloud server, and screening out lyrics or music elements related to emotions; forming sentiment label-element pairs, and merging and unifying similar or related labels to reduce the complexity of the knowledge graph; using a data integration engine to integrate the unified label-element pairs into a simplified data framework; after the data integration is completed, triggering the update process of the knowledge graph, and adding this simplified data framework to the music knowledge graph; according to the music knowledge graph, analyzing the melody or lyrics that the music producer or lyricist is concerned about through a decision tree algorithm; based on the prediction results of the decision tree algorithm, generating a lyrics modification plan in combination with the music knowledge graph for modifying the melody or lyrics. However, the application relies too much on the quality and diversity of training data, and the decision tree algorithm lacks flexibility and accuracy when dealing with highly nonlinear relationships and complex decision-making scenarios, especially when it comes to the coupling of multiple physical quantities or non-logical operations in power scenarios, and cannot fully capture the content and characteristics of data information.

[0006] Chinese patent application publication number CN117094387A provides a method and system for constructing a knowledge graph based on big data, the method comprising: determining the current sample project conversation data according to the sample learning value, and loading the corresponding target conversation association vector into the initialization neural network for knowledge learning, generating a first knowledge relationship estimation network, and optimizing the learning value of each sample project conversation data using the first knowledge relationship estimation network, forming a self-learning and improvement cycle process, and improving the network estimation accuracy. When the first knowledge relationship estimation network converges, a second knowledge relationship estimation network is generated, which can estimate the knowledge relationship attribute characteristics corresponding to the loaded conversation text segment and the loaded project focus node information, and construct a knowledge graph. However, this application relies on the training of neural networks, especially iterative self-learning and improvement cycles, so it requires a lot of computing resources and computing time, which limits its scope of application. At the same time, determining the learning value of the sample project conversation data requires the intervention of subjective judgment, which is easy to cause information inconsistency or error, and is not suitable for the power profession with high trust and transparency.

[0007] Chinese patent application publication number CN117875424A provides a knowledge graph completion method and system based on entity description and symmetric relationship, the method comprising: S1: obtaining a knowledge graph, preprocessing the knowledge graph, obtaining a triple set and an entity description set; S2: performing entity and relationship modeling on the triple set through a knowledge graph embedding model, obtaining a symmetric relationship set; S3: encoding the triple set, the symmetric relationship set and the entity description set through an encoder, obtaining a coding vector set; S4: decoding the coding vector set through a decoder, obtaining a candidate solution set; S5: scoring and sorting the candidate solution set, obtaining the optimal solution, and using the optimal solution as the completed knowledge graph. However, this method involves multiple complex steps, including knowledge graph preprocessing, embedding model training, encoding and decoding processes, etc. When applied to large-scale knowledge graphs, it will result in longer processing time and higher hardware configuration requirements, thus limiting its practicality in resource-constrained environments. In addition, this method significantly relies on the accuracy and richness of entity descriptions. When the entity description information is missing, inaccurate or too brief, it will affect the encoding effect, thereby reducing the accuracy and completeness of the completed knowledge graph. Summary of the invention

[0008] In order to solve the deficiencies existing in the prior art, namely, the prior art has the problems of poor information organization, poor flexibility, poor accuracy, poor practicality, serious fragmentation and the like, the present invention provides a method and system for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph. The present invention combines informatization-based computational science with the operation and maintenance requirements and equipment characteristics of the CLCC converter, organizes, aggregates and maps electrical and non-electrical information such as physical information, time-space status, equipment status, technical parameters, two-dimensional drawings, control behavior logic, etc., and constructs a transparent information carrier to realize multi-scale and transparent display of the operation and maintenance information of the CLCC converter.

[0009] The present invention adopts the following technical solution.

[0010] A first aspect of the present invention provides a method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph, the method comprising the following steps:

[0011] Step 1: Acquire CLCC converter related data, use the converter related data to construct triples, and construct knowledge topics, knowledge facets, and knowledge nodes of the CLCC converter based on the triples; the triples are expressed in the form of entity-attribute-attribute value;

[0012] Step 2: constructing a knowledge hierarchy of the converter based on the CLCC converter device entity; and vectorizing and encoding each knowledge facet and knowledge node;

[0013] Step 3: Based on the encoding results of step 2, construct the CLCC converter multi-source information knowledge graph and the CLCC converter transparent information carrier;

[0014] Step 4: Realize the dynamic flow of CLCC converter transparency information based on the knowledge hierarchy of CLCC converter multi-source information;

[0015] Step 5: Based on the graph database, the network topology modeling of the CLCC converter multi-source information knowledge graph is carried out, and the transparent information carrier constructed in step 3 is used to realize the transparent display of the CLCC converter multi-source information.

[0016] More preferably,

[0017] In step 1, the relevant data of the CLCC converter includes sensor data of the CLCC converter valve hall, event messages, equipment drawings, nameplates, and electrical and non-electrical data representing the operating status of the converter.

[0018] More preferably,

[0019] In step 1, the triples are constructed by using the data, which is to export the data into a common information model file, and then construct the triples based on the common information model file.

[0020] More preferably,

[0021] In step 2, the hierarchical structure of the converter knowledge includes upper and lower levels and affiliations of the knowledge topics;

[0022] The upper and lower levels of the knowledge topics and the affiliation relationships are described by defining a tree structure.

[0023] More preferably,

[0024] In the tree structure, a knowledge subject is defined as a parent node, i.e., an upper layer in the hierarchy; multiple knowledge facets corresponding to the knowledge subject are defined as child nodes, i.e., lower layers in the hierarchy; and the affiliation between the knowledge subject and each knowledge facet is displayed through the tree structure.

[0025] More preferably,

[0026] In step 2, attribute constraints are established based on the knowledge hierarchy relationship; the attribute constraints include attribute type, allowed value range and number of data.

[0027] More preferably,

[0028] In step 2, the weight of each knowledge topic is as follows:

[0029] ;

[0030] in, For the qi The knowledge weight value of each knowledge topic, wk For the qi The total number of knowledge facets associated with each knowledge topic. Indicates qi The total number of knowledge nodes corresponding to the knowledge topics and the knowledge facets associated with them, Representing knowledge topics X The weight threshold of .

[0031] More preferably,

[0032] In step 2, the vectorized encoding of each knowledge facet and knowledge node is implemented based on a multi-layer perceptron.

[0033] More preferably,

[0034] In step 2, the knowledge facets corresponding to different knowledge topics are connected and aligned in pairs.

[0035] More preferably,

[0036] In step 3, based on the vectorized coding of each knowledge facet and knowledge node obtained in step 2, as well as the set of association relationships between each knowledge topic and its corresponding knowledge facet, a comprehensive description of the relationship between device entities is implemented in the knowledge graph, thereby constructing a multi-source information knowledge graph of CLCC converters;

[0037] The comprehensive description of the relationship between the equipment entities is shown in the following formula:

[0038] ;

[0039] in, It is a comprehensive description matrix of the relationship between equipment entities; The set of vectorized codes for knowledge nodes; Represents the relationship between knowledge nodes after vector encoding. When the first Knowledge nodes and When the knowledge nodes are related, ,when Knowledge nodes and If there is no association between knowledge nodes .

[0040] More preferably,

[0041] In step 3, the transparent information carrier of the CLCC converter is constructed based on each knowledge source. Z As shown below:

[0042] ;

[0043] in, It is a transparent information carrier for CLCC converters. B A set of knowledge source quantization values ​​for the knowledge nodes constructed in the knowledge graph at the current moment; For the ti The quantitative value of a knowledge source, represents the variables related to all knowledge sources in the knowledge source quantization value set, For the ti The conditional probability distribution parameter of the knowledge source of a CLCC converter, K is the total number of knowledge nodes constructed in the knowledge graph at the current moment, and E is the entity set of CLCC converters corresponding to the aforementioned knowledge nodes.

[0044] More preferably,

[0045] In step 4, the method of using the main entity structuring method to realize the dynamic flow of CLCC converter transparency information is: constructing a knowledge propagation mechanism, and constructing a dynamic knowledge model based on the aforementioned propagation mechanism, so as to realize the dynamic flow of CLCC converter transparency information.

[0046] More preferably,

[0047] In step 4, after satisfying the dynamic knowledge flow of CLCC converters, it is necessary to establish a mathematical model of dynamic knowledge topics to meet the actual operation and maintenance needs;

[0048] Mathematical model of dynamic knowledge topics within a certain period of time DM for:

[0049] ;

[0050] In the formula, The first ai Behavioral state, The first aj Behavioral state; an is the total number of behavioral states of the knowledge subject; ET-ST for a set time period; Indicates the th knowledge topic corresponding to the The total number of times a behavior state occurs within a set time period.

[0051] More preferably,

[0052] In step 5, the method for performing network topology modeling on the CLCC converter multi-source information knowledge graph based on the graph database is:

[0053] Based on the CLCC converter multi-source information knowledge graph and the mapping path of triple information, the entities and inter-entity relationships in the knowledge nodes are identified and converted into nodes and inter-node relationships in the graph database;

[0054] Then, the nodes and the relationships between nodes in the graph database are implanted into the graph database network to complete the network topology modeling.

[0055] More preferably,

[0056] The mapping path of the triplet information includes an entity set of CLCC converter operation and maintenance objects, a relationship set between entities, an attribute set, a set of association relationships between entities and attributes, and a relationship set between triplet.

[0057] The second aspect of the present invention provides a CLCC converter multi-source information carrier construction system based on a knowledge graph based on a CLCC converter multi-source information carrier construction method, including a CLCC converter knowledge information construction module, a CLCC converter hierarchy construction module, a CLCC converter multi-source information knowledge graph construction module, a CLCC converter knowledge dynamic flow construction module and a CLCC converter network topology modeling module.

[0058] The CLCC converter knowledge information construction module is used to obtain CLCC converter related data, and use the converter related data to construct triples, and construct knowledge topics, knowledge facets, and knowledge nodes of the CLCC converter based on the triples;

[0059] The CLCC converter hierarchy construction module constructs the knowledge hierarchy of the converter based on the CLCC converter device entity; and performs vectorized coding on each knowledge facet and knowledge node;

[0060] The CLCC converter multi-source information knowledge graph construction module constructs the CLCC converter multi-source information knowledge graph based on the encoding result of step 2;

[0061] The CLCC converter knowledge dynamic flow construction module realizes the dynamic flow of CLCC converter transparent information based on the knowledge hierarchy of CLCC converter multi-source information;

[0062] The CLCC converter network topology modeling module performs network topology modeling on the CLCC converter multi-source information knowledge graph based on a graph database.

[0063] Another aspect of the present application provides an electronic device, including a processor and a storage medium:

[0064] The storage medium is used to store instructions;

[0065] The processor is used to operate according to the instructions to execute the method for constructing a multi-source information carrier of a CLCC converter.

[0066] The present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a multi-source information carrier of a CLCC converter.

[0067] Compared with the prior art, the present invention has the following advantages:

[0068] 1. The present invention solves the serious problems of information fragmentation, disordered information structure, and low resource integration in traditional complex operation and maintenance network information by constructing a transparent information carrier for CLCC converters based on knowledge graphs. Through orderly information carriers, it provides technical support for business needs such as problem node tracing and status analysis of CLCC converters.

[0069] 2. The present invention can more intuitively display the multi-source operation and maintenance knowledge relationship of CLCC converters. By constructing a transparent information carrier of CLCC converters based on knowledge graphs, operation and maintenance personnel can be assisted to fully and intuitively understand the multi-source information status of CLCC converters.

[0070] 3. The transparent information carrier of the CLCC converter established by the present invention is particularly helpful for analyzing the related states, so that prevention and maintenance can be carried out in advance, which helps to extend the life of the CLCC converter equipment.

[0071] 4. The present invention solves the problems of low efficiency in multi-source operation and maintenance information processing of CLCC converters, serious fragmentation problems, inconsistency in data structure, technical architecture, construction standards and construction content, and difficulty in forming an effective data organization framework. It effectively characterizes the complex operation and maintenance information of CLCC converters and establishes an orderly and structured organization. The knowledge fusion results can describe the complex operation and maintenance conditions of CLCC converters from the perspective of multi-source information, improve the readability, reliability, organization efficiency and analysis efficiency of CLCC converter information, and realize efficient use of CLCC converter information.

[0072] 5. The present invention combines the information-based computing science with the operation and maintenance requirements and equipment characteristics of the CLCC converter, organizes, aggregates, and maps the electrical and non-electrical information such as physical information, time-space status, equipment status, technical parameters, two-dimensional drawings, and control behavior logic, and constructs a transparent information carrier to realize the transparency of multi-scale operation and maintenance information. Most of the previous inventions are based on information acquisition, data quality, simulation dimensions, and state evaluation, such as the construction of ubiquitous Internet of Things, information extraction of multi-agent systems, and virtualized entity characteristics. The information complexity, organization, and coupling node effects of power equipment are not considered. Information transmission and processing mainly rely on manual monitoring and analysis, and there are problems such as information confusion and low processing efficiency. In contrast, the transparent information carrier of the CLCC converter can provide information decision support with multi-scale influences, thereby describing the complex operation and maintenance conditions of the CLCC converter from a panoramic perspective, improving the readability, reliability, organization efficiency, and analysis efficiency of the CLCC converter information, so as to achieve efficient use of the CLCC converter information. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a schematic flow chart of a method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to the present invention;

[0074] Figure 2 A schematic diagram of the framework of a method for constructing a multi-source information carrier for a CLCC converter;

[0075] Figure 3 This is a schematic diagram of the CIM / E file analysis process;

[0076] Figure 4 Schematic diagram of multi-source knowledge structured alignment method for CLCC converter;

[0077] Figure 5 A schematic diagram of a system framework for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to the present invention. DETAILED DESCRIPTION

[0078] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0079] This application discloses a method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph, see Attachment Figure 1 ,include:

[0080] Step 1: Acquire CLCC converter related data, use the converter related data to construct triples, and construct knowledge topics, knowledge facets, and knowledge nodes of the CLCC converter based on the triples; the triples are expressed in the form of entity-attribute-attribute value;

[0081] The CLCC converter related data includes sensor data, event messages, equipment drawings, nameplates of the CLCC converter valve hall, and electrical and non-electrical data characterizing the operating status of the converter.

[0082] The triples are constructed by using the data, namely, exporting the data into a common information model file, and then constructing the triples based on the common information model file.

[0083] Step 2: constructing a knowledge hierarchy of the converter based on the CLCC converter device entity; and vectorizing and encoding each knowledge facet and knowledge node;

[0084] The hierarchical structure of the inverter knowledge includes upper and lower levels and affiliations of the knowledge topics;

[0085] The upper and lower levels of the knowledge topics and the affiliation relationships are described by defining a tree structure.

[0086] In the tree structure, a knowledge subject is defined as a parent node, i.e., an upper layer in the hierarchy; multiple knowledge facets corresponding to the knowledge subject are defined as child nodes, i.e., lower layers in the hierarchy; and the affiliation between the knowledge subject and each knowledge facet is displayed through the tree structure.

[0087] Attribute constraints are established based on the hierarchical structure of knowledge; the attribute constraints include attribute type, allowed value range and number of data; the attribute types include numeric type, enumeration type and Boolean type.

[0088] The weight of each knowledge topic is shown in the following formula:

[0089] ;

[0090] in, For the qi The knowledge weight value of each knowledge topic, wk For the qi The total number of knowledge facets associated with each knowledge topic. Indicates qi The total number of knowledge nodes corresponding to the knowledge topics and the knowledge facets associated with them, Representing knowledge topics X The weight threshold of .

[0091] The vectorized encoding of each knowledge facet and knowledge node is implemented based on a multi-layer perceptron.

[0092] The knowledge facets corresponding to different knowledge topics are connected and aligned in pairs.

[0093] Step 3: Based on the encoding results of step 2, construct the CLCC converter multi-source information knowledge graph and the CLCC converter transparent information carrier;

[0094] In step 3, based on the vectorized encoding of each knowledge facet and knowledge node obtained in step 2, as well as the set of association relationships between each knowledge topic and its corresponding knowledge facet, a comprehensive description of the relationship between device entities is realized in the knowledge graph, thereby constructing a CLCC converter multi-source information knowledge graph; the association relationship between the knowledge topic and its corresponding knowledge facet is recorded as G X =( ),in( ) represents knowledge facets , for knowledge topics The related wj knowledge facets; where wj is an integer, and wj Can represent any knowledge topic The value range of knowledge facets with association relationship is between 1 and The total number of knowledge facets with associated relationships.

[0095] The comprehensive description of the relationship between the equipment entities is shown in the following formula:

[0096] ;

[0097] in, It is a comprehensive description matrix of the relationship between equipment entities; The set of vectorized codes for knowledge nodes; Represents the relationship between knowledge nodes after vector encoding. When the first Knowledge nodes and When the knowledge nodes are related, ,when Knowledge nodes and If there is no association between knowledge nodes .

[0098] The transparent information carrier of CLCC converter is constructed based on various knowledge sources Z As shown below:

[0099] ;

[0100] in, It is a transparent information carrier for CLCC converters. B A set of knowledge source quantization values ​​for the knowledge nodes constructed in the knowledge graph at the current moment; is the quantitative value of the ti-th knowledge source, represents the variables related to all knowledge sources in the knowledge source quantization value set, is the conditional probability distribution parameter of the knowledge source of the ti-th CLCC converter, K is the total number of knowledge nodes constructed in the knowledge graph at the current moment, and E is the entity set of CLCC converters corresponding to the aforementioned knowledge nodes.

[0101] Step 4: Realize the dynamic flow of CLCC converter transparency information based on the knowledge hierarchy of CLCC converter multi-source information;

[0102] The method of using the main entity structuring method to realize the dynamic flow of CLCC converter transparency information is: constructing a knowledge propagation mechanism, and constructing a dynamic knowledge model based on the aforementioned propagation mechanism, thereby realizing the dynamic flow of CLCC converter transparency information.

[0103] After satisfying the dynamic knowledge flow of CLCC converters, it is necessary to establish a mathematical model of dynamic knowledge topics to meet the actual operation and maintenance needs;

[0104] Mathematical model of dynamic knowledge topics within a certain period of time DM for:

[0105] ;

[0106] In the formula, The first ai Behavioral state, The first aj Behavioral state; an is the total number of behavioral states of the knowledge subject; ET-ST for a set time period; Indicates th The knowledge topic corresponds to The total number of times a behavior state occurs within a set time period; th Can represent any knowledge subject;

[0107] Preferably, the set time period is 1 hour.

[0108] Step 5: Based on the graph database, the network topology modeling of the CLCC converter multi-source information knowledge graph is carried out, and the transparent information carrier constructed in step 3 is used to realize the transparent display of the CLCC converter multi-source information.

[0109] The method for network topology modeling of the CLCC converter multi-source information knowledge graph based on the graph database is:

[0110] Based on the CLCC converter multi-source information knowledge graph and the mapping path of triple information, the entities and inter-entity relationships in the knowledge nodes are identified and converted into nodes and inter-node relationships in the graph database;

[0111] The mapping path of the triplet information includes an entity set of CLCC converter operation and maintenance objects, a relationship set between entities, an attribute set, a set of association relationships between entities and attributes, and a relationship set between triplet.

[0112] The nodes and relationships between nodes in the graph database are implanted into the graph database network to complete the network topology modeling. Example

[0113] like Figure 2 As shown, the present invention provides a method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph, by mapping the multi-source operation and maintenance information of the CLCC converter, such as event messages, alarm information, image and audio, into multi-source knowledge, performing entity naming and knowledge pre-storage knowledge processing, and then performing entity relationship extraction and hierarchical classification, thereby completing knowledge modeling; the present invention realizes the construction of a multi-source knowledge graph through triple knowledge mapping, knowledge orderly organization, and knowledge structured alignment; further, through knowledge fusion, a multi-source information carrier of a CLCC converter is constructed in an ordered and structured form, providing guidance for the operation and maintenance and information retrieval of the CLCC converter. Wherein, knowledge fusion includes knowledge vectorization and similarity calculation.

[0114] The method is as Figure 1 The following steps are shown:

[0115] Step 1: Obtain relevant data of the CLCC converter, use the data to construct triple information in the form of entity-attribute-attribute value, and construct knowledge topics, knowledge facets, and knowledge nodes of the CLCC converter based on the triple information;

[0116] Relevant data of the CLCC inverter is obtained, and triplet information is constructed based on the data in the form of entity-attribute-attribute value to map the CLCC inverter entity and the attribute association relationship between its entities, that is, to describe the correspondence between each entity and entity attribute in the CLCC inverter, and pre-store the data in the graphic database according to the logical structure before and after the data in the triplet information and the physical structure of the corresponding device entity.

[0117] It should be noted that the equipment entities include the low-voltage insulated gate bipolar transistor valve, high-voltage insulated gate bipolar transistor valve, main branch, auxiliary branch, main thyristor valve module, auxiliary thyristor valve module, high-voltage and low-current fully-controlled valve, bridge arm, commutation bus, lightning arrester, winding, valve layer wiring, commutation valve tower and valve control and pole control equipment of the CLCC converter.

[0118] Preferably, in step 1, the relevant data of the CLCC converter includes sensor data, event messages, equipment drawings, nameplates, and electrical and non-electrical data of the CLCC converter valve hall.

[0119] In a preferred embodiment of the present invention, the acquired relevant data of the CLCC inverter is first converted into a common information model file, i.e., a CIM / E file, and then a triple information mapping path of entity-attribute-attribute value of the CLCC inverter is constructed based on the basic data types of the CIM / E file using a computer programming language, such as Python. The multi-source information knowledge unit of the CLCC inverter, i.e., the triple information, is encapsulated in the CIM / E file, and the physical entity drawings, CAD drawings, three-dimensional models and manufacturer models corresponding to the CLCC inverter equipment are migrated from the computer memory to a graphic database, such as Neo4j.

[0120] Preferably, in step 1, model training is an important part of constructing a multi-source information carrier of CLCC converters based on knowledge graphs. First, information related to CLCC converters is collected from multiple data sources, including sensor data, event messages, and equipment drawings, and the data is converted into triples, that is, entity-attribute-attribute value forms, to map the relationship between the device and its attributes. Then, a graph structure is constructed using triples, in which nodes represent entities and edges represent the relationship between entities. A feature vector is initialized for each node, and the vector feature is an attribute value or a generated vector. A graph convolutional network is selected as a model for training because it can effectively capture the correlation between nodes in graph structure data. During the training process, the constructed graph structure and feature vector are input into the graph convolutional network model, and a cross entropy loss function is defined to evaluate the accuracy of the model prediction. The model is iteratively trained using a stochastic gradient descent optimization algorithm, and the weights are updated by minimizing the loss function. After the model training is completed, the model performance is evaluated using a validation set, the accuracy, recall, and F1-score indicators are calculated, and the hyperparameters are adjusted according to the evaluation results for retraining. Ultimately, the trained model is used to infer and analyze new data, extract relevant features, and update the knowledge graph in a timely manner to ensure its accuracy and timeliness in a dynamic environment.

[0121] It should be noted that the triple information mapping path can also be constructed using C language. There is no difference in the technical effects brought by the two. Using different programming languages ​​to construct the triple information mapping path will not have any impact on the technical effects of the present invention. The image database can also use other image databases such as OrientDB, ArangoDB or JanusGraph.

[0122] like Figure 3 As shown, taking Python language as an example: the acquired relevant data of the CLCC inverter is converted into a CIM / E file, and then the CIM / E file is opened using Python language to read the relevant data of the CLCC inverter, and then the triplet information is constructed and exported to the CIM / E file. In the process of converting the data into a CIM / E file, the CIM / E file will be automatically generated, but the construction and formatting of the content need to be manually defined and converted. The conversion process involves parsing and organizing the relevant data of the CLCC inverter to ensure that the generated CIM / E file meets specific grammatical structure and format requirements. This process requires manual code writing and definition of data structure based on the data source and model design. This solution uses a block function to identify the class names in the file as data blocks, and the attributes are guided by "@";

[0123] Those skilled in the art should know that data block is a professional term for data processing operations in Python. It is a record arranged continuously in the order of triple mapping. It is a data unit transmitted between the main memory and the input device, output device or external memory. The calculation operations are mainly performed by block functions.

[0124] Use Python language to read the relevant data of CLCC converter in CIM / E file, and then construct the optimal solution of triple information as follows Figure 3 As shown, based on Python parsing of CIM / E files, the basic grammatical structure of the CIM / E files describing the CLCC converter is first analyzed, and then the model data in the CIM / E files are parsed into a data structure in the computer memory; according to the basic data types of Python, a data structure named "Class" is constructed and defined to store the multi-source information knowledge unit of the CLCC converter, namely, triple information.

[0125] See attached Figure 3, parsing the model data in the CIM / E file into a data structure in computer memory is to determine whether the file is complete after starting to read and write the file; if not, then determine whether it starts with the character "< / " and ends with ">\\n"; if so, then parse and obtain the hierarchical name, and then set CIM / E["tag"] to initialize the content of the CIM / E file, completing one round of reading and writing the file; if not, then continue to determine whether it starts with "@"; if it is determined that it starts with "@", then parse and obtain the attributes, and then set CIM / E["attribute"] to initialize the content of the CIM / E file, completing one round of reading and writing the file; if it is determined that it does not start with "@", then continue to determine whether it starts with "#"; parse and obtain the data starting with "#", and then set CIM / E["value"] to initialize the content of the CIM / E file, completing one round of reading and writing the file; after completing one round of reading and writing the file, continue to determine whether the file is complete until the judgment result is yes and then close the file.

[0126] Preferably, the relevant data of the CLCC converter can be divided into structured data, semi-structured data, and unstructured data; for different data types in the relevant data, they are stored in the form of rows and columns to form data entries; where each row represents a record and each column represents a field, and they are stored in different relational databases such as Oracle, MySQL, SQLServer, PostgreSQL, etc. according to actual needs;

[0127] Among them, the structured data has a clear row and column structure and can be conveniently stored in a relational database. For the relevant data of the CLCC converter, the structured data includes sensor data, alarm information, inventory records, power data of the operating status, etc.

[0128] Among them, the sensor data includes measurement data such as temperature, pressure, current, and voltage. These data usually have fixed fields such as timestamps, measurement values, and units, and can be stored in the form of a two-dimensional table.

[0129] The alarm information includes the time, type, severity, processing status, etc. of the alarm, and the information has fixed fields.

[0130] The inventory records include the installation date, service life, maintenance records, etc. of the equipment, and these records are stored in a fixed format in the management system.

[0131] The power data of the operating status includes the monitoring records of parameters such as electric energy, power factor, and reactive power, and is presented as numerical data collected periodically.

[0132] The semi-structured data does not have a strict row and column structure, but contains tags or markers to describe the attributes of the data, which is convenient for data parsing and processing, and includes event messages, CIM / E files, maintenance record documents, configuration information files, etc.

[0133] The event message is usually recorded in log file, JSON or XML format, including the event's timestamp, type, status and other label information.

[0134] As a common information model file, CIM / E file is used to describe various attributes and relationships of converter equipment. Its data format has fixed tags, but the data structure is relatively flexible.

[0135] Maintenance record documents can use XML or JSON format to record maintenance processes and results, with tags and unfixed data fields.

[0136] The configuration information file includes the setting values ​​of system parameters and is recorded in a semi-structured format such as INI, JSON, or XML.

[0137] The unstructured data has no fixed format and is not suitable for direct storage in a traditional relational database, including equipment drawings, nameplate information, non-electrical data of operating status, document materials, etc.

[0138] Equipment drawings include CAD drawings, 3D model files, etc., which exist in the form of graphics or model files, and the data structure is complex and unpredictable.

[0139] Nameplate information includes pictures, complex text descriptions, serial numbers and other information. The format is relatively flexible and there are no fixed fields.

[0140] The non-electrical data of the operating status includes sound recordings, video monitoring data, etc., which are usually stored in the form of multimedia files.

[0141] Documents include equipment manuals, maintenance logs in PDF format, operating manuals, etc. They have no uniform structure and contain a mixture of text, tables, and images.

[0142] Those skilled in the art should know that the knowledge source refers to the source of data, such as operating data collected by sensors, maintenance records and operating manuals. These data provide basic information for subsequent knowledge extraction and topic analysis. The knowledge theme is the core concept extracted from the knowledge source. In the present invention, the knowledge theme mainly focuses on the operation and maintenance of the CLCC converter. This theme is extracted from various knowledge sources and forms the focus of research; for example, the temperature data recorded by the temperature sensor is a knowledge source, and the operating state of the CLCC converter is the focus of research extracted from the knowledge source and other knowledge sources, that is, the operating state of the CLCC converter is one of the knowledge topics associated with the knowledge source. Knowledge facets are a detailed description of the knowledge topic, indicating different dimensions or aspects of the knowledge topic. For example, for the knowledge topic of the CLCC converter, knowledge facets can be defined as "operating status", "fault handling" and "maintenance method". Knowledge nodes are the basic units that constitute the knowledge graph, representing knowledge facets, knowledge topics or related entities. Each knowledge node can contain multiple attributes and their values, which are used to specifically describe the characteristics of the node. A triple is the basic information structure in the knowledge graph, usually represented as entity-attribute-attribute value. For example, a triple can be represented as: CLCC converter-operating status-normal, which is used to describe the state and attributes of an entity. and Ci The relationship between them is one-to-one. Ci Indicates i The relevant knowledge of a CLCC converter device entity is composed of relevant triplet information. Can be mapped to a set of multiple triples of information, these triples are related to i Associated with a device entity.

[0143] The relevant knowledge of the first CLCC converter device entity is recorded as C 1. k The relevant knowledge of a CLCC converter device entity is recorded as C k , define the CLCC converter equipment entity knowledge set C ={ C 1,…, , …, C k},in C is the knowledge set of CLCC converter equipment entity, is any element in the knowledge set, indicating the i Knowledge of the CLCC converter equipment entity, C i ∈ℝ n´di , ℝ n´diFor relational database i A structured data entry related to a CLCC converter device. i Knowledge of CLCC converter equipment entities Including all i The triplet related to the CLCC converter device, and the i The relationship between the CLCC converter device and other entities. i The triplet related to the CLCC converter device, and the i The associations between each CLCC converter device and other entities are included in the i CLCC converter equipment related structured data items ℝ n´di ; among them i The structured data entries related to each CLCC converter device are structured representations of the device entity in a relational database, including attribute constraints, classification information or hierarchical structure, etc.

[0144] Based on the knowledge set C={C1,…,Ck} of CLCC converter, relevant knowledge topics are defined, and the triplet information constructed above is divided according to different knowledge topics. C 1 For the first CLCC converter equipment entity related knowledge, C k For the k Knowledge of CLCC converter equipment entities;

[0145] The knowledge subject is defined as follows:

[0146] ;

[0147] in, For the first CLCC converter equipment qi Entities have knowledge topics associated with triples. It is composed of related triple information and represented as graph structure data. is the mapping path of triple information; GCN is a graph convolutional network, that is, using this network to achieve Path mapping; the device entity-attribute-attribute value triple information mapping path of the CLCC converter is expressed as G=( E , R , A , T R , T A ) ,in E It is the entity collection of CLCC converter operation and maintenance objects. Ris the set of relations between entities, A is the attribute set, T A = E A is the attribute of the entity, T R = E R E is the relationship between the triples, T R Represents the head entity of the CLCC converter h With tail entity t There is a specific relationship between r ; is the normalized adjacency matrix of the graph, is the degree matrix, W is the weight matrix, σ is the activation function;

[0148] It should be noted that two constraints should be considered when performing relational extraction and triple mapping of CLCC converter knowledge, namely consistency constraint and exclusivity constraint; the consistency constraint represents that the knowledge features of the same CLCC converter operation and maintenance entity have similarity, reflecting the actual working conditions of CLCC converter operation and maintenance; the extraction of knowledge features is carried out from the acquired triple information, and the specific operation includes the following steps: first, obtain the relevant triple information of all CLCC converters, which are expressed in the form of entity-attribute-attribute value, such as converter A-temperature-75°C, converter B-operating status-normal, etc. Then, according to the features to be extracted, filter out the triples of specific attributes, for example, select "temperature" and "operating status" as the attributes of interest. Then, directly extract the attribute values ​​from the filtered triples, such as "75°C" and "normal", as the characteristic values ​​of the equipment. Furthermore, according to the analysis requirements, the extracted attribute values ​​can be normalized, classified or digitized to facilitate subsequent analysis and model input, such as normalizing the temperature value to the range of 0 to 1, or converting the operating status to "0" normal or "1" abnormal. Finally, the processed eigenvalues ​​are summarized into a feature matrix, where each row represents a device and each column represents the characteristics of a specific attribute for further analysis and processing. The exclusive constraint means that a knowledge source can only map the operation and maintenance information of one entity. Since the same device may have different status attributes, knowledge confusion caused by the same knowledge is avoided.

[0149] Knowledge topic collection based on CLCC converter , Represents the total number of knowledge questions, qi is an integer, and qi ∈[1, ]; Use the knowledge subject set X and the entity-attribute-attribute value triple information mapping path G of the CLCC converter, and use the association module ,in, m is the maximum length or maximum number of hops of the mapping path, when expressed as j ≥1 and j ≤ m hour, j is the number of edges in the path, that is, the number of steps, which means starting from the starting point, passing through at least one edge, that is j =1 to max m Edge, that is j = m All possible paths that can be reached will be considered, and the knowledge facets of CLCC converters will be established based on the association criteria, as shown in the following formula:

[0150] ;

[0151] in, The first wi knowledge facets, in the established knowledge subject set, It is an association module, which is used to establish the attribute attribution of a certain element or a certain condition in the knowledge subject X. The subscript j of represents the value range, which is a threshold value set based on the actual working conditions. Here, the value range is set considering the deviation value of the fault point in the actual working conditions; among them, the state association refers to the association relationship between the knowledge subject of the CLCC converter and the specific knowledge facet, which reflects the characteristics and performance of the knowledge subject under different states or conditions. Specifically, the knowledge subject of the CLCC converter involves the state information, operating parameters or other factors that affect its performance of the equipment. The above-mentioned knowledge subject will change its related attributes or characteristics under different operating conditions such as normal operation, failure, maintenance, load change, etc. The role of state association is to establish a connection between the state information of the knowledge subject and the corresponding knowledge facet through the association module to reflect the attribute attribution under specific conditions or states. For example, in the case of large load fluctuations, the knowledge subject of the CLCC converter may include changes in parameters such as current, temperature, and voltage, and the knowledge facets corresponding to these changes are "current state facet" or "temperature state facet". Through state association, the changes in current parameters when the load fluctuates can be associated with the "current state facet", so as to establish the organization and representation of the above states at the level of knowledge nodes. The attribution of the attributes of a certain element or condition is related to the triple content mentioned above. It refers to the attribution of the attributes of a specific entity or the relationship between entities in these triple information. The attributes of a specific entity are T AThe information in; the relationship between entities is T R This means that to determine the position or classification of a knowledge topic or its corresponding knowledge facet in the knowledge facet set of CLCC converter, it is necessary to classify it according to its association with a certain entity, attribute, or relationship. A knowledge facet may involve a specific attribute or a specific relationship between entities, all of which can be represented by triple information.

[0152] Knowledge topic X and its knowledge facet set matrix for CLCC converters , wi is an integer, and wi∈[1,wk]; wk Represents the total number of knowledge facets;

[0153] The knowledge node of the CLCC converter is established as shown in the following formula:

[0154] ;

[0155] in, is the knowledge node of CLCC converter, The knowledge processing module for the knowledge subject and its knowledge facet set represents the information integration of the knowledge subject set and the knowledge facet set, including data cleaning and standardization. Next, the features related to the equipment performance are selected by statistical methods, and numerical features, time series features and graph structure features are extracted. Subsequently, the feature association matrix is ​​constructed, and the association rule mining is performed using the association rule learning method to identify the strong association between the knowledge subject and the knowledge facet. At the same time, the relationship between the equipment status and the features is analyzed, and the key influencing features are identified through regression analysis or cluster analysis.

[0156] A collection of different knowledge topics, knowledge facets and knowledge nodes of CLCC converters is established as shown in the following formula:

[0157] ;

[0158] in, A collection of knowledge topics, knowledge facets, and knowledge nodes. It is a collection function of knowledge topics, knowledge facets and knowledge nodes.

[0159] Preferably, the subject of the knowledge X The sub-contents are divided into different weights under the same dimension, which is convenient for the subsequent calling of conditional random fields to model the knowledge sequence of the multi-source information of the CLCC converter. The multi-source information of the CLCC converter is extracted by using E language and the CIM / E file of the CLCC converter, and the time series of the knowledge source of the CLCC converter is established. , state sequence Corresponding knowledge source probability model ;in is the time series of the first knowledge source, is the time series of the first knowledge source; these time series describe the changes of the data of each knowledge source at different time points; each state in the state sequence y i Represents the operating status of the CLCC converter at the corresponding time point. Specifically, these states can include multiple types, such as: normal operating state, indicating that the converter is working within the expected range and there are no abnormal conditions, and the normal state is quantified as 0; minor fault state, indicating that there are some minor problems with the system, such as slight deviations in sensor readings, but it does not affect the main functions; serious fault state, indicating that there are major problems with the system, such as overheating of components or failure of key sensors, which may require immediate shutdown for maintenance; maintenance state, indicating that the converter is in the maintenance or overhaul stage and does not perform normal operating functions temporarily; and standby state, that is, the system is not running but is ready to start at any time. By analyzing the changes in characteristics such as temperature, operating current, voltage, etc. in the data, these states can be judged and marked, thereby describing the specific conditions of the converter at different time points.

[0160] Those skilled in the art should know that the state sequence can be generated based on a conditional random field or a Markov model or other existing technologies, and those skilled in the art can construct it according to actual conditions.

[0161] According to the probability model P( y ︱ x ), the weights of different knowledge topics can be obtained. The knowledge source probability model is shown in the following formula:

[0162] ;

[0163] in, ei is an integer, and ei ∈[1,n]; ej is an integer, and ej ∈[1,n], n Indicates the number of sub-contents under the knowledge topic of CLCC converter. and is a hyperparameter of the conditional random field, [ P ]and[ V ] respectively represent and Real-time sequence status.

[0164] ;

[0165] ;

[0166] Step 2: Create the hierarchical structure of CLCC converter equipment entities and knowledge, including the upper and lower levels and affiliations of knowledge topics, and perform attribute constraints based on the knowledge hierarchy, including attribute type, allowed value range, and number of data.

[0167] In step 2.1.1, the upper and lower levels of the knowledge subject and the subordinate relationship are described by defining a tree structure. In a preferred embodiment of the present invention, the knowledge subject can be set as the upper level, and the specific knowledge facets corresponding to the knowledge subject are set as the lower level, thereby forming a top-down hierarchical structure. In this structure, the parent node represents the knowledge subject, and the child node represents its subordinate knowledge facet. The subordinate relationship between the knowledge subject and its facet can be clearly displayed through the tree structure, reflecting the organization of the various attributes of the device entity and their mutual relationships.

[0168] For example, the knowledge subject is the operating status, and the knowledge facets associated with the knowledge subject include normal, minor fault, and major fault; then the operating status can be set as the upper level, and normal, minor fault, and major fault as the lower level, thus forming a top-down hierarchical structure.

[0169] When performing attribute constraints based on the knowledge hierarchy, the following aspects should be considered: attribute types include numeric types such as temperature and pressure, enumeration types such as fault status type and operating status, and Boolean types such as normal and correct; the range of allowed values ​​is limited to each attribute value, for example, the temperature attribute can be specified between -40°C and 85°C, and the voltage attribute can be limited to between 0V and 800V; the number of data can be set within a certain range, for example, for a certain equipment fault record, a maximum of 100 records are retained to avoid data redundancy. Through the above constraints, it is ensured that during the data storage and analysis process, all input attribute values ​​meet the requirements of the actual working conditions, thereby ensuring the validity and accuracy of the data.

[0170] In step 2.1.2, the division of upper and lower levels and affiliation and the corresponding attribute constraints are intended to achieve knowledge reduction through orderly knowledge organization and characterize knowledge association relationships; the calculation of knowledge topic weights is shown in the following formula:

[0171] ;

[0172] in, For the qi The knowledge weight value of the knowledge topic, wk, is qi The total number of knowledge facets associated with each knowledge topic. Indicates qiThe number of knowledge nodes corresponding to the knowledge topics and the knowledge facets associated with them, including qi knowledge topic, and qi knowledge facets that have associated relationships with knowledge topics, and the total number of entities that have associated relationships with the aforementioned knowledge facets, Representing knowledge topics X To effectively calculate the weight threshold, the setting of the threshold needs to comprehensively consider the specific working conditions and operation and maintenance application requirements of the CLCC equipment. =0.6; historical data analysis shows that in similar projects in the past, the average effective similarity between knowledge facets fluctuated between 0.5 and 0.7, so choosing 0.6 as the threshold can ensure that most important knowledge facets are retained. In addition, using the similarity normalization method, if the similarity between any two knowledge facets is greater than 0.6, they are considered to be meaningfully related, thereby improving the overall weight of the knowledge topic.

[0173] Through the prior calculation of knowledge facet weights, let F t1 and F t2 For any knowledge topic t 1 and t 2 knowledge facet set, then t 1 and t 2 Knowledge Similarity S , as shown in the following formula:

[0174] ,

[0175] in, , Representing knowledge topics t 1 and t 2The sum of the weights of all knowledge facets in the public knowledge facets, , Indicates knowledge topics only t The sum of the knowledge facet weights of 1, , Indicates knowledge topics only t The sum of the knowledge facet weights of 2, represents the weight of a single knowledge facet, For knowledge topics t The total weight value is 1, For knowledge topics t The total weight value is 2;

[0176] By setting the weight threshold a, establish an ordered knowledge organization with upper and lower weighted hierarchies, and provide an organizational basis for the knowledge dependencies and implicit knowledge mining of CLCC converters. Among them, the construction of the upper and lower hierarchical structure of knowledge mainly relies on the similarity calculation between knowledge topics to determine their relationship. By calculating the similarity of different knowledge topics, a similarity threshold can be set to divide the upper and lower knowledge. Specifically, when the similarity between two knowledge topics exceeds the threshold, it can be considered that there is a correlation between them, thereby determining the hierarchical relationship. Upper-level knowledge usually represents a higher degree of abstraction and covers more sub-topics, while lower-level knowledge has more specific attributes or functions to support the implementation of upper-level knowledge. For example, if the knowledge topic t The similarity of 1 is t 2 exceeds the threshold, then t 1 can be regarded as t 2's upper knowledge; on the contrary, if t 2 Yes t 1, and its similarity also exceeds the threshold, then t 2 is considered t 1. In this way, the hierarchical structure between knowledge topics is clearly established, providing a structured basis for subsequent knowledge management and information retrieval.

[0177] Preferably, the knowledge topic parameters of multiple models such as semantic analysis and image recognition of CLCC converters are and its knowledge facets The relationship between G X =( ),in( ) represents knowledge facets For knowledge topics The related wj knowledge nodes. H The characteristic vector of G X Input the multi-layer perceptron MLP separately to realize the facet analysis of each knowledge A and knowledge nodes H The vectorized encoding is shown in the following formula:

[0178] ,

[0179] ,

[0180] in, Represents knowledge facet ri The feature vector of a knowledge node, Indicates ri The associated vectorized encoding of knowledge nodes, Indicates With The associated vector encoding of each knowledge facet. The feature vector of a knowledge node is the value corresponding to the knowledge node. Specifically, the feature vector is used to represent the characteristic attributes of the knowledge node. The attribute values ​​of the node are encoded into a high-dimensional space in the form of vectorization, so that they can be input into the multi-layer perceptron MLP for calculation and training. Therefore, the feature vector can be understood as the numerical expression of the knowledge node. A wi Indicates the first wi Each knowledge facet can be considered as a specific feature or sub-content under the knowledge topic. For example, the knowledge facet of CLCC inverter can include the operating status of the equipment, sensor readings, maintenance records, etc. Each knowledge facet has specific attributes to describe different aspects of the knowledge topic. Therefore, A wi Represents the content of a knowledge facet related to a specific knowledge topic.

[0181] The multi-layer perceptron MLP needs to be trained. First, the initial value of the vectorized encoding needs to be set according to the specific characteristics of the knowledge topics and knowledge facets of the CLCC converter. The goal of this step is to H The corresponding attributes are given meaningful initial codes. The specific implementation method is as follows:

[0182] According to the needs of actual applications, the vector dimension is divided into multiple parts, each of which corresponds to a specific device feature. For example, assuming the vector length is N ,in:

[0183] No. 1 to No. k The dimension is used to represent the operating status of the CLCC converter, and the value is set to binary code. For example, "normal" is [1,0,0], "abnormal" is [0,1,0], and "warning" is [0,0,1].

[0184] No. ( k +1) to m Dimensions are used to represent sensor data, including temperature, voltage, pressure, etc. These values ​​are set as continuous real values; for example, the range of temperature is [0, 100] degrees Celsius.

[0185] No. ( m +1) to p The dimension is used to describe the fault information of the device, such as fault type, fault location, etc. It is mapped to a specific vector encoding through the enumeration type; for example, "Type A fault" is [1,0,0,0], and "Type B fault" is [0,1,0,0].

[0186] No. ( p +1) to N Dimensions are used to encode maintenance records, including the most recent maintenance time interval, maintenance operation type, etc. By mapping time difference and maintenance type, the information is converted into numerical features.

[0187] Building knowledge topics X and its facets A Relationship mapping G X :The knowledge subject X and its knowledge facet A, and is expressed as G X =( A , X This step involves graph structuring of the knowledge organization of CLCC converters, that is, building a directed graph by describing the relationship between knowledge topics and knowledge facets. G X , nodes represent knowledge topics or knowledge facets, and edges represent their association relationships.

[0188] Initialization and assignment of vectorized coding: In the multi-layer perceptron MLP model, the number of nodes in the input layer should be equal to the length of the vector N. The assignment method of the initial vectorized coding is set by the feature definition, and the corresponding vectorized data is entered in the input layer node. The value of each vector dimension is assigned according to the specific attribute information:

[0189] The initial vector value of the operating state is assigned according to the actual operating conditions of the device, for example, "normal" is [1, 0, 0] and the fault state is [0, 1, 0].

[0190] Sensor data such as temperature and pressure are normalized according to the actual values ​​of the device and scaled to the range of [0,1] or [-1,1].

[0191] The fault information is converted into a corresponding encoding vector through a mapping table.

[0192] Maintenance records are quantified based on the actual maintenance time intervals, and the time difference is converted into corresponding numerical inputs.

[0193] Knowledge Node H The characteristic vector of G X As the input of the multi-layer perceptron MLP, the set initial value of the vectorized code is used as the output of the multi-layer perceptron MLP for training to obtain a trained multi-layer perceptron.

[0194] Training and optimization of vectorized coding: During the training process of the MLP model, the model is trained using labeled samples in the data set, and the weights of the vectorized coding are adjusted through the back-propagation algorithm. The training process can be implemented in the following steps:

[0195] The input layer receives the vectorized knowledge node encoding and inputs it into the hidden layer of the MLP.

[0196] The hidden layer transforms features through a nonlinear activation function to generate a higher-level feature representation. The activation function can be ReLU, sigmoid, etc.

[0197] The output layer generates prediction results, which can represent the operating status of the equipment or the predicted fault type.

[0198] Using known labeled data, the error of the output layer is calculated using mean square error or cross entropy loss, and the weights of the model are adjusted through the back-propagation algorithm.

[0199] The training process is repeated until the model error converges to within the set tolerance range.

[0200] Prediction and analysis of new data through trained MLP: After model training is completed, the trained MLP model can be used to predict and analyze new CLCC converter data. Input new vectorized code, generate output results through forward propagation of the model, judge the status of the device or possible fault information based on the output results, and interpret the eigenvalues ​​of the code. Preferably, for any knowledge topic t 1, t 2, and the corresponding knowledge facet is recorded as and , connect and align the knowledge facets corresponding to different knowledge topics in pairs; the knowledge facets corresponding to the knowledge topics are the knowledge facets that have an association relationship with the knowledge topics; connect and align the knowledge facets corresponding to different knowledge topics in pairs, as shown in the following formula:

[0201] ,

[0202] in, Representing knowledge topics t 1, t 2 is the concatenation vector of the knowledge nodes of the corresponding knowledge facets, concat represents the model concatenation algorithm, Representing knowledge facets The knowledge node feature set, Representing knowledge facets The knowledge node feature set;

[0203] When the splicing algorithm connects and aligns the knowledge nodes of two knowledge facets, it can be achieved by cascading the feature vectors of each knowledge facet. t 1 The corresponding knowledge facet feature set is , knowledge topics t 2 The corresponding knowledge facet feature set is , the splicing algorithm is expressed as:

[0204] ;

[0205] in, Representing knowledge facets The set of feature vectors of m Knowledge facets t The number of features is 1. Representing knowledge facets The set of feature vectors of n Knowledge facets t The number of features is 2. Z is the concatenated vector, which represents the corresponding connection of the two knowledge facet features. During the concatenation process, the concat operation connects the two feature vector sets into a longer vector and maintains the original order. This method can merge the feature vectors of different knowledge facets into an overall input for the subsequent logistic regression binary classification detector to determine whether their structures are aligned.

[0206] For details, see Appendix Figure 4 ,In the splicing process, first the attribute structures of different knowledge objects are analyzed, where the attribute structure of knowledge objects is to analyze the association relationship between the attributes and connect them with line segments; secondly, structural feature representation is added to the attribute structure of each knowledge object, that is, based on the constructed knowledge object attribute structure, the attribute local structure module is constructed; then the overall and local structure of each attribute local structure module is separated; finally, the overall and local structures of the two pairs of knowledge objects are aligned.

[0207] The splicing vector Input the logistic regression binary classification detector to determine the topic t 1st Aspects and Topics t 2nd Whether the knowledge facets are structurally aligned, so as to establish the unstructured to structured transformation of knowledge topics and knowledge facets of CLCC converter.

[0208] Step 3: Perform knowledge fusion based on knowledge features, build a knowledge graph of multi-source information of CLCC converters based on knowledge fusion, and build a transparent model of knowledge features and knowledge relationships.

[0209] In step 3.1, based on the structured knowledge topics and knowledge facets in step 2, a CLCC converter operation relationship model containing time series and knowledge information is established. The specific implementation of this step is to vectorize the knowledge nodes into a set after encoding in the knowledge fusion process. H and the set of knowledge-based quantized encodings A Combined knowledge topics X and its associated relationship set G X , thereby building a running relationship model R In this model, the relationship between knowledge facets is determined and the adjacency matrix is ​​used to P The adjacent relationship between knowledge concepts is represented, and the knowledge relationship connection between entities is realized. Due to the dependence of CLCC converters on time in dynamic operation and maintenance, an operation relationship model that includes both time series and knowledge characteristics of CLCC converters centered on knowledge information is established, as shown in the following formula:

[0210] ,

[0211] in, For the operating relationship model, is the set of vectorized encoded knowledge nodes, is the set of knowledge-based quantized encodings, For knowledge topics X and its knowledge facets A A collection of association relationships. A collection of knowledge nodes H It is composed of vectorized encoded node data, representing various operation and maintenance characteristics of the converter equipment, such as status and parameters, and providing basic data support for the model; knowledge facet collection A It represents multiple feature dimensions related to equipment operation and maintenance, such as status information, fault characteristics, etc., which are compared and integrated in one space through vectorized coding, enriching the description of equipment status. G X As a bridge between knowledge topics and knowledge facets, it constructs the structural relationship between topics and their facets and describes the interrelationships between various knowledge facets. H , A and G X , a comprehensive description of the relationship between equipment entities can be achieved in the knowledge graph, and a transparent knowledge model for dynamic operation and maintenance can be constructed, thereby realizing the fusion analysis of multi-source information and the correlation disclosure of spatiotemporal characteristics.

[0212] Constructing knowledge relationship connections between CLCC converter device entities based on the operation relationship model , as shown in the following formula,

[0213] ; ;

[0214] in, Represents the relationship between knowledge facets after vector encoding. Knowledge facets and When the knowledge facets are related, ,when Knowledge facets and When the knowledge facets are not related, , is the adjacency matrix derived from the association relationship of knowledge facets, as shown in the following formula:

[0215] ,

[0216] Among them, the adjacency matrix Represents the adjacent relationship of knowledge concepts, is the knowledge node vector, through the weighted distance measure The concept correlation is shown in the following formula:

[0217]

[0218] in, is the modulus of the first knowledge topic vector, is the modulus of the second knowledge topic vector, is the weighted distance, For the qi The mapping value of the first entity content in the knowledge topic corresponding to the triple of entity-attribute-attribute value, Collection of knowledge topics X The average of the first entity content mapping values ​​in the triple, rk is the number of knowledge nodes; when calculating weighted distance or relevance, you need to select a node as the reference point for calculation, which is the current node. rn is the current node degree value of the knowledge node, P X1,X2 The value is between -1 and 1, representing the positive and negative correlation of knowledge concepts; knowledge topics XThe device entities and related attributes of the CLCC inverter are included in the above measurement method. The structural information and node vectors of the CLCC inverter are represented as the content information of the CLCC inverter through the adjacency matrix and the correlation strength of any two knowledge nodes is comprehensively evaluated; the knowledge relevance is comprehensively calculated through the directly or indirectly connected node information, and the weights on the mapping path are accumulated. Among them, the topic vector is the encoding result mentioned above; the mapping value of the first entity content refers to the specific numerical value corresponding to the entity content in the entity-attribute-attribute value structure in the triple. This mapping value is the value of the entity on a specific attribute, which is obtained by calculation or conversion to reflect its characteristics on the knowledge topic.

[0219] It actually represents a specific component of the entity's representation value or feature vector under a specific attribute extracted from the knowledge representation of the triple. These values ​​are obtained through mapping functions such as encoding or hashing.

[0220] When a symbol has multiple values, these values ​​can be combined into a vector to be substituted into the formula for calculation. During calculation, the multi-valued vector can be subjected to norm calculation such as L2 norm to obtain a single scalar value.

[0221] By using the above method, the time-space sequence labels of knowledge are calculated from the operation relationship model of the CLCC converter, and the adjacency matrix under different knowledge topics is converted into a time graph to realize the linking of entity relationships. When constructing the graph of multi-source information of the CLCC converter, the time-space sequence labels of knowledge are calculated through the operation relationship model, and the adjacency matrix under the knowledge topic is converted into a time graph to realize the linking of entity relationships. This step measures the association strength of two knowledge nodes through the weighted distance in the adjacency matrix, and further establishes a transparent model of knowledge features and knowledge relationships.

[0222] Preferably, knowledge fusion is defined as the input-output problem of multilinear functions, using high-dimensional random variables ω Each dimension of , … Describe the knowledge sources of voltage, power, temperature, current, electromagnetic and electric heat of CLCC converter. The knowledge source set of knowledge node is . Constructing a knowledge source set requires quantifying the data of tk knowledge sources: First, it is necessary to determine the quantification standard and range for each knowledge source, including selecting appropriate units and determining data types. Next, collect the original data of each knowledge source through sensors or other monitoring equipment, record information such as voltage, current, and temperature at different time points, and ensure the integrity and consistency of these data. Among them, choose the appropriate unit, and you can choose the unit commonly used for the corresponding data, such as volts V for voltage, watts W for power, degrees Celsius °C for temperature, amperes A for current, etc.; data types include continuous data, discrete data, etc.

[0223] Before quantification, data preprocessing must be performed. This step includes data cleaning, which is to remove outliers and handle missing values, which can usually be done through methods such as interpolation or mean substitution. In addition, the data needs to be standardized or normalized to ensure that the data from different knowledge sources are in the same range for subsequent analysis.

[0224] In terms of specific quantification methods, for some discrete knowledge sources, continuous data can be divided into several intervals, such as temperature into intervals of 0-10°C, 10-20°C, etc., and a corresponding integer value can be assigned to each interval, or categorical variables can be converted into numerical form using methods such as one-hot encoding. For continuous data, the original values ​​can be used directly for quantification, or the data can be normalized to a unified range to ensure comparability between various knowledge sources.

[0225] Ultimately, each quantized value will be defined as a random variable, such as ω 1 Represents the quantized value of voltage, ω 2 Represents the quantified value of power, ω 3 A quantized value representing temperature, and so on. ω 1 , ω 2 , ω 3 Used for subsequent model analysis and knowledge fusion.

[0226] Based on the prior knowledge of CLCC converter operation and maintenance rules, topology structure, etc., the relevant , use Representation; Multimodal knowledge variables Available probability plots Z ( B , E ) description; the multimodal knowledge variable is the aforementioned high-dimensional random variable.

[0227] The probability graph is the information carrier of transparent knowledge, as shown in the following formula:

[0228] ,

[0229] in, Knowledge source for CLCC converters The probability function of Indicates ti Knowledge source for CLCC converters, are variables related to all knowledge sources, , is the joint probability distribution parameter, is the conditional probability distribution parameter, K is the current position of the knowledge node, E is the entity set of CLCC converter operation and maintenance objects; variables related to all knowledge sources By integrating data from different knowledge sources, we assume that we are monitoring the operating status of a CLCC converter and consider several key factors such as voltage, power, and temperature. These factors interact with each other and may affect the performance of the converter. We can define a composite variable , which integrates voltage, power, and temperature. For example, It can be expressed as a weighted sum of voltage and power, minus the effect of temperature; such an expression can help us evaluate the overall operating condition of the converter. Reasonable initialization is required, which can be done by random selection or by adopting some heuristic strategies, such as using the mean or maximum value. Then, the value of parameter 𝜃 is derived by maximum likelihood estimation MLE, specifically by maximizing the likelihood function of the training data under given parameters to obtain the optimal parameters. In addition, Bayesian inference can also be used to obtain the posterior distribution of parameters, combining the prior distribution with the observed data, and using Bayes' theorem to calculate the posterior distribution to update the parameters; the current position of k represents the position of the current knowledge node in the knowledge graph or model. When constructing the knowledge graph of CLCC converters, each knowledge node represents a specific information or feature, and the position of k can be defined by the number or identification of the knowledge node in the graph structure. In addition, the position of k based on the real-time operation and maintenance rules and status monitoring of CLCC converters can also be updated by real-time data. When the system collects real-time data during operation, k can be dynamically adjusted to reflect the status of the current knowledge node in the entire operation and maintenance process. During the training process, the position of k will change with the training iteration, and each iteration will update the status of the knowledge node to better adapt to new data input and environmental changes.

[0230] Those skilled in the art should know that is the conditional probability, which can be obtained as follows:

[0231] First, collect the operating data related to the CLCC converter. Make sure the recorded data includes information such as timestamp, measurement value, sensor location, etc. The data can include multiple variables such as voltage, power, temperature, current, etc. Then perform data preprocessing:

[0232] Clean the data: remove outliers and missing values. Use data interpolation methods such as linear interpolation to fill missing values ​​or mean substitution to handle outliers.

[0233] Standardization: All data were standardized using z-score standardization or min-max scaling methods so that different variables were at the same level for subsequent analysis.

[0234] Infer conditional probability: use an inference algorithm to calculate the conditional probability, where the inference algorithm preferably uses belief propagation or Gibbs sampling.

[0235] It can also be calculated according to the following formula:

[0236] ;

[0237] in yes , and The joint distribution of yes and The joint distribution of .

[0238] Unified integration of CLCC knowledge It is expressed as follows:

[0239] ;

[0240] in, Knowledge source The probability set of Knowledge source The variable set, , is the fusion representation set of the CLCC converter knowledge source after knowledge fusion, m is the type or number of parameters, such as voltage, power, and temperature, then m=3, and n represents the number of variables used in the knowledge fusion process. Variables are features selected from m knowledge sources, and more than one variable may come from the same knowledge source. For example, if we extract two variables from each knowledge source for fusion analysis, then n is 2m, that is, in the knowledge fusion process, more variables are used to fully describe the state of the CLCC converter. The above-mentioned unified knowledge fusion representation method can be described as: according to the knowledge source variables Describe the multimodal knowledge of CLCC converters with probability graphs and establish the joint probability distribution of knowledge nodes Z ( B , E ),based on Z ( B , E ) and the representation set The fusion representation of multimodal knowledge is calculated, and the value with the largest probability is the unified fusion description result of CLCC knowledge. The description result combines the knowledge features of different knowledge topics through the fusion of probability and knowledge concepts, and forms an information carrier that transforms CLCC converter information into CLCC converter transparent knowledge in the form of a knowledge graph. In the process of unified knowledge fusion, the multimodal knowledge source of CLCC converter is described based on the high-dimensional random variable 𝜔, and the probability graph is used Z ( B , E ) calculates the joint probability distribution of knowledge nodes, thereby realizing a unified fusion representation of CLCC converter knowledge. This step combines different knowledge features through the probability fusion method to form an information carrier that transforms CLCC converter information into transparent knowledge, and finally presents the multi-source heterogeneous data of CLCC converters in the form of a knowledge graph.

[0241] The probability set is obtained by analyzing and counting the historical data of the CLCC converter during actual operation. First, relevant data needs to be collected from the sensors and monitoring systems of the equipment. These data may include voltage, power, temperature, current and other variables. Next, the quality of the data is ensured by preprocessing the collected data, such as removing noise and missing values. Then, statistical methods such as frequency analysis and maximum likelihood estimation are used to calculate the probability distribution of these variables to form a probability set. For example, for historical voltage data, its probability distribution can be established by counting the frequency of occurrence of each voltage value. Finally, these probability distributions are aggregated to form a probability set of knowledge sources to support subsequent knowledge fusion and decision-making.

[0242] The variable set specifically refers to the knowledge source variables used as input in the model, which have a direct impact on the state and performance of the CLCC converter. Although the probability set and the variable set have intersections to some extent, their definitions and functions are different. The probability set focuses on the probability distribution characteristics of each knowledge source and emphasizes its statistical properties, while the variable set focuses more on the variables used to describe and analyze the state of the CLCC converter in actual operation. These variables may include different physical quantities such as voltage, power, temperature, etc. By combining these variables, the operating state of the CLCC converter can be captured more comprehensively, thereby performing more effective knowledge fusion.

[0243] Preferably, the visualization software system for transparent operation and maintenance information of CLCC converters is constructed as follows:

[0244] Undirected edges are used to represent association relationships, including multi-source heterogeneous data such as physical structure, spatial location, and state information composed of triples. The theme information and second- and third-level facet information are marked by circles of different colors and sizes, and the common landing points are marked as cluster aggregations. At the same time, the hierarchical setting is realized by the association relationship of logical connection, that is, undirected edges are used to connect circles of different colors and sizes to illustrate the association relationship between the nodes represented by each circle;

[0245] Specifically, in the visualization software system of transparent operation and maintenance information of CLCC converters, if you click on one of the circles, you can see other related circles, that is, related knowledge, which is essentially divided into levels.

[0246] The visualization software system of transparent operation and maintenance information of CLCC converter is the presentation method of the final carrier. The system includes the organization of different data sources. Specifically, the system is mainly used to display the relationship between knowledge nodes and their aggregation structure. Undirected edges represent the association between different knowledge nodes. These nodes constitute multi-source heterogeneous data in the form of triples such as physical structure, spatial position, and status information. The color and size of the node are used to distinguish different subject information and their secondary and tertiary facet information, helping users to quickly identify different knowledge sources and their degree of association. Preferably, the size and color of the circle represent the same meaning; the large circle represents the knowledge subject, which may be a key knowledge source or core concept, such as the overall performance indicators of the CLCC converter, the transformer, etc.; the middle circle represents the secondary knowledge facet, including various parameters related to the main node, such as voltage, current, etc.; the small circle represents a more specific knowledge source or detailed information, which may be the real-time data or status information of a specific monitoring point.

[0247] For structured data such as information association, semi-structured data such as CLCC converter operation and maintenance network system, reports, and unstructured data such as monitoring videos and event messages, the data is represented and stored in the form of nodes, edges, and attributes to construct the CLCC converter knowledge ontology;

[0248] Through the association between CLCC converter entities and knowledge topics, knowledge data is classified into a knowledge forest from trunk to branches, taking into account the dynamic characteristics of CLCC converter power data; sequence, label and other variable information are used to represent the time-space state and define the knowledge distribution structure, and multi-granularity time labels are formulated for different businesses such as asset attributes, task flow and other time-varying information to meet the needs of dynamic knowledge processing. The multi-granularity classification is knowledge source, knowledge topic, knowledge facet and knowledge node; the purpose of formulating multi-granularity time labels is to meet the needs of dynamic knowledge processing, so that different types of information can be effectively managed and analyzed in the time dimension. First, the time granularity needs to be identified, and the required time granularity, such as seconds, minutes, hours, days, weeks, months, etc., is determined according to business needs. For example, for the real-time monitoring data of CLCC converters, labels at the second or minute level may be required, while historical maintenance records can be in units of days or weeks. Next, create corresponding time labels for each data type. Taking current monitoring data as an example, the time tag can be a current value recorded once a minute in the format of: time, current value; for example: 2024-10-2022:00,150A. At the same time, the event log can be marked in hours to record the specific time point when the event occurred, and the maintenance record can mark the execution time of the maintenance task in days or weeks to help manage maintenance plans and periodic inspections. Finally, according to data changes and business needs, the granularity of the time tag is adjusted in real time, so that the granularity of the time tag can be shortened for more frequent monitoring under high load conditions, and the granularity of the time tag can be expanded during low load periods to reduce data storage and processing pressure.

[0249] Step 4: Construct multiple fragmented knowledge mappings based on a single information source, and use the main entity structuring method in combination with the different structures of the initial multi-source information to realize the dynamic flow of transparent information of the CLCC converter, that is, after information or data is collected and processed from multiple sources, it can be transmitted, updated and applied between different parts of the system in real time or near real time.

[0250] The main entity structuring method includes the construction of knowledge propagation mechanism, dynamic representation of features, formation of dynamic knowledge model, dynamic update of behavior state and application of stacked autoencoder.

[0251] In step 4.1, to realize the dynamic flow of transparent information of CLCC converters, we must first build a knowledge dissemination mechanism, and then characterize the dynamic knowledge model based on the dissemination mechanism;

[0252] In the propagation mechanism, since the knowledge information of the CLCC converter has a coupling relationship, the propagation mechanism is expressed by the following formula:

[0253] ;

[0254] in, and Knowledge Node H Trainable parameters in the graph, is the relational representation of knowledge node H, H-1 represents the previous knowledge node associated with the current knowledge node H, which is realized by linear propagation or complex gate control; wherein the trainable parameters represent the trainable weights of knowledge node H in the process of knowledge propagation. They are used to adjust the importance of node H to the associated knowledge in the information flow. These parameters affect the relational representation between nodes, that is, how to integrate and propagate the information of the previous nodes associated with node H. In the initial setting stage, the parameters use small random numbers of uniform distribution or normal distribution as initial values, such as a mean of 0 and a standard deviation of 0.01; during the training process, these parameters should be dynamically adjusted according to the feedback of the loss function;

[0255] In the present invention, a graph neural network GNN is used for training.

[0256] The periodicity of the digital image is transformed by the transformation matrix , average estimated value d and deviation Mapped into the latent space,

[0257] ;

[0258] ;

[0259] ;

[0260] in, For periodic The transformation matrix, is the average estimate d The transformation matrix, Degree of deviation The transformation matrix, , and are knowledge nodes respectively H In periodic , average estimate d and deviation The trainable transformation matrix, , and For periodic , average estimate d and deviation Corresponding graph theory representation. The trainable transformation matrix is ​​used to map the features of knowledge nodes such as the periodic p average estimate d and the deviation v to the latent space during the dynamic propagation of knowledge of the CLCC converter. The acquisition of these transformation matrices is mainly divided into three steps: First, before starting training, the matrix is ​​usually initialized to a random small value, such as a normally distributed random number, so that it can gradually learn and adapt to the data characteristics during the training process. Then, during the training process, back propagation algorithms and optimization methods such as stochastic gradient descent or Adam optimization algorithm are used to continuously adjust the values ​​of these matrices, and the elements of the matrix are optimized by calculating the gradient of the propagation error to enhance the effect of knowledge propagation. Finally, if a deep learning model such as a neural network or a recursive neural network is used, these transformation matrices can be regarded as the weights of the network layer. As the training progresses, the weights are continuously updated to more accurately express and propagate the multi-source knowledge of the CLCC converter.

[0261] The digitization of graph representation is a key step in achieving knowledge dissemination and dynamic information processing. This involves digitizing the nodes, edges and their attributes of the graph. The specific method is as follows: First, the nodes are represented as feature vectors, which contain multiple attributes, such as the operating status of the CLCC converter, such as temperature and voltage, and a digital node representation is formed through standardization or feature extraction. Secondly, the digitization of edges usually uses an adjacency matrix to represent the connection relationship between nodes, and the matrix elements are assigned values ​​according to physical properties such as signal strength or correlation. For the representation of periodicity p, the time series data on the node or edge can be analyzed by Fourier transform, the periodic characteristics of the signal can be extracted, and converted into a digital representation in the frequency domain, such as the fluctuation amplitude of the node state. The deviation v reflects the deviation of the node state from the reference value. The deviation can be digitized and used as an additional attribute of the node, such as the deviation of the node voltage from the long-term average value. The average estimate d can be obtained by calculating the average value of the node attributes within a certain time window, and digitizing it using a sliding average or weighted moving average method. Through these steps, the multi-source data of the CLCC converter can be effectively converted into a digital form to achieve the dissemination and fusion of knowledge features.

[0262] It should be noted that in actual application, the difficulty in realizing dynamic propagation of knowledge of CLCC converters lies mainly in solving the problems under complex equilibrium constraints, and it is necessary to simplify features to enable knowledge nodes and knowledge topics to learn collaboratively. Therefore, the established knowledge feature representation and dynamic propagation of association relationships can convey various relationships such as time, accuracy, granularity, etc. in different scenarios, which is suitable for the actual operation and maintenance scenarios of CLCC converters.

[0263] Preferably, after satisfying the dynamic flow of knowledge of CLCC converters, it is necessary to establish dynamic update of CLCC converter knowledge information to meet actual operation and maintenance needs; atIndicates CLCC inverter knowledge topic th Knowledge of facets in a certain period of time, For this knowledge topic th The behavioral status of the knowledge ai exist ET - ST The total number of occurrences within a time period, ET - ST is the length of the time period, then the dynamic knowledge topics in this time period DM The mathematical model is:

[0264] ;

[0265] In the formula, Behavioral status of the subject of the knowledge ai , an is the number of behavioral states of the knowledge; the mathematical model of the dynamic knowledge subject is used to realize the dynamic update of CLCC converter knowledge information, which can meet the actual operation and maintenance needs.

[0266] Those skilled in the art should know that the behavior state is obtained by analyzing and processing the state sequence. The state sequence records the operating state changes of the CLCC converter at different times, such as power parameters including voltage, current, and power, environmental parameters including temperature and humidity, and equipment states including start-stop state, load change, etc. These state sequences can be extracted through time series segmentation, trend detection or pattern recognition to obtain at least one behavior state.

[0267] ET - ST The average behavior label under length needs to be based on the time interval ad The state average estimation is performed as shown in the following formula:

[0268] ;

[0269] in, For the si The total amount or measurement value of behavior tags in a period, s According to the time interval ad The total number of time periods divided, assuming that the time distance of the CLCC converter's operating state in a certain time period is , , …, , the time distance is divided into s interval, forming s Level , ,… , …, ; Among them, behavior refers to the specific state changes or events of the knowledge subject of the CLCC converter within a specific time period. Specifically, behavior can reflect the transition of different states of the CLCC converter during the operation and maintenance process, the execution of operation events, and the dynamic changes of related parameters, such as power data fluctuations, equipment operation state switching, fault occurrence and its processing. These behaviors are important dynamic features in the CLCC converter knowledge model, which are used to describe the operation and maintenance of the system at different time points. Behavior labels are used to classify and identify these behaviors. The setting of labels includes the division and annotation of time series, the periodic description of state changes, and the mapping and integration of multi-source heterogeneous data;

[0270] The division and labeling of time series is the division based on time interval ad mentioned in formula B, that is, by dividing the operating time period of the CLCC converter into several sub-intervals, each sub-interval represents a specific period, and the behavior label represents the number or frequency of occurrence of a specific behavior in the period. This division helps to capture and label the behavioral characteristics of the system in different time ranges, and set different behavior labels to reflect the frequency and characteristics of state changes.

[0271] The behavior label describing the periodicity of state changes can be set according to the periodic changes of the operating state of the CLCC converter. For example, if a state appears repeatedly within a certain time interval, this repeated behavior can be set as a periodic behavior label. In this way, the regular operation of the system can be effectively marked and analyzed.

[0272] Mapping and integration of multi-source heterogeneous data Through the integration of structured, semi-structured and unstructured data, the system can set different behavior labels to represent the behavior characteristics from different data sources. For example, statistical data from reports may correspond to real-time information from surveillance videos, and the fusion of these multi-source data can help label complex behavior states.

[0273] like Indicates the behavior state of knowledge subject th in the interval The number of occurrences, then the state periodicity of the knowledge behavior , as shown in the following formula:

[0274] ;

[0275] For the cycle of state changes of the CLCC converter's operating state, the frequency, change and change interval of the knowledge of different behavioral states of the CLCC converter are labeled and described. By updating different knowledge nodes and facet states, the changes of the CLCC converter's state at different times are constructed in the form of a dynamic knowledge model. represents the total amount or measurement value of the behavior label of the knowledge topic th in a subinterval divided by ad. Here, ,th is the number or cumulative value of the behavior labels that appear in the subinterval si. This measurement reflects the frequency or total number of behaviors that occur in each subinterval after the time series is divided, so it emphasizes the behavior statistics of a specific time interval. It represents the total number of occurrences of the behavior state of knowledge topic th in the entire time period ET−ST. ,th reflects the cumulative occurrence of a certain behavior state over the entire time period. Therefore, this metric tends to measure the overall frequency or total number of occurrences of a behavior over a longer time span.

[0276] The state periodicity of the knowledge behavior The formula describes the periodic changes in the operating state of the CLCC converter and the statistical process of its behavior state. This process characterizes the behavior pattern of the CLCC converter in different time periods through time interval division and frequency statistics. The specific explanation is as follows:

[0277] The measurement of time division and behavior label is to divide the entire time period into multiple sub-intervals s1…sn according to the time interval ad, and each interval represents a time period. In this way, there are a total of sn sub-intervals, that is, the total number of time periods, and the behavior label in each time period is Represents the total amount or measurement value of the behavior label in the si-th period.

[0278] The classification of operating status time distance means that the operating status of the CLCC converter in a certain period of time is recorded as a series of time distances, such as t1…tn. The distances between these time points can be classified according to the divided sn intervals, thus forming sn levels. Each level reflects the behavior in different time periods.

[0279] The statistics of behavior states are calculated by counting the number of times the behavior of knowledge subject th occurs in each subinterval si, that is, ,th, the occurrence frequency of the behavior in each time interval can be obtained.

[0280] Preferably, the multi-source knowledge of the CLCC converter is encoded using a stacked autoencoder. First, when training the stacked autoencoder, the input data is derived from the multi-source knowledge data of the CLCC converter, including structured data such as voltage and current values ​​in operating parameters, semi-structured data such as maintenance logs, event messages, and records, and unstructured data such as video surveillance information, pictures, and other multimedia materials. Before entering the stacked autoencoder, these data need to be normalized and standardized to convert different types of data into a unified numerical vector representation. Each sample in the input data set can be represented as a high-dimensional vector, and each dimension of the vector corresponds to a certain feature or information source. The goal of training is to enable the autoencoder to learn the potential feature representation of the input data. Through the encoding process, the original data is first compressed layer by layer into a lower-dimensional hidden feature space. These hidden features can better reveal the inherent structural relationship of the CLCC converter data, such as the coupling between the operating states of the equipment, the characteristics of trend changes, etc. In the decoding process, the autoencoder decodes the hidden features into a reconstructed output that is as similar to the input as possible. The purpose of training is to minimize the difference between the original input and the reconstructed output, that is, the reconstruction error. This training method continuously optimizes the network parameters including weight matrix and bias term through back propagation and gradient descent algorithm to improve the model's ability to represent data. Through the above training process, the stacked autoencoder can extract feature representations suitable for dynamic knowledge processing from multi-source heterogeneous data. These feature representations can reflect the behavior patterns and state changes of CLCC converters in different time periods. The multi-source knowledge features are extracted through the encoding results, and then the corresponding hidden layer features are obtained; the encoding results are knowledge features, which are extracted after the stacked autoencoder processes the multi-source data of the CLCC converter, and can reveal the complex relationship between the system operation status, behavior patterns and data. The extracted features include equipment operation parameter features, such as voltage fluctuations and current change trends; behavior pattern features, reflecting state changes and fault characteristics under different operation modes; time series features, capturing periodic, trending and sudden changes; coupling features of multi-source data, revealing the association between structured data, semi-structured data and unstructured data; spatial location features, reflecting the temperature distribution of physical components or the heat dissipation of the cooling system; and state transition features, showing the transition law between normal operation state and fault state.

[0281] The hidden layer features are shown as follows:

[0282] ;

[0283] ;

[0284] in, is the hidden layer feature of the multi-source knowledge attribute value of the CLCC converter, is the encoding label corresponding to the state cycle, is the hidden layer feature of the correlation between multi-source knowledge of CLCC converter, is the encoding label corresponding to the knowledge node.

[0285] The setting of the coding label of the state cycle is completed by feature extraction and classification of the state cycle characteristics of the CLCC converter. Specifically, it is first necessary to analyze the operating data of the equipment in different time periods to determine the pattern of periodic changes. For example, the periodic fluctuations of parameters such as voltage, current or temperature can be used as feature inputs, and these data are encoded by the stacked automatic encoder to extract their potential periodic characteristics. The coding label will be set according to the type, frequency and amplitude of these periodic characteristics to represent the category of state characteristics for each time period. During the setting process, the classification of labels can be based on different criteria, such as normal operating state, abnormal fluctuation state or different operating modes such as startup, stable operation, and shutdown. In addition, the division of time intervals such as hours, days, and weeks will also affect the setting of coding labels, so that the coding labels not only reflect the state characteristics, but also take into account the timing. The final coding label is used to identify and distinguish different state cycles to support dynamic monitoring and fault prediction of the converter.

[0286] The coding labels of knowledge nodes are set based on the analysis and feature extraction of the CLCC converter state cycle characteristics. The setting of the coding labels of knowledge nodes mainly relies on processing the operation data of the equipment in a specific time period to identify and describe different behavior states.

[0287] Specifically, the coding label setting process includes the following steps:

[0288] Data collection: Collect operating data from sensors or monitoring systems of CLCC converters. The data includes various operating parameters such as current, voltage, and temperature.

[0289] Feature extraction: Analyze the collected data to extract periodic features and behavior states. For example, identify high-frequency and low-frequency changes in data through statistical analysis to determine the operating characteristics of the device in different states.

[0290] Tag classification: Different status cycles are classified according to the results of feature extraction. Tags are set according to the operating status of the equipment, such as normal, warning, fault, etc., and can also be defined based on specific operating modes such as startup, stable operation, and shutdown.

[0291] Coding mapping: The classified state cycles are mapped to labels through a specific coding method. The normal state is represented by "0", the abnormal state is represented by "1", and different types of operation modes can be marked with different numbers or letters.

[0292] , Represent the network weight matrices of multi-source knowledge and association relationships, , are used to represent the bias vectors corresponding to multi-source knowledge and association relationships, f() is the Sigmoid nonlinear activation function. In the initial stage of deep learning, the bias vector is usually set to zero or a value close to zero to facilitate the stability of the training process. As the training progresses, the bias vector will be gradually adjusted and optimized to adapt to the distribution characteristics of the input data and the requirements of the model;

[0293] The network weight matrices of the multi-source knowledge and association relationships are learned from the data through the training process. The origins of these two weight matrices can be summarized in several aspects. First, when establishing a deep learning model such as a stacked autoencoder or other neural network, the weight matrix is ​​usually randomly initialized, and the initial value may be randomly generated according to a certain rule such as normal distribution or uniform distribution, so that the model can start learning from different starting points. Secondly, during the training phase, the model optimizes these weight matrices through the back propagation algorithm. Specifically, input data such as multi-source knowledge and its association relationships are transmitted through the network, and the output is calculated and compared with the true label, that is, the target value, and then the loss function is calculated. The model adjusts the weight matrix by minimizing this loss function, and the updating process involves optimization algorithms such as gradient descent. Furthermore, the learning of the weight matrix enables the model to extract effective features from the input data. For multi-source knowledge, the weight matrix can capture the mutual influence and importance between different knowledge sources; while for association relationships, the weight matrix is ​​used to describe the degree of connection between knowledge points. This feature learning ability enables the model to better understand and process complex data structures. Finally, the bias vectors are used in conjunction with the weight matrix to adjust the output of the model so that the model can better fit the data. The bias term can be regarded as an additional adjustment to the output, helping the network to respond more flexibly to changes in input data. By introducing the bias vector, the model can learn the distribution characteristics of the data more effectively. Therefore, the origin of the weight matrix is ​​dynamically learned through the training process of the model, reflecting the characteristics of multi-source knowledge and its correlation.

[0294] Preferably, the multi-source knowledge and association relationships encoded by the stacked autoencoder are subjected to canonical correlation analysis (CCA); thereby obtaining the overall correlation between the multi-source knowledge and the association relationships. Specifically, the goal of CCA is to find a method to project the two knowledge domains, namely, multi-source knowledge and association relationships, onto new coordinate axes respectively, so as to maximize the correlation between these projected features. Through this analysis, the potential correlation structure of the two data sets can be revealed, thereby mapping the features of the multi-source knowledge and association relationships into a shared semantic space. This helps to understand the intrinsic connection between multi-source data and enhance the overall expression ability of the knowledge system.

[0295] According to the knowledge characteristics, the maximum correlation coefficient matrix between the two knowledge domains can be obtained. and . The reason why there are two maximum correlation coefficient matrices between the two knowledge domains is that the canonical correlation analysis projects and optimizes the two knowledge domains separately. Specifically, one matrix is ​​used to represent the correlation of multi-source knowledge in the new projection space, and the other matrix is ​​used to represent the correlation of the association relationship in the new projection space. These two matrices respectively reflect the feature expressions of the two knowledge domains in the shared subspace, so that the correlation between the two is maximized in the projected space. The difference between them is that one is for the feature mapping of the multi-source knowledge domain, and the other is for the feature mapping of the association relationship domain, but the goal is to achieve the strongest correlation in the shared semantic space.

[0296] The maximum correlation coefficient matrix can be used to map the knowledge features of multi-source knowledge and association relationships into the semantic shared subspace;

[0297] The knowledge features refer to the mapping features in the semantic shared subspace after canonical correlation analysis. The knowledge features processed by CCA are the original features of multi-source knowledge and association relationships projected into the new shared semantic space to maximize the correlation between the two. These features are more representative and relevant in the new subspace, and better characterize the potential correlation between knowledge and relationships. Therefore, these knowledge features can be regarded as high-correlation features optimized by CCA, which are used for further analysis or fusion and prediction in subsequent model training.

[0298] Using the Matrix U S(2) and U T(2) For multi-source knowledge and association relationships ( U S , b S , U T , b T ) to merge:

[0299] ;

[0300] ;

[0301] in, is the fused knowledge feature matrix, For knowledge association and feature set, the process is based on the representation in the semantic shared subspace. is the feature map in the semantic shared subspace, which is H s Together we build the final fusion result. is the mapping value of the spatial output layer, are network training parameters, where the weight parameters are initialized to small random values ​​in the range of [-0.1, 0.1], the bias vector is initialized to a zero vector, and the learning rate range is 0.001 to 0.01. , They are respectively represented as the reconstruction errors of multi-source knowledge and association relationship after encoding and training of deep belief network; using the maximum correlation coefficient matrix and The projection vector V T and V S , the knowledge features of multi-source knowledge and association relationships can be converted into the problem of maximizing feature correlation, and the Lagrange multiplier and stochastic gradient descent method are used to respectively , , ( W S , b S )、( W T , b T )and( V T , V S ) is optimized, where W s for The relevant weight matrix of W T for The brackets are used to group the relevant model parameters, which makes it easier to distinguish different types of parameter sets, so as to more clearly indicate the specific role of each set of parameters in the process of integrating multi-source knowledge and association relationships. This grouping helps improve the readability and structure of the information in the text, allowing readers to quickly identify which parameters are commonly associated.

[0302] Previous studies have mainly focused on the construction of knowledge graphs and the extraction of knowledge features. These processes do not involve deep data pattern learning, so the intervention of deep belief networks (DBNs) is not required. However, when fusing features from different knowledge sources, the introduction of DBNs helps to enhance the overall performance of the model through its powerful feature learning capabilities, ensuring that the fused knowledge feature matrix can better represent the complexity of multi-source knowledge and associations.

[0303] The specific implementation of training and reconstruction is completed in the training process of the deep belief network. DBN continuously adjusts the weights and biases in the network through layer-by-layer pre-training to minimize the reconstruction error, thereby enhancing the model's ability to capture the characteristics of the input data. In this context, the fusion process is implemented based on formulas, and the introduction of DBN provides the necessary learning mechanism for the process, so as to optimize the representation and relevance of knowledge features through training.

[0304] Step 5: Use the identifier mRID to model the network topology of the knowledge graph in any graph database such as Neo4j, use the knowledge node selection function, i.e. NodeSelector, to associate knowledge nodes, initialize the knowledge entity object according to the relationship type between knowledge nodes, and call the generation function, i.e. create, to establish the association relationship of CLCC converter operation and maintenance knowledge. The association relationship of the operation and maintenance knowledge is the knowledge node association

[0305] Step 5.1: Analyze the CLCC converter knowledge unit obtained from the CIM / E file analysis according to the triple path mapping rule.

[0306] The main purpose of analyzing the CLCC converter knowledge units obtained from the analysis of CIM / E files is to extract and map the relationship between the knowledge units and the knowledge nodes in the graph database. Specifically, this analysis aims to identify and understand the structure, attributes and location of the knowledge units in the knowledge graph, so that knowledge can be effectively associated and integrated. Therefore, the focus of the analysis does include mapping relationships, but more broadly involves how to effectively transform knowledge units into usable information in the knowledge graph.

[0307] Knowledge units refer to specific information units extracted from CIM / E files, which include various attributes, states, and operation information. Knowledge units are not directly equivalent to knowledge nodes or triples, but their combination, which is converted into knowledge nodes or represented in the form of triples in further processing and analysis. Knowledge nodes refer to specific entities in the graph database, while triples are structured representations that describe these entities and their relationships. Therefore, knowledge units can be regarded as raw materials that provide basic data for building knowledge graphs.

[0308] Step 5.2: Identify the entities and relationships in the knowledge unit through mapping relationships and convert them into nodes and relationships in the graph database.

[0309] Step 5.3: Populate the nodes and relationships into a graph database such as neo4j network.

[0310] Step 5.4: Use the graph database operation module, such as the function provided by py2neo, to establish the network graph that ultimately corresponds to the CLCC converter entity model based on multi-source knowledge, and associate the knowledge relationship nodes.

[0311] Step 5.5: Visually display the structural information of the whole station system configuration file, such as the SCD file, and associate the knowledge information such as the structure function and operation status of the primary and secondary devices of the CLCC converter, traverse the multi-source knowledge nodes to form a visual connection between the knowledge graph and the entity model. The SCD file is a file format formulated by the power grid. The requirements of most power grid substations are file formats defined according to the IEC 61850 standard to facilitate system configuration, information exchange and interoperability between devices.

[0312] Preferably, in step 5, the constructed information carrier knowledge unit is stored in a graph database, and the objects and relationships between nodes of the triple mapping are imported to convert the text representation into a knowledge graph representation.

[0313] It should be noted that in step 5, the create function provided by py2neo is used to establish the final network graph corresponding to the CLCC converter entity model based on multi-source knowledge, and the knowledge topics corresponding to the CLCC converter components and their associated operating status, data information and other knowledge facets are associated;

[0314] The structural information of the SCD knowledge file is displayed visually, and the knowledge information such as the structure, function and operation status of the primary and secondary devices of the CLCC converter is associated. The multi-source knowledge nodes are traversed to form a visual connection between the knowledge graph and the entity model.

[0315] The traversal starts from step 5.1. In this step, after the CLCC converter knowledge unit is obtained by analyzing the CIM / E file, the relationship identification stage of the knowledge unit is entered, that is, step 5.2; in step 5.2, the entities and relationships in the knowledge unit are identified through mapping relationships and converted into nodes and relationships in the graph database. Then, in step 5.3, the nodes and relationships are implanted into the graph database to form a knowledge graph. In step 5.4, through the graph database operation module, a network graph corresponding to the CLCC converter entity model is established based on multi-source knowledge, and finally the visual connection between the knowledge graph and the entity model is realized in step 5.5.

[0316] The present application also discloses a CLCC converter multi-source information carrier construction system based on a knowledge graph based on a CLCC converter multi-source information carrier construction method, such as Figure 5 As shown, it includes a CLCC converter knowledge information building module, a CLCC converter hierarchy building module, a CLCC converter multi-source information knowledge graph building module, a CLCC converter knowledge dynamic flow building module and a CLCC converter network topology modeling module.

[0317] The CLCC converter knowledge information construction module is used to obtain CLCC converter related data, and use the converter related data to construct triples, and construct knowledge topics, knowledge facets, and knowledge nodes of the CLCC converter based on the triples;

[0318] The CLCC converter hierarchy construction module constructs the knowledge hierarchy of the converter based on the CLCC converter device entity; and performs vectorized coding on each knowledge facet and knowledge node;

[0319] The CLCC converter multi-source information knowledge graph construction module constructs the CLCC converter multi-source information knowledge graph based on the encoding result of step 2;

[0320] The CLCC converter knowledge dynamic flow construction module realizes the dynamic flow of CLCC converter transparent information based on the knowledge hierarchy of CLCC converter multi-source information;

[0321] The CLCC converter network topology modeling module performs network topology modeling on the CLCC converter multi-source information knowledge graph based on a graph database.

[0322] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0323] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the above. The computer-readable storage medium used herein is not to be interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0324] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0325] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0326] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph, characterized in that: The method comprises the following steps: Step 1: Acquire CLCC converter related data, use the converter related data to construct triples, and construct knowledge topics, knowledge facets, and knowledge nodes of the CLCC converter based on the triples; the triples are expressed in the form of entity-attribute-attribute value; Step 2: constructing a knowledge hierarchy of the converter based on the CLCC converter device entity; and vectorizing and encoding each knowledge facet and knowledge node; Step 3: Based on the vectorized encoding of each knowledge facet and knowledge node obtained in step 2, as well as the set of association relationships between each knowledge topic and its corresponding knowledge facet, a comprehensive description of the relationship between device entities is realized in the knowledge graph, thereby constructing a CLCC converter multi-source information knowledge graph; the transparent information carrier of the CLCC converter is constructed according to the following formula: Z : ； in, is the transparent information carrier of CLCC converter, and B is the knowledge source quantization value set of the knowledge node constructed in the knowledge graph at the current moment; is the quantitative value of the ti-th knowledge source, represents the variables related to all knowledge sources in the knowledge source quantization value set, is the conditional probability distribution parameter of the knowledge source of the ti-th CLCC converter, K is the total number of knowledge nodes constructed in the knowledge graph at the current moment, and E is the entity set of the CLCC converter corresponding to the aforementioned knowledge node; Step 4: Realize the dynamic flow of CLCC converter transparency information based on the knowledge hierarchy of CLCC converter multi-source information; Step 5: Based on the graph database, the network topology modeling of the CLCC converter multi-source information knowledge graph is carried out, and the transparent information carrier constructed in step 3 is used to realize the transparent display of the CLCC converter multi-source information.

2. According to claim 1, a method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph is characterized in that: In step 1, the relevant data of the CLCC converter includes sensor data of the CLCC converter valve hall, event messages, equipment drawings, nameplates, and electrical and non-electrical data representing the operating status of the converter.

3. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 1, the triples are constructed by using the data, which is to export the data into a common information model file, and then construct the triples based on the common information model file.

4. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 2, the hierarchical structure of the converter knowledge includes upper and lower levels and affiliations of the knowledge topics; The upper and lower levels of the knowledge topics and the affiliation relationships are described by defining a tree structure.

5. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 4, characterized in that: In the tree structure, a knowledge subject is defined as a parent node, i.e., an upper layer in the hierarchy; multiple knowledge facets corresponding to the knowledge subject are defined as child nodes, i.e., lower layers in the hierarchy; and the affiliation between the knowledge subject and each knowledge facet is displayed through the tree structure.

6. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 2, attribute constraints are established based on the knowledge hierarchy relationship; the attribute constraints include attribute type, allowed value range and number of data.

7. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 2, the weight of each knowledge topic is as follows: ； in, For the qi The knowledge weight value of each knowledge topic, wk For the qi The total number of knowledge facets associated with each knowledge topic. Indicates qi The total number of knowledge nodes corresponding to the knowledge topics and the knowledge facets associated with them, Representing knowledge topics X The weight threshold of .

8. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 2, the vectorized encoding of each knowledge facet and knowledge node is implemented based on a multi-layer perceptron.

9. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 2, the knowledge facets corresponding to different knowledge topics are connected and aligned in pairs.

10. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 3, The comprehensive description of the relationship between the equipment entities is shown in the following formula: ； in, It is a comprehensive description matrix of the relationship between equipment entities; The set of vectorized codes for knowledge nodes; Represents the relationship between knowledge nodes after vector encoding. When the first Knowledge nodes and When the knowledge nodes are related, ,when Knowledge nodes and If there is no association between knowledge nodes .

11. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 4, the method of using the main entity structuring method to realize the dynamic flow of CLCC converter transparency information is: constructing a knowledge propagation mechanism, and constructing a dynamic knowledge model based on the aforementioned propagation mechanism, so as to realize the dynamic flow of CLCC converter transparency information.

12. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 4, after satisfying the dynamic knowledge flow of CLCC converters, it is necessary to establish a mathematical model of dynamic knowledge topics to meet the actual operation and maintenance needs; Mathematical model of dynamic knowledge topics within a certain period of time DM for: ； In the formula, The first ai Behavioral state, The first aj Behavioral state; an is the total number of behavioral states of the knowledge subject; ET-ST For a set time period; Indicates th The knowledge topic corresponds to The total number of times a behavior state occurs within a set time period.

13. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: In step 5, the method for performing network topology modeling on the CLCC converter multi-source information knowledge graph based on the graph database is: Based on the CLCC converter multi-source information knowledge graph and the mapping path of triple information, the entities and inter-entity relationships in the knowledge nodes are identified and converted into nodes and inter-node relationships in the graph database; The nodes and relationships between nodes in the graph database are implanted into the graph database network to complete the network topology modeling.

14. The method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to claim 1, characterized in that: The mapping path of the triplet information includes an entity set of CLCC converter operation and maintenance objects, a relationship set between entities, an attribute set, a set of association relationships between entities and attributes, and a relationship set between triplet.

15. A CLCC converter multi-source information carrier construction system based on a knowledge graph for implementing the CLCC converter multi-source information carrier construction method according to any one of claims 1 to 14, comprising a CLCC converter knowledge information construction module, a CLCC converter hierarchy construction module, a CLCC converter multi-source information knowledge graph construction module, a CLCC converter knowledge dynamic flow construction module and a CLCC converter network topology modeling module; characterized in that: The CLCC converter knowledge information construction module is used to obtain CLCC converter related data, and use the converter related data to construct triples, and construct knowledge topics, knowledge facets, and knowledge nodes of the CLCC converter based on the triples; The CLCC converter hierarchy construction module constructs the knowledge hierarchy of the converter based on the CLCC converter device entity; and vectorizes and encodes each knowledge facet and knowledge node; The CLCC converter multi-source information knowledge graph construction module constructs the CLCC converter multi-source information knowledge graph based on the encoding result of step 2; The CLCC converter knowledge dynamic flow construction module realizes the dynamic flow of CLCC converter transparent information based on the knowledge hierarchy of CLCC converter multi-source information; The CLCC converter network topology modeling module performs network topology modeling on the CLCC converter multi-source information knowledge graph based on a graph database.

16. An electronic device, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph according to any one of claims 1-14.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the method for constructing a multi-source information carrier of a CLCC converter based on a knowledge graph as described in any one of claims 1 to 14.

Images