Electric power measurement large model training method, apparatus and device, and storage medium
By constructing a knowledge graph for power metering and screening sample corpora, a large-scale power metering model was trained, which solved the problem of insufficient knowledge in the application of general large-scale models in the power industry and achieved highly accurate and reliable power metering services.
Patent Information
- Application Number
- CN202511639293.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-10
- Publication Date
- 2026-02-06
AI Technical Summary
Existing technologies make it difficult to build professional and reliable large-scale metering models for the power industry. General-purpose large-scale models suffer from insufficient knowledge and low credibility when applied in the power industry, resulting in feedback results that do not match the facts or specific business scenarios.
By acquiring initial electricity metering knowledge from the power industry, a target electricity metering knowledge graph is constructed. The sample corpus is expanded and filtered out. An initial large model is trained using the target sample corpus to form a target electricity metering large model. The model is then optimized by combining an interactive interface and a feedback traceability module.
This enhances the professionalism and reliability of the large-scale power metering model, meets the high accuracy and reliability requirements of power metering services, and reduces the probability of errors.
Smart Images
Figure CN121480718A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of electric power metering, in particular to an electric power metering large model training method, device and equipment and storage medium. BACKGROUND
[0002] At present, electric power enterprises have used automation technology in daily operation and business implementation, but in the work of metering data collection and processing, business quality control and supervision, abnormal risk investigation and disposal, which are highly professional and have high technical threshold, intelligent means still play a supporting role more. The lack of understanding and processing ability of professional knowledge limits the further application of emerging technologies.
[0003] To solve this problem, knowledge graph technology has attracted attention from the electric power industry. With the advantages of structured knowledge expression and reasoning, knowledge graph has quickly realized the deep application of electric power business. However, at the same time, problems such as low quality of basic data, high training and maintenance cost, and difficult knowledge annotation have also emerged. In recent years, although large model technology has stronger knowledge cognition and generation ability, its too complex reasoning logic makes its decision-making process similar to a "black box" process, which limits the credibility and acceptability of the information provided by the large model, and it is difficult to meet the application requirements of high accuracy, high timeliness and high traceability of metering services. Moreover, the widely used large models in the industry are general large models or general-purpose large models. Their training process and training corpus pay more attention to the understanding, memory and application ability of general knowledge, and it is difficult to achieve good results for metering professional problems in the electric power industry which are highly professional, traceable and highly interpretable. The feedback results do not match the facts, the general logic or the specific business scenarios due to the illusion problem often occur, which has objectively limited the application prospect of large model technology in the metering profession of the electric power industry.
[0004] In summary, how to build a professional and reliable electric power metering large model is a technical problem to be solved at present. SUMMARY
[0005] Therefore, the purpose of the present application is to provide an electric power metering large model training method, device, equipment and storage medium, which can build a professional and reliable electric power metering large model. The specific solutions are as follows:
[0006] In a first aspect, the present application provides an electric power metering large model training method, comprising:
[0007] Obtaining a plurality of initial electric power metering knowledge of the electric power industry, and constructing a target electric power metering knowledge graph based on each of the initial electric power metering knowledge;
[0008] The initial power metering knowledge is expanded to obtain a plurality of to-be-verified power metering knowledge, and an initial sample corpus is constructed based on the to-be-verified power metering knowledge.
[0009] Based on the target power metering knowledge graph, each of the to-be-verified power metering knowledge in the initial sample corpus is screened out to obtain a plurality of target power metering knowledge, and a target sample corpus is determined based on each of the target power metering knowledge.
[0010] Each of the target power metering knowledge in the target sample corpus is used to train an initial large model to obtain a target power metering large model, so as to provide metering business services in the power industry by using the target power metering large model.
[0011] Optionally, the obtaining of the plurality of initial power metering knowledge in the power industry comprises:
[0012] The legal regulations, industry regulations, technical standards and business cases corresponding to the power metering data of the power industry are determined, and based on the legal regulations, the industry regulations, the technical standards and the business cases, a plurality of the initial power metering knowledge in the power industry is obtained.
[0013] Optionally, the constructing of the target power metering knowledge graph based on each of the initial power metering knowledge comprises:
[0014] A plurality of initial triples corresponding to each of the initial power metering knowledge are extracted;
[0015] Each of the target entities in each of the initial triples is identified, and the relationship between each of the target entities in each of the initial triples is extracted to obtain a corresponding relationship extraction result;
[0016] Each of the target entities is taken as a node, and based on the relationship extraction result, a line is established between the nodes corresponding to different target entities having an association relationship, so as to construct an initial power metering knowledge graph corresponding to the initial power metering knowledge;
[0017] The target power metering knowledge graph is determined based on the initial power metering knowledge graph.
[0018] Optionally, the determining of the target power metering knowledge graph based on the initial power metering knowledge graph comprises:
[0019] The nodes with conflicts in the initial power metering knowledge graph are determined, and the nodes with conflicts are disambiguated to obtain adjusted nodes;
[0020] The target power metering knowledge graph is determined based on the adjusted nodes in the initial power metering knowledge graph.
[0021] Optionally, the expansion of each of the initial power metering knowledge to obtain several power metering knowledge to be verified includes:
[0022] The initial power metering knowledge described above is formally transformed to obtain several formally transformed power metering knowledge;
[0023] The transformed electricity metering knowledge in each of the aforementioned forms is randomly combined to obtain several combined electricity metering knowledge;
[0024] Based on the initial power metering knowledge, the power metering knowledge after transformation, and the combined power metering knowledge, several power metering knowledge to be verified are determined.
[0025] Optionally, the step of filtering out each of the unverified power metering knowledge in the initial sample corpus based on the target power metering knowledge graph to obtain several target power metering knowledge includes:
[0026] Determine each target triplet corresponding to the target power metering knowledge graph;
[0027] The consistency verification results are obtained by comparing each of the power metering knowledge to be verified in the initial sample corpus with each of the target triples in the target power metering knowledge graph.
[0028] Based on the consistency verification results, power metering knowledge that conflicts with any of the target triples is identified from each of the power metering knowledge to be verified and is to be screened out.
[0029] The electricity metering knowledge to be screened out is removed from the initial sample corpus to obtain several target electricity metering knowledge.
[0030] Optionally, the large-scale electricity metering model training method further includes:
[0031] Construct the interactive interface corresponding to the target large-scale electricity metering model;
[0032] Through the interactive interface, the target power metering big model is used to receive target metering business questions from users in the power industry and generate corresponding answer information.
[0033] Obtain feedback information generated by the user that corresponds to the answer information, and determine the target triplet corresponding to the answer information from the target power metering knowledge graph based on the feedback information;
[0034] If it is determined based on the feedback information that the solution information contains an error, then the target triple corresponding to the solution information is corrected based on the feedback information, the target power metering knowledge graph is updated based on the corrected target triple, and the target power metering large model is updated based on the updated target power metering knowledge graph.
[0035] Secondly, this application provides a large-scale power metering model training device, comprising:
[0036] The knowledge graph construction module is used to acquire several initial power metering knowledge in the power industry and construct a target power metering knowledge graph based on each of the initial power metering knowledge.
[0037] The initial sample corpus construction module is used to expand the initial power metering knowledge to obtain several power metering knowledge to be verified, and to construct the initial sample corpus based on the power metering knowledge to be verified.
[0038] The target sample corpus determination module is used to filter out each of the power measurement knowledge to be verified in the initial sample corpus based on the target power measurement knowledge graph to obtain a number of target power measurement knowledge, and to determine the target sample corpus based on each of the target power measurement knowledge.
[0039] The power metering large model training module is used to train the initial large model using the target power metering knowledge in the target sample corpus to obtain the target power metering large model, so as to provide metering business services for the power industry using the target power metering large model.
[0040] Thirdly, this application provides an electronic device, comprising:
[0041] Memory, used to store computer programs;
[0042] A processor is used to execute the computer program to implement the aforementioned large-scale power metering model training method.
[0043] Fourthly, this application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned large-scale power metering model training method.
[0044] In this application, firstly, several initial power metering knowledge items are acquired from the power industry, and a target power metering knowledge graph is constructed based on these initial power metering knowledge items. Then, the initial power metering knowledge items are expanded to obtain several power metering knowledge items to be verified, and an initial sample corpus is constructed based on these initial power metering knowledge items. Subsequently, based on the target power metering knowledge graph, the initial sample corpus is filtered to obtain several target power metering knowledge items, and a target sample corpus is determined based on these target power metering knowledge items. Finally, the initial large-scale model is trained using the target power metering knowledge items in the target sample corpus to obtain a target power metering large-scale model, which is then used to provide metering services for the power industry. As can be seen from the above, this application first acquires initial power metering knowledge and constructs a target power metering knowledge graph. Then, it expands the initial power metering knowledge to obtain power metering knowledge to be verified, thus constructing an initial sample corpus. Next, it uses the target knowledge graph to filter out the power metering knowledge to be verified, thereby determining the target sample corpus. Finally, it uses the target corpus to train the initial large-scale model to obtain the target power metering large-scale model, which can be used to provide power metering business services. In this way, this application achieves the screening and conflict disambiguation of power metering knowledge through the target power metering knowledge graph, solving the problem that the training corpus of the general large-scale model focuses on general knowledge but lacks sufficient power metering knowledge, thus improving the reliability of the target power metering large-scale model. Simultaneously, the target sample corpus obtained through the screening of the target power metering knowledge graph ensures the professionalism and accuracy of the large-scale model training data, enabling the target power metering large-scale model to possess stronger metering professional analysis and reasoning capabilities, meeting the high accuracy and high reliability requirements of power metering services. Attached Figure Description
[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0046] Figure 1 A flowchart of a large-scale power metering model training method provided in this application;
[0047] Figure 2 A flowchart illustrating a specific method for training a large-scale power metering model is provided in this application.
[0048] Figure 3 A schematic diagram of a large-scale power metering model training device provided in this application;
[0049] Figure 4 This application provides a structural diagram of an electronic device. Detailed Implementation
[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0051] Currently, knowledge graph technology has attracted attention in the power industry. With its advantages in structured knowledge representation and reasoning, knowledge graphs have quickly achieved deep application in power business. However, problems such as low quality of basic data, high training and maintenance costs, and difficulties in knowledge annotation have also emerged. While large-scale modeling technology has demonstrated stronger knowledge recognition and generation capabilities in recent years, its overly complex reasoning logic makes its decision-making process essentially a "black box" process. This limitation restricts the credibility and acceptability of information provided by large-scale models, making it difficult to meet the application requirements of high accuracy, high timeliness, and high traceability in metering services. Furthermore, the large-scale models widely used in the industry are generally general-purpose or generic models. Their training process and training corpora focus more on understanding, memorizing, and applying general knowledge, making it difficult to achieve good results for highly specialized, traceable, and interpretable metering problems in the power industry. The "illusion problem" frequently leads to feedback results that are inconsistent with facts, general logic, or specific business scenarios, objectively limiting the application prospects of large-scale modeling technology in the metering field of the power industry. Therefore, this application provides a large-scale power metering model training scheme that can construct a professional and reliable large-scale power industry metering model.
[0052] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for training a large-scale electricity metering model, which may include:
[0053] Step S11: Obtain several initial power metering knowledge from the power industry, and construct a target power metering knowledge graph based on each of the initial power metering knowledge.
[0054] In this embodiment, it is first necessary to acquire some initial power metering knowledge of the power industry. The specific process may include: determining the laws, regulations, industry rules, technical standards, and business cases corresponding to the power industry's power metering data, and acquiring this initial power metering knowledge based on these laws, regulations, technical standards, and business cases. Specifically, power metering expertise is the data source used for constructing and training large-scale metering models, mainly composed of data in the form of power metering-related laws, regulations, industry rules, technical standards, typical designs, business cases, and calculation examples.
[0055] It should be noted that this embodiment can construct a target power metering knowledge graph based on collected initial power metering knowledge. The specific process may include: first, extracting several initial triples corresponding to each initial power metering knowledge; then, identifying each target entity in each initial triple and extracting the relationships between the target entities in each initial triple to obtain corresponding relationship extraction results; subsequently, using each target entity as a node, and establishing connections between nodes corresponding to different target entities with related relationships based on the relationship extraction results to construct an initial power metering knowledge graph corresponding to the initial power metering knowledge; finally, determining the target power metering knowledge graph based on the initial power metering knowledge graph. Specifically, the power metering knowledge graph is a graph database constructed based on power metering knowledge, mainly extracting and representing entities in the metering profession and the relationships between entities. Semantic association and disambiguation are performed during the construction process to ensure the high accuracy of its association reasoning chain. To construct the power metering knowledge graph, it is first necessary to extract initial triples corresponding to each initial power metering knowledge, identify the target entities in the initial triples, extract the relationships between the target entities, and obtain relationship extraction results. Next, an initial power metering knowledge graph is constructed, with the target entity as the node and the relationships extracted as edges. Finally, the initial power metering knowledge graph needs to be further refined to determine the target power metering knowledge graph.
[0056] In this embodiment, the process of determining the target power metering knowledge graph based on the initial power metering knowledge graph can include: first, identifying conflicting nodes in the initial power metering knowledge graph and disambiguating these nodes to obtain adjusted nodes; then, determining the target power metering knowledge graph based on the adjusted nodes in the initial power metering knowledge graph. In one specific implementation, a power metering knowledge priority rule can be established to clarify the validity hierarchy of knowledge from different sources, such as current standards taking precedence over older cases. Next, entities and relationships in the knowledge graph can be cross-validated, comparing multi-source data to identify conflicting nodes in the initial power metering knowledge graph. Then, for conflicting content, correct information is determined according to the priority rule, and erroneous entries are removed or corrected to obtain adjusted nodes; for example, when the verification cycle description is inconsistent, the latest industry standard takes precedence. Finally, the target power metering knowledge graph is determined based on the adjusted nodes and the nodes in the initial power metering knowledge graph that did not originally conflict.
[0057] Step S12: Expand the initial power metering knowledge to obtain several power metering knowledge to be verified, and construct an initial sample corpus based on the power metering knowledge to be verified.
[0058] In this embodiment, to expand the training sample data volume, a sample expansion module is designed to combine and formally transform the limited training sample data, thereby expanding the training sample data volume, adapting to the training requirements of large models, and improving the training effect of large models. The specific workflow of the sample expansion module may include: firstly, formally transforming each initial power metering knowledge to obtain several formally transformed power metering knowledge; then, randomly combining each formally transformed power metering knowledge to obtain several combined power metering knowledge; finally, determining several power metering knowledge to be verified based on each initial power metering knowledge, each formally transformed power metering knowledge, and each combined power metering knowledge. Specifically, in this embodiment, the sample expansion module can be used to formally transform the initial power metering knowledge, including steps such as synonym replacement and sentence transformation, to obtain several formally transformed power metering knowledge. Then, the formally transformed power metering knowledge can be randomly combined to obtain several combined power metering knowledge. Finally, the expanded power metering knowledge to be verified can be obtained based on the original initial power metering knowledge, the formally transformed power metering knowledge, and the combined power metering knowledge.
[0059] It should be noted that after obtaining the electricity metering knowledge to be verified, an initial sample corpus can be built for training the large model.
[0060] Step S13: Based on the target power metering knowledge graph, filter out each of the power metering knowledge to be verified in the initial sample corpus to obtain several target power metering knowledge, and determine the target sample corpus based on each of the target power metering knowledge.
[0061] It should be noted that the data in the initial sample corpus may contain errors or contradictions. Therefore, a pre-constructed target power metering knowledge graph can be used to verify the power metering knowledge to be verified in the initial sample corpus. The specific process may include: first, determining each target triplet corresponding to the target power metering knowledge graph; then, performing consistency checks between each power metering knowledge to be verified in the initial sample corpus and each target triplet in the target power metering knowledge graph to obtain the corresponding consistency check results; subsequently, based on the consistency check results, identifying power metering knowledge that conflicts with any of the target triplets from the power metering knowledge to be verified and requiring elimination; finally, eliminating the power metering knowledge to be eliminated from the initial sample corpus to obtain a number of target power metering knowledge.
[0062] Specifically, firstly, the target triples corresponding to the current target electricity metering knowledge graph are identified, ensuring that these target triples are accurate information. Next, the consistency of each electricity metering knowledge to be verified in the initial sample corpus is checked against each target triple. If, after verification, it is determined that any of the electricity metering knowledge to be verified conflicts with any target triple—that is, contains errors or contradictions—it can be removed from the initial sample corpus, yielding the remaining target electricity metering knowledge. A target sample corpus is then constructed based on this remaining target electricity metering knowledge.
[0063] Step S14: Train the initial large model using the target power metering knowledge in the target sample corpus to obtain the target power metering large model, so as to provide metering business services for the power industry using the target power metering large model.
[0064] In this embodiment, target electricity metering knowledge from the screened target sample corpus can be introduced into the initial large model. During reinforcement training, the target electricity metering knowledge is learned with emphasis. That is, when new knowledge contradicts existing knowledge, the new knowledge is uniformly selected as the output result; when new knowledge does not contradict existing knowledge but is related to the user's question, the new knowledge is output first or placed in a higher position, thus obtaining the reinforced target electricity metering large model. It should be noted that the initial large model is a deep learning model that has been trained through a large language model and has certain semantic understanding and data retrieval capabilities. It is the direct basis for the reinforcement training of the target electricity metering large model and is mainly responsible for providing basic support capabilities such as natural language recognition and multimodal data fusion for the target electricity metering large model. The target power metering big model is a large model trained with power metering knowledge enhancement. Besides possessing the capabilities of general / applicable big models in areas such as intent understanding, automatic querying, intelligent analysis, and data fusion, it also grasps the general knowledge, rules, legal regulations, and institutional norms of power metering, and is familiar with the reasoning and analysis logic of various typical business scenarios. Therefore, it possesses business analysis and processing capabilities in the power metering field and can be used to provide metering services to the power industry. It is understood that in the construction of the power metering big model in this embodiment, a knowledge embedding model and big model joint training mode were adopted. Data screening and topic enhancement of the target sample corpus were carried out through the power metering knowledge graph. This allows the power metering big model to focus more on knowledge in the power metering field while being less affected by low-quality or erroneous data caused by source errors or corpus expansion biases, thus improving the responsiveness and performance of the power metering big model in the power metering field.
[0065] It should be noted that this embodiment also includes a feedback tracing module to optimize and improve the target power metering knowledge graph and the target power metering large model. In one specific implementation, firstly, an interactive interface corresponding to the target power metering large model is constructed; then, through the interactive interface, the target power metering large model receives target metering business questions from users in the power industry and generates corresponding answer information; subsequently, feedback information generated by the user corresponding to the answer information is obtained, and the target triplet corresponding to the answer information is determined from the target power metering knowledge graph based on the feedback information; if the answer information is determined to be incorrect based on the feedback information, the target triplet corresponding to the answer information is corrected based on the feedback information, and the target power metering knowledge graph is updated based on the corrected target triplet, and the target power metering large model is updated based on the updated target power metering knowledge graph. Specifically, this embodiment can construct an interactive interface corresponding to the target power metering large model. The interactive interface is a graphical interface or operation window for users to input questions and feedback information, and for the target power metering large model to output the thought process and feedback results. Subsequently, the system can receive target metering business questions from users in the power industry through an interactive interface, and generate corresponding answer information using the target power metering big data model. Then, the system can obtain user feedback on the answer information through the interactive interface and input this feedback information into the target power metering big data model. This allows the target power metering big data model to perform reverse tracing based on the feedback information, locating the thought process and logical steps involved in making the judgment, and determining the target triplet corresponding to the answer information from the target power metering knowledge graph. After completing the reverse tracing, for correct answer information, the existing knowledge graph relationships and thought processes will be enhanced; for incorrect answer information, the incorrect target triplet will be corrected based on the feedback information, and the target power metering knowledge graph and the target power metering big data model will be updated simultaneously, thereby achieving the goal of timely error correction and continuous functional improvement. It is understandable that this embodiment fully leverages the interpretability and traceability advantages of the power metering knowledge graph during the use of the large-scale power metering model. It traces potential errors in the thought process or reasoning biases of the large-scale power metering model back to the smallest unit of the "triple". Based on the feedback results, it is enhanced or corrected in a targeted manner. Compared with the "black box" reasoning process of general / general-purpose large models, its error location and correction are faster and more accurate, and the probability of similar errors recurring in the future is lower.
[0066] As can be seen from the above, in this embodiment, firstly, several initial power metering knowledges of the power industry are acquired, and a target power metering knowledge graph is constructed based on each of the initial power metering knowledges; then, the initial power metering knowledges are expanded to obtain several power metering knowledges to be verified, and an initial sample corpus is constructed based on each of the power metering knowledges to be verified; subsequently, based on the target power metering knowledge graph, the power metering knowledges to be verified in the initial sample corpus are filtered to obtain several target power metering knowledges, and a target sample corpus is determined based on each of the target power metering knowledges; finally, the initial large model is trained using the target power metering knowledges in the target sample corpus to obtain a target power metering large model, so as to provide metering business services for the power industry using the target power metering large model. As can be seen from the above, this embodiment first acquires initial power metering knowledge and constructs a target power metering knowledge graph. Then, it expands the initial power metering knowledge to obtain power metering knowledge to be verified, thus constructing an initial sample corpus. Next, it uses the target knowledge graph to filter out the power metering knowledge to be verified, thereby determining the target sample corpus. Finally, it uses the target corpus to train the initial large-scale model to obtain the target power metering large-scale model, which can be used to provide power metering business services. In this way, this embodiment achieves the filtering and conflict disambiguation of power metering knowledge through the target power metering knowledge graph, solving the problem that the training corpus of the general large-scale model focuses on general knowledge but lacks sufficient power metering knowledge, thereby improving the reliability of the target power metering large-scale model. Simultaneously, the target sample corpus obtained through the filtering of the target power metering knowledge graph ensures the professionalism and accuracy of the large-scale model training data, enabling the target power metering large-scale model to possess stronger metering professional analysis and reasoning capabilities, meeting the high accuracy and high reliability requirements of power metering services.
[0067] In one specific implementation, see Figure 2 As shown, the specific process of training a large-scale electricity metering model can be described as follows:
[0068] 1. Collect electricity metering knowledge in advance to form electricity metering knowledge (1);
[0069] 2. Based on electricity metering knowledge (1), extract entities and relationships related to metering knowledge, and use entity association to complete the construction of electricity metering knowledge graph (2) and disambiguation of conflict information;
[0070] 3. Import the knowledge of electricity metering (1) into the sample expansion module (3) to improve the richness and scale of the training corpus of the large-scale electricity metering model. The expanded data is used to construct the initial sample corpus.
[0071] 4. Perform consistency matching between the initial sample corpus and the power metering knowledge graph (2), and remove the samples that are contradictory or erroneous to obtain the target sample corpus (4).
[0072] 5. The screened target sample corpus (4) is introduced into the initial general / general-purpose large model (5), and this part of the knowledge is studied in a focused manner during the reinforcement training process to form a large model of electricity metering (6).
[0073] 6. During the use of the system, users can ask questions and obtain information from the power metering big data model through the interactive interface (7), and provide feedback on the answers.
[0074] 7. The feedback tracing module (8) will collect user feedback information and input the feedback information into the power metering knowledge graph (2) for reverse tracing to locate the thought chain and logical link that made the judgment;
[0075] 8. After completing the reverse tracing, for correct answers, the existing knowledge graph relationships and thought chains will be enhanced; for incorrect answers, the triples that caused the errors will be corrected based on the feedback information, and the power metering knowledge graph (2) and the power metering big model (6) will be updated simultaneously to achieve the goal of timely correction of errors and continuous improvement of functions.
[0076] Accordingly, see Figure 3 As shown in the illustration, this application also provides a large-scale power metering model training device, which may include:
[0077] The knowledge graph construction module 11 is used to acquire several initial power metering knowledge in the power industry and construct a target power metering knowledge graph based on each of the initial power metering knowledge.
[0078] The initial sample corpus construction module 12 is used to expand each of the initial power metering knowledge to obtain several power metering knowledge to be verified, and to construct an initial sample corpus based on each of the power metering knowledge to be verified.
[0079] The target sample corpus determination module 13 is used to filter out each of the power measurement knowledge to be verified in the initial sample corpus based on the target power measurement knowledge graph to obtain a number of target power measurement knowledge, and to determine the target sample corpus based on each of the target power measurement knowledge.
[0080] The power metering large model training module 14 is used to train the initial large model using the target power metering knowledge in the target sample corpus to obtain the target power metering large model, so as to provide metering business services for the power industry using the target power metering large model.
[0081] In some specific embodiments, the knowledge graph construction module 11 may include:
[0082] The power metering knowledge acquisition unit is used to determine the laws, regulations, industry rules, technical standards, and business cases corresponding to the power metering data of the power industry, and to acquire a number of initial power metering knowledge of the power industry based on the laws, regulations, industry rules, technical standards, and business cases.
[0083] In some specific embodiments, the knowledge graph construction module 11 may include:
[0084] The triplet extraction submodule is used to extract several initial triplets corresponding to each of the initial power metering knowledge.
[0085] The relation extraction submodule is used to identify each target entity in each of the initial triples and extract the relationship between each target entity in each of the initial triples to obtain the corresponding relation extraction result;
[0086] The knowledge graph construction submodule is used to take each of the target entities as nodes and establish connections between nodes corresponding to different target entities with related relationships based on the relationship extraction results, so as to construct an initial power metering knowledge graph corresponding to the initial power metering knowledge.
[0087] The knowledge graph determination submodule is used to determine the target power metering knowledge graph based on the initial power metering knowledge graph.
[0088] In some specific implementations, the knowledge graph determination submodule may include:
[0089] The node disambiguation unit is used to identify conflicting nodes in the initial power metering knowledge graph and disambiguate the conflicting nodes to obtain adjusted nodes.
[0090] The knowledge graph determination unit is used to determine the target power metering knowledge graph based on the adjusted nodes in the initial power metering knowledge graph.
[0091] In some specific embodiments, the initial sample corpus construction module 12 may include:
[0092] The form conversion unit is used to perform form conversion on each of the initial power metering knowledge to obtain several form-converted power metering knowledge.
[0093] The knowledge combination unit is used to randomly combine the electricity metering knowledge transformed from each of the above forms to obtain several combined electricity metering knowledge.
[0094] The knowledge to be verified determination unit is used to determine a number of the power metering knowledge to be verified based on each of the initial power metering knowledge, each of the power metering knowledge after transformation, and each of the combined power metering knowledge.
[0095] In some specific embodiments, the target sample corpus determination module 13 may include:
[0096] The target triplet determination unit is used to determine each target triplet corresponding to the target power metering knowledge graph.
[0097] The consistency verification unit is used to perform consistency verification between each of the power metering knowledge to be verified in the initial sample corpus and each of the target triples in the target power metering knowledge graph, and obtain the corresponding consistency verification results.
[0098] The knowledge to be screened out determination unit is used to determine, based on the consistency verification result, the power metering knowledge to be screened out that conflicts with any of the target triplet from each of the power metering knowledge to be verified;
[0099] The target knowledge determination unit is used to filter out the electricity metering knowledge to be removed from the initial sample corpus to obtain a number of target electricity metering knowledge.
[0100] In some specific embodiments, the large-scale power metering model training device may further include:
[0101] An interactive interface construction unit is used to construct the interactive interface corresponding to the target power metering large model.
[0102] The solution information generation unit is used to receive target metering business questions from users in the power industry through the interactive interface and the target power metering big model, and generate corresponding solution information.
[0103] The feedback information acquisition unit is used to acquire feedback information generated by the user that corresponds to the answer information, and to determine the target triplet corresponding to the answer information from the target power metering knowledge graph based on the feedback information.
[0104] The large model update unit is used to correct the target triplet corresponding to the solution information based on the feedback information if it is determined that there is an error in the solution information based on the feedback information, so as to update the target power metering knowledge graph based on the corrected target triplet, and update the target power metering large model based on the updated target power metering knowledge graph.
[0105] Furthermore, embodiments of this application also disclose an electronic device, Figure 4This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the power metering large model training method disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.
[0106] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0107] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.
[0108] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the large-scale power metering model training method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.
[0109] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed method for training a large-scale electricity metering model. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.
[0110] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0111] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0112] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0113] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0114] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for training a large-scale electricity metering model, characterized in that, include: Acquire some initial power metering knowledge from the power industry, and construct a target power metering knowledge graph based on each of the initial power metering knowledge. The initial power metering knowledge is expanded to obtain several power metering knowledge to be verified, and an initial sample corpus is constructed based on the power metering knowledge to be verified. Based on the target power metering knowledge graph, the power metering knowledge to be verified in the initial sample corpus is screened out to obtain a number of target power metering knowledge, and the target sample corpus is determined based on each of the target power metering knowledge. The initial large model is trained using the target power metering knowledge in the target sample corpus to obtain the target power metering large model, which is then used to provide metering services for the power industry.
2. The power metering large-scale model training method according to claim 1, characterized in that, The acquisition of certain initial power metering knowledge in the power industry includes: Identify the laws, regulations, industry rules, technical standards, and business cases corresponding to the power metering data of the power industry, and based on the laws, regulations, industry rules, technical standards, and business cases, acquire some of the initial power metering knowledge of the power industry.
3. The power metering large-scale model training method according to claim 1, characterized in that, The construction of the target power metering knowledge graph based on the initial power metering knowledge includes: Extract several initial triplets corresponding to each of the initial power metering knowledge; Identify each target entity in each of the initial triples and extract the relationships between each target entity in each of the initial triples to obtain the corresponding relationship extraction results; Each of the target entities is used as a node, and a connection is established between the nodes corresponding to different target entities with related relationships based on the relationship extraction results, so as to construct an initial power metering knowledge graph corresponding to the initial power metering knowledge. The target power metering knowledge graph is determined based on the initial power metering knowledge graph.
4. The power metering large model training method according to claim 3, characterized in that, The process of determining the target power metering knowledge graph based on the initial power metering knowledge graph includes: Identify the conflicting nodes in the initial power metering knowledge graph and disambiguate the conflicting nodes to obtain the adjusted nodes. The target power metering knowledge graph is determined based on the adjusted nodes in the initial power metering knowledge graph.
5. The power metering large-scale model training method according to claim 1, characterized in that, The expansion of the initial power metering knowledge yields several power metering knowledge sets to be verified, including: The initial power metering knowledge described above is formally transformed to obtain several formally transformed power metering knowledge; The transformed electricity metering knowledge in each of the aforementioned forms is randomly combined to obtain several combined electricity metering knowledge; Based on the initial power metering knowledge, the power metering knowledge after transformation, and the combined power metering knowledge, several power metering knowledge to be verified are determined.
6. The method for training a large-scale power metering model according to claim 1, characterized in that, The process of filtering out the unverified power metering knowledge in the initial sample corpus based on the target power metering knowledge graph yields several target power metering knowledge items, including: Determine each target triplet corresponding to the target power metering knowledge graph; The consistency verification results are obtained by comparing each of the power metering knowledge to be verified in the initial sample corpus with each of the target triples in the target power metering knowledge graph. Based on the consistency verification results, power metering knowledge that conflicts with any of the target triples is identified from each of the power metering knowledge to be verified and is to be screened out. The electricity metering knowledge to be screened out is removed from the initial sample corpus to obtain several target electricity metering knowledge.
7. The method for training a large-scale electricity metering model according to any one of claims 1 to 6, characterized in that, Also includes: Construct the interactive interface corresponding to the target large-scale electricity metering model; Through the interactive interface, the target power metering big model is used to receive target metering business questions from users in the power industry and generate corresponding answer information. Obtain feedback information generated by the user that corresponds to the answer information, and determine the target triplet corresponding to the answer information from the target power metering knowledge graph based on the feedback information; If it is determined based on the feedback information that the solution information contains an error, then the target triple corresponding to the solution information is corrected based on the feedback information, the target power metering knowledge graph is updated based on the corrected target triple, and the target power metering large model is updated based on the updated target power metering knowledge graph.
8. A large-scale power metering model training device, characterized in that, include: The knowledge graph construction module is used to acquire several initial power metering knowledge in the power industry and construct a target power metering knowledge graph based on each of the initial power metering knowledge. The initial sample corpus construction module is used to expand the initial power metering knowledge to obtain several power metering knowledge to be verified, and to construct the initial sample corpus based on the power metering knowledge to be verified. The target sample corpus determination module is used to filter out each of the power measurement knowledge to be verified in the initial sample corpus based on the target power measurement knowledge graph to obtain a number of target power measurement knowledge, and to determine the target sample corpus based on each of the target power measurement knowledge. The power metering large model training module is used to train the initial large model using the target power metering knowledge in the target sample corpus to obtain the target power metering large model, so as to provide metering business services for the power industry using the target power metering large model.
9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the power metering large model training method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the large-scale power metering model training method as described in any one of claims 1 to 7.