A method for recommending business functions of measurement systems based on knowledge graph

Abstract

The present invention discloses a method for recommending business functions for a measurement system based on a knowledge graph. The method comprises: step S1, obtaining user historical access archive records, preprocessing them, and importing them into a knowledge graph; step S2, constructing a subgraph based on the topological relationships of the knowledge graph; step S3, modeling the subgraph and training model parameters based on a hierarchical attention model to obtain a trained recommendation model; and step S4, deploying the trained recommendation model. Whenever a business user accesses the system, the recommendation model is used to intelligently push relevant information about the business corresponding to their role to the user based on real-time fault monitoring status and real-time access data status of other users. Implementing the present invention can reduce algorithm complexity, improve reliability, and enhance recommendation accuracy.

Description

Technical Field

[0001] The present invention relates to the field of electric power measurement technology, and in particular to a method for recommending business functions of a measurement system based on a knowledge graph. Background Art

[0002] Due to the increasing demand for measurement services and the expansion of energy data application modules, the existing measurement system has multiple functional layers and complex data query paths. Even for professionals familiar with the business, daily operations require considerable time and effort to query relevant data, establish connections between different data, and identify relevant business needs. The existing system can only locate specific modules through conditional or fuzzy queries and lacks business support functions based on user attributes of business personnel, which limits the efficiency and quality of operations and maintenance. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method for recommending business functions of a measurement system based on a knowledge graph, which can reduce the complexity of the algorithm and improve reliability and recommendation accuracy.

[0004] To solve the above technical problems, as one aspect of the present invention, a method for recommending business functions of a measurement system based on a knowledge graph is provided, which comprises at least the following steps:

[0005] Step S1: Obtain the archival records of all operation and maintenance platform users accessing the system from the system background, focusing on recording the access time, access path, and whether the data was downloaded. The data is cleaned and preprocessed, and imported into the predefined knowledge graph.

[0006] Step S2: Generate a subgraph representing the relationship between users and business modules based on the knowledge graph topology relationship;

[0007] Step S3: Based on the hierarchical attention model, the relationship between users and business modules in the sub-graph is modeled and the model parameters are trained to obtain a trained recommendation model;

[0008] Step S4: deploy the trained recommendation model in the existing measurement system. Whenever a business user accesses the system, the recommendation model is used to intelligently push relevant information about the business corresponding to the role to the user based on the real-time monitoring status of faults and the real-time access data status of other users.

[0009] Preferably, in step S1, the predefined knowledge graph is a knowledge graph based on the CIM 61970 power grid information model, which includes content related to geographical areas, substations, voltage level areas, various types of equipment entities, and connection methods between entities.

[0010] Preferably, the step S2 further comprises:

[0011] Step S21: For each "user-business module" combination, use the DeepWalk algorithm to find all paths from the user to the business module in the knowledge graph with a length of less than 6, exclude paths with more than a predetermined number of intermediate links, and retain directly connected business chains;

[0012] Step S22: assign a corresponding weight to each path based on the path length using the following method:

[0013]

[0014] Among them, |P i | is the length of the i-th path between the user and the business module, so that the model assigns higher weights to entities that are directly related or closely associated with the user role;

[0015] Step S23: Recombine all filtered paths to form a sub-knowledge graph.

[0016] Preferably, step S3 further comprises:

[0017] Step S31: Based on semantic propagation and semantic aggregation modeling, semantic information of the paths of the sub-knowledge graph is deduced and aggregated layer by layer, so that the model fully contains all semantic information related to the user entity, and the business knowledge in the field of power measurement is represented by the parameters in the model;

[0018] Step S32: For each entity vector calculated in step S31, the importance weight is controlled through the attention model mechanism A(u;i), so that the model learns the relationship between important users and business module data;

[0019] Step S33, optimizing all parameters in steps S2 to S3 by using a gradient descent algorithm, calculating the probability of transmission between the user entity and the technical document / material entity based on the user access business module record, and using the cross entropy function C as the objective function;

[0020] c=Σ(σ(y*ln(p ij )+(1-y)*ln(1-p ij )))

[0021] Among them, e i Represents the user entity vector, e j Represents the business module entity vector, using the semantic model to calculate e i Passed to the business module entity e through the path in the knowledge graph j The probability is p ij , y indicates whether the user has access records to the business module. If yes, y = 1, otherwise y = 0.

[0022] Preferably, the step S31 further includes:

[0023] Step S311: Initialize the embedding vectors of all entities, types, and relationships in the subgraph, denoted as e h , t h and r h、k , concatenate the following initialization entity vectors:

[0024]

[0025] in, Operation represents the concatenation of vectors, σ represents the sigmoid operation, W0 and b represent the parameters to be trained;

[0026] Step S312: semantic propagation and semantic aggregation are performed l times, where l is the length of the path from the user to the data. The specific steps of each cycle are as follows:

[0027] Join all adjacent entities of entity e k With the relationship vector r h、k , the semantic vector is calculated using the following formula:

[0028]

[0029] The entity vectors of the next layer are calculated by combining all semantic vectors and the entity vectors of the previous layer using the following formula:

[0030]

[0031] Preferably, the step S32 further includes:

[0032] Get the entity vector by dot product The correlation between the semantic vector and other semantic vectors is normalized by weighting the result of the dot product through the Softmax function, so that the converted semantic vector can represent the importance of the semantics in the context of the overall knowledge graph.

[0033] The implementation of the embodiments of the present invention has the following beneficial effects:

[0034] The present invention provides a method for recommending business functions of a measurement system based on a knowledge graph, which obtains user historical access archive records, performs preprocessing, and imports them into a knowledge graph; constructs a subgraph based on the topological relationship of the knowledge graph; models the subgraph based on a hierarchical attention model and trains model parameters to obtain a trained recommendation model; deploys the trained recommendation model, and whenever a business personnel user accesses the system, utilizes the recommendation model based on real-time monitoring of fault conditions and real-time access data status of other users to intelligently push relevant information about the business corresponding to the role to the user. The technical solution of the example of the present invention solves the problem that there are many interfaces in the existing business system, and users need to spend a lot of working time searching for business-related supporting materials, so that business personnel can concentrate their time on analyzing and handling business problems, concentrate operation and maintenance resources, and greatly improve business efficiency;

[0035] By selecting parameterized entity relationship paths, the model assigns weights to path importance based on path length. Relying on a knowledge graph of typical power measurement services, the present invention uses a knowledge graph embedding method based on a hierarchical attention model to construct a recommendation algorithm model. Building on existing technologies, this approach reduces unnecessary complexity in existing algorithms and enhances the reliability, rationality, and recommendation accuracy of existing optimal algorithms.

[0036] At the same time, the present invention is applicable to the functional application of measurement systems. Developing measurement business systems or upgrading existing systems based on the core technology designed by the present invention can effectively reduce the repetitive workload of measurement business personnel. In response to query tasks for measurement business emergencies, the recommendation algorithm involved in the present invention helps assist business personnel in quickly locating crisis paths and searching for information related to crisis resolution, thereby improving the efficiency of business personnel in resolving emergency events and supporting the reliable operation of power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, other drawings obtained based on these drawings still fall within the scope of the present invention.

[0038] Figure 1 A schematic diagram of the main process of an embodiment of a method for recommending business functions of a measurement system based on a knowledge graph provided by the present invention. DETAILED DESCRIPTION

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in further detail below with reference to the accompanying drawings.

[0040] like Figure 1 FIG. 1 shows a schematic diagram of the main process of an embodiment of a method for recommending business functions of a measurement system based on a knowledge graph provided by the present invention. In this embodiment, the method includes at least the following steps:

[0041] Step S1: Obtain archival records of all users accessing the system from the system backend, focusing on features such as access time, access path, and whether data was downloaded. The data is cleaned and preprocessed, and imported into a predefined knowledge graph. In a specific example, the predefined knowledge graph can be a knowledge graph based on the CIM 61970 power grid information model, which includes entities such as geographical regions, substations, voltage level zones, various types of equipment, and the relationships between the connections between entities.

[0042] Step S2: Generate a subgraph representing the relationship between users and business modules based on the knowledge graph topology relationship;

[0043] In a specific example, step S2 specifically includes:

[0044] Step S21: For each "user-business module" combination, use the DeepWalk algorithm to find all paths from the user to the business module in the knowledge graph with a total length less than a predetermined number (e.g., 6), exclude paths with more than a predetermined number of intermediate links, and focus on directly connected business chains;

[0045] Step S22: assign a weight to each path based on the path length using the following method:

[0046]

[0047] Among them, |P i | is the length of the i-th path between the user and the business module, so that the model assigns higher weights to entities that are directly related or closely associated with the user role;

[0048] Step S23: Recombine all filtered paths to form a sub-knowledge graph.

[0049] Step S3: Based on the hierarchical attention model, the relationship between users and business modules in the sub-graph is modeled and the model parameters are trained to obtain a trained recommendation model;

[0050] In a specific example, step S3 further includes:

[0051] Step S31: Based on semantic propagation and semantic aggregation modeling, the semantic information of the paths in the sub-knowledge graph in step S22 is deduced and aggregated layer by layer, so that the model fully contains all semantic information related to the user entity, and the business knowledge in the field of power measurement is represented by the parameters in the model;

[0052] Specifically, step S31 further includes:

[0053] Step S311: Initialize the embedding vectors of all entities, types, and relationships in the subgraph, denoted as e h , t h and r h、k , concatenate the following initialized entity vectors:

[0054]

[0055] in, Operation represents the concatenation of vectors, σ represents the sigmoid operation, W0 and b represent the parameters to be trained;

[0056] S312: Perform semantic propagation and semantic aggregation l times, where l is the length of the path from the user to the data. The specific steps of each cycle are as follows:

[0057] Join all adjacent entities of entity e k With the relationship vector r h、k , calculate the semantic vector:

[0058]

[0059] Calculate the entity vector of the next layer through all semantic vectors and the entity vector of the previous layer:

[0060]

[0061] Step S32: For each semantic vector calculated in step S31, the importance weight is controlled by the attention model mechanism A(u;i). The specific implementation method is: the entity vector is obtained by dot product. The correlation between the semantic vector and other semantic vectors is normalized by weighting the result of the dot product through the Softmax function, so that the converted semantic vector can represent the importance of the semantics in the context of the overall knowledge graph;

[0062] In step S33, all parameters in step S2 to step S3 are optimized by the gradient descent algorithm, and the probability of transmission between the user entity and the technical document / data entity is calculated based on the user access business module record, with the cross entropy function C as the objective function.

[0063] c=Σ(σ(y*ln(p ij )+(1-y)*ln(1-p ij )))

[0064] Among them, e i Represents the user entity vector, e jRepresents the business module entity vector, using the semantic model to calculate e i Passed to the business module entity e through the path in the knowledge graph j The probability is p ij , y indicates whether the user has access records to the business module. If yes, y = 1, otherwise y = 0.

[0065] Step S4: Design and deploy a container module for the recommendation model trained in step S3 in the existing measurement system. Whenever a business user accesses the system, the recommendation model in step S3 is based on real-time monitoring of faults and real-time access to data status by other users, and intelligently pushes business-related information corresponding to the role to the user, such as pushing business demand work orders and support and reference materials related to the user to the operation and maintenance system user.

[0066] It can be understood that the recommendation model described in step S3 is based on the current fault situation and the user entity vector e i The model begins semantic propagation by filtering business module entities with a propagation probability above a threshold α and intelligently pushes them to users. The α value is set by the user based on current business needs. A higher α value results in fewer recommended content, but with higher business relevance, while a lower α value results in more recommended content, but with a lower relevance threshold.

[0067] The implementation of the embodiments of the present invention has the following beneficial effects:

[0068] The present invention provides a method for recommending business functions of a measurement system based on a knowledge graph, which obtains user historical access archive records, performs preprocessing, and imports them into a knowledge graph; constructs a subgraph based on the topological relationship of the knowledge graph; models the subgraph based on a hierarchical attention model and trains model parameters to obtain a trained recommendation model; deploys the trained recommendation model, and whenever a business personnel user accesses the system, utilizes the recommendation model based on real-time monitoring of fault conditions and real-time access data status of other users to intelligently push relevant information about the business corresponding to the role to the user. The technical solution of the example of the present invention solves the problem that there are many interfaces in the existing business system, and users need to spend a lot of working time searching for business-related supporting materials, so that business personnel can concentrate their time on analyzing and handling business problems, concentrate operation and maintenance resources, and greatly improve business efficiency;

[0069] By selecting parameterized entity relationship paths, the model assigns weights to path importance based on path length. Relying on a knowledge graph of typical power measurement services, the present invention uses a knowledge graph embedding method based on a hierarchical attention model to construct a recommendation algorithm model. Building on existing technologies, this approach reduces unnecessary complexity in existing algorithms and enhances the reliability, rationality, and recommendation accuracy of existing optimal algorithms.

[0070] At the same time, the present invention is applicable to the functional application of measurement systems. Developing measurement business systems or upgrading existing systems based on the core technology designed by the present invention can effectively reduce the repetitive workload of measurement business personnel. In response to query tasks for measurement business emergencies, the recommendation algorithm involved in the present invention helps assist business personnel in quickly locating crisis paths and searching for information related to crisis resolution, thereby improving the efficiency of business personnel in resolving emergency events and supporting the reliable operation of power systems.

[0071] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] The above disclosure is only a preferred embodiment of the present invention and certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for recommending business functions of a measurement system based on knowledge graph, characterized in that: At least the following steps are included: Step S1: Obtain the archival records of all operation and maintenance platform users accessing the system from the system background, record the access time, access path, and whether the data is downloaded, clean and pre-process the data, and import it into the predefined knowledge graph; Step S2: Generate a subgraph representing the relationship between users and business modules based on the knowledge graph topology relationship; Step S3: Based on the hierarchical attention model, the relationship between users and business modules in the sub-graph is modeled and the model parameters are trained to obtain a trained recommendation model; Step S4: Deploy the trained recommendation model in the existing measurement system. Whenever a business user accesses the system, the recommendation model is used to intelligently push relevant information about the business corresponding to the role to the user based on the real-time monitoring status of faults and the real-time access data status of other users. Wherein, the step S3 further includes: Step S31: Based on semantic propagation and semantic aggregation modeling, semantic information of the paths of the sub-knowledge graph is deduced and aggregated layer by layer, so that the model fully contains all semantic information related to the user entity, and the business knowledge in the field of power measurement is represented by the parameters in the model; Step S32: For each entity vector calculated in step S31, the importance weight is controlled through the attention model mechanism A(u; i) so that the model learns the relationship between important users and business module data; Step S33, optimize all parameters in step S2 to step S3 by gradient descent algorithm, calculate the probability of transmission between user entity and technical document or data entity according to the record of user accessing business module, and use cross entropy function C As the objective function; in, e i represents the user entity vector, e j Represents the entity vector of the business module, calculated using the semantic model e i Passed to business module entities through paths in the knowledge graph e j The probability of p ij , y indicates whether the user has access records to the business module. If yes, y=1, otherwise y=0.

2. The method according to claim 1, wherein In step S1, the predefined knowledge graph is a knowledge graph based on the CIM 61970 power grid information model, which includes the relationship between geographical areas, substations, voltage level areas, various types of equipment entities and connection methods between entities.

3. The method according to claim 2, wherein The step S2 further comprises: Step S21: For each "user-business module" combination, use the deep walk algorithm to find all paths from the user to the business module in the knowledge graph whose length is less than a predetermined number, exclude paths with more than a predetermined number of intermediate links, and retain directly connected business chains; Step S22: assign a corresponding weight to each path based on the path length using the following method: in, The length of the i-th path between the user and the business module enables the model to assign higher weights to entities that are directly related or closely associated with the user role; Step S23: Recombine all filtered paths to form a sub-knowledge graph.

4. The method according to claim 3, wherein The step S31 further comprises: Step S311: Initialize the embedding vectors of all entities, types, and relationships in the subgraph, denoted as e h , t h and r h、k , concatenate the following initialization entity vectors: Among them, the ⊕ operation represents the concatenation of vectors, σ Represents the sigmoid operation, W 0 and b Represents the parameters to be trained; Step S312: semantic propagation and semantic aggregation l Second-rate, l is the length of the path from the user to the data, where each loop step is as follows: Join all adjacent entities of entity e k The embedding vector r of the relationship h、k , the semantic vector is calculated using the following formula: The entity vectors of the next layer are calculated by combining all semantic vectors and the entity vectors of the previous layer using the following formula: 。 5. The method according to claim 4, wherein The step S32 further includes: Get the entity vector by dot product The correlation between the semantic vector and other semantic vectors is normalized by weighting the result of the dot product through the Softmax function, so that the converted semantic vector represents the importance of the semantics in the context of the overall knowledge graph.

Images