A method for supervising illegal behavior of power grid, terminal equipment and storage medium

Abstract

The present invention relates to a method for supervising power grid violations, a terminal device and a storage medium, the method comprising: constructing a word vector model, and semantically aligning the output result of the word vector model for a problem with the output result of the detailed rules corresponding to the problem; extracting the accident type and stage type corresponding to each detailed rule to construct a graph; encoding each node in the graph through the word vector model; combining the adjacency matrix of the graph and the encoding results of each node, re-encoding the detailed rule content nodes through a graph neural network; predicting the stage type and accident type of the input problem through a large model, screening the corresponding detailed rules from a standard file, and obtaining the re-encoding results in the graph corresponding to the screening detailed rules, calculating the similarity between the word vector encoding result of the input problem and the re-encoding results of each screening detailed rule, and outputting the N detailed rules with the highest similarity. The present invention solves the defects of the existing manual operation mode and improves the overall performance and reliability of power grid supervision.

Description

Technical Field

[0001] The present invention relates to the field of power grid supervision, and in particular to a power grid violation supervision method, terminal equipment and storage medium. Background Art

[0002] At present, in power grid supervision, the identification and matching of violations mainly rely on manual operations. In the case of insufficient power grid automation, this method mainly has the following problems: 1. Inefficiency: The manual matching process is usually time-consuming and inefficient, especially when the data volume is large and the events are complex. The processing time is long, resulting in slow supervision response. 2. Human resource limitations: It is highly dependent on experienced professionals for data analysis and judgment, and faces the problem of insufficient number of experts. Especially in the case of the need to quickly expand supervision capabilities, human resource limitations become a significant bottleneck. 3. Consistency and accuracy issues: The results of manual analysis are easily affected by personal experience and subjective judgment, and may miss or misjudge certain violations. Although existing technologies provide basic functional support in the field of power grid technical supervision, there are still considerable challenges in terms of efficiency, resource utilization and accuracy. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a method for supervising power grid violation behavior, a terminal device and a storage medium.

[0004] The specific plan is as follows:

[0005] A method for supervising power grid violations comprises the following steps:

[0006] S1: Collect the corresponding violations of each rule in the standard document for supervision of power grid violations;

[0007] S2: Construct a training set based on each rule and the corresponding question. The training set includes positive examples and negative examples. The positive example is that the question and the rule are in a corresponding relationship, and the negative example is that the question and the rule are not in a corresponding relationship.

[0008] S3: Build a word vector model and train the word vector model using the training set so that the output result of the word vector model for the question is semantically aligned with the output result of the details corresponding to the question;

[0009] S4: extract the accident types and stage types corresponding to each detail in the standard document to build a graph, which includes three types of nodes: stage type node, accident type node and detail content node;

[0010] S5: Encode the content of each node in the graph through the trained word vector model to obtain the encoding result of each node;

[0011] S6: Combining the adjacency matrix of the graph and the encoding results of each node, re-encode the rule content node through the graph neural network to obtain the feature vector of the rule content node;

[0012] S7: Build a large model for predicting the stage type and accident type corresponding to the input problem;

[0013] S8: When an input question is received, the phase type and accident type corresponding to the input question are first predicted through the large model; then, the details with the same phase type and accident type are filtered out from the standard file, and the feature vectors of each filtered detail are extracted based on the feature vectors of the content nodes of each detail in the figure; then, the input question is encoded using the trained word vector model; finally, the similarity between the encoded result of the input question and the feature vectors of each filtered detail is calculated respectively, and the N details with the highest similarity are output.

[0014] Furthermore, the corresponding violations of the rules are collected through the deepseek-67B-chat model.

[0015] Furthermore, the word vector model adopts text2vec-base-chinese.

[0016] Furthermore, the graph neural network needs to be trained in advance before use, so that the final feature vector after the detailed content node of the trained word vector model is encoded by the graph neural network is consistent with the output result obtained by passing the content of the input question through the trained word vector model.

[0017] Furthermore, when building a large model, the prompt for the large model's stage type prediction is set to: "Based on the input information, the output scenario belongs to which stage. The stage types are {design stage, infrastructure stage, operation stage}; select a result from the above brackets to answer; answer example: operation stage; the following is the input information:".

[0018] Furthermore, when constructing the large model, the prompt for the large model's prediction of accident types is set as: "Based on the input information, the output scenario is required to be of what type of accident. The accident types are {GIS, switchgear accident, transformer damage accident, personal injury accident,..., transmission line accident}; select a result from the above brackets to answer; answer example: power cable damage accident; the following is the input information:".

[0019] Furthermore, the large model uses internlm2-chat-7b.

[0020] A power grid violation monitoring terminal device comprises a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method described above in the embodiment of the present invention are implemented.

[0021] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described above in an embodiment of the present invention are implemented.

[0022] The present invention adopts the above technical solution to solve the defects of the existing manual operation mode and improve the overall performance and reliability of power grid supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Shown is a flow chart of a method according to a first embodiment of the present invention. DETAILED DESCRIPTION

[0024] To further illustrate various embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, which are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementations and advantages of the present invention.

[0025] The present invention will now be further described with reference to the accompanying drawings and specific implementation methods.

[0026] Embodiment 1:

[0027] The embodiment of the present invention provides a method for monitoring illegal behaviors of a power grid, such as Figure 1 As shown, the method comprises the following steps:

[0028] S1: Collect the corresponding violations of each rule (i.e., regulations) in the standard document for supervision of grid violations.

[0029] In the standard documents on power grid safety supervision, the corresponding details are located in the third level of the directory. The first three levels are: accident type, equipment type and stage type (such as operation stage, maintenance stage).

[0030] In order to facilitate the collection of problems in this embodiment, it is preferred to generate actual problems that violate various rules through the deepseek-67B-chat model. In other embodiments, the problems can also be collected through historical violation cases and expert knowledge, which is not limited here.

[0031] S2: Construct a training set based on each rule and the corresponding question. The training set includes positive examples and negative examples. The positive example is that the question and the rule are in correspondence, and the negative example is that the question and the rule are not in correspondence.

[0032] S3: Build a word vector model and train the word vector model through the training set so that the output results of the word vector model for the question are semantically aligned with the output results of the details corresponding to the question.

[0033] In this embodiment, the word vector model adopts text2vec-base-chinese.

[0034] S4: Extract the accident types and stage types corresponding to each detail in the standard document to construct a graph, which includes three types of nodes: stage type nodes, accident type nodes, and detail content nodes.

[0035] In order to facilitate the construction of the graph, in this embodiment, a search directory corresponding to each rule is constructed in advance according to the directory of the standard file. The search directory contains fields corresponding to the contents of the above three types of nodes, namely, the type of accident, the type of stage and the content of the rules. Other fields corresponding to the description can also be added thereon, and the fields are arranged in a certain order, such as: "System stability destruction--power supply--infrastructure stage--for situations with special stability requirements such as point-to-network and long-distance transmission of large power sources, special research on the impact of the excitation system on the power grid should be carried out, and the research results are used to guide the selection of the excitation system."

[0036] S5: Encode the content of each node in the graph through the trained word vector model to obtain the encoding result of each node.

[0037] In this step, it is necessary to obtain the encoding results of all nodes in the graph (i.e., three types of nodes).

[0038] S6: Combining the adjacency matrix of the graph and the encoding results of each node, the detail content node is re-encoded through the graph neural network to obtain the feature vector of the detail content node.

[0039] The graph neural network needs to be trained in advance so that the feature vector of the re-encoded detail content node is consistent with the output result obtained by the trained word vector model for the detail content of the detail content node.

[0040] S7: Build a large model for predicting the stage type and accident type corresponding to the input problem.

[0041] In this embodiment, the large model uses internlm2-chat-7b. When constructing the large model, the prompt for the large model prediction for the stage type is set as: "Based on the input information, the output scenario belongs to which stage, the stage type is {design stage, infrastructure stage, operation stage}; select a result from the above brackets to answer; answer example: operation stage; the following is the input information:". The prompt for the large model prediction for the accident type is set as: "Based on the input information, the output scenario belongs to which accident type, the accident type is {GIS, switchgear accident, transformer damage accident, personal injury accident,..., transmission line accident}; select a result from the above brackets to answer; answer example: power cable damage accident; the following is the input information:".

[0042] S8: When an input question is received, the phase type and accident type corresponding to the input question are first predicted through the large model; then, the details with the same phase type and accident type are filtered out from the standard file, and the feature vectors of each filtered detail are extracted based on the feature vectors of the content nodes of each detail in the figure; then, the input question is encoded using the trained word vector model; finally, the similarity between the encoded result of the input question and the feature vectors of each filtered detail is calculated respectively, and the N details with the highest similarity are output.

[0043] The similarity calculation algorithm may adopt a conventional algorithm, and in this embodiment, a pre-phase velocity algorithm is adopted.

[0044] The size of N can be set by those skilled in the art as needed, and is set to 5 in this embodiment.

[0045] The embodiment of the present invention uses a large language model combined with a graph neural network model and a word vector model to achieve automatic matching of on-site inspection issues and violation regulations, which is different from the traditional manual matching method in the following aspects:

[0046] 1. Model framework: It uses advanced neural network architecture to process large-scale data sets, while traditional manual methods cannot handle such large and complex data structures.

[0047] 2. Automated processing: Automatically identify and match inspection questions with relevant violation regulations by training large language models. This method uses machine learning algorithms and a large amount of training data to enable the model to learn to understand and parse the deep semantic connection between inspection text and regulations. Automated models can process and update large amounts of data in real time, while manual methods are limited in data processing speed and real-time updates.

[0048] 3. Semantic understanding ability: Compared with manual matching, large models can more deeply understand the implicit meaning and complex context in the text, thereby improving the accuracy of matching. Manual matching often relies on superficial keyword matching or intuitive judgment, which can easily miss or misjudge complex semantic relationships.

[0049] Embodiment 2:

[0050] The present invention also provides a power grid violation supervision terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps in the above method embodiment of the first embodiment of the present invention when executing the computer program.

[0051] Further, as an executable solution, the power grid violation monitoring terminal device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The power grid violation monitoring terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the composition structure of the above-mentioned power grid violation monitoring terminal device is only an example of the power grid violation monitoring terminal device, and does not constitute a limitation on the power grid violation monitoring terminal device. It can include more or less components than the above-mentioned, or combine certain components, or different components. For example, the power grid violation monitoring terminal device can also include input and output devices, network access devices, buses, etc., and the embodiments of the present invention do not limit this.

[0052] Further, as an executable solution, the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the power grid violation monitoring terminal device, and uses various interfaces and lines to connect various parts of the entire power grid violation monitoring terminal device.

[0053] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the power grid violation supervision terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required for a function; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0054] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method in the embodiment of the present invention are implemented.

[0055] If the module / unit integrated in the power grid violation supervision terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory) and software distribution medium, etc.

[0056] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, it should be understood by those skilled in the art that various changes may be made to the present invention in form and details without departing from the spirit and scope of the present invention as defined by the appended claims, all of which are within the scope of protection of the present invention.

Claims

1. A method for supervising illegal behaviors of a power grid, characterized in that: The following steps are involved: S1: Collect the corresponding violations of each rule in the standard document for supervision of power grid violations; S2: Construct a training set based on each rule and the corresponding question. The training set includes positive examples and negative examples. The positive example is that the question and the rule are in a corresponding relationship, and the negative example is that the question and the rule are not in a corresponding relationship. S3: Build a word vector model and train the word vector model using the training set so that the output result of the word vector model for the question is semantically aligned with the output result of the details corresponding to the question; S4: extract the accident types and stage types corresponding to each detail in the standard document to build a graph, which includes three types of nodes: stage type node, accident type node and detail content node; S5: Encode the content of each node in the graph through the trained word vector model to obtain the encoding result of each node; S6: Combining the adjacency matrix of the graph and the encoding results of each node, re-encode the rule content node through the graph neural network to obtain the feature vector of the rule content node; S7: Build a large model for predicting the stage type and accident type corresponding to the input problem; S8: When an input question is received, the phase type and accident type corresponding to the input question are first predicted through the large model; then, the details with the same phase type and accident type are filtered out from the standard file, and the feature vectors of each filtered detail are extracted based on the feature vectors of the content nodes of each detail in the figure; then, the input question is encoded using the trained word vector model; finally, the similarity between the encoded result of the input question and the feature vectors of each filtered detail is calculated respectively, and the N details with the highest similarity are output.

2. The method for supervising power grid violations according to claim 1, characterized in that: The corresponding violations of the rules are collected through the deepseek-67B-chat model.

3. The method for supervising power grid violations according to claim 1, characterized in that: The word vector model uses text2vec-base-chinese.

4. The method for monitoring power grid violations according to claim 1, characterized in that: The graph neural network needs to be trained in advance before use, so that the final feature vector after the detailed content node of the trained word vector model is encoded by the graph neural network is consistent with the output result obtained by the trained word vector model for the content of the input question.

5. The method for supervising power grid violations according to claim 1, characterized in that: When building a large model, set the prompt for the large model's stage type prediction to: "Based on the input information, the output scenario belongs to which stage. The stage types are {design stage, infrastructure stage, operation stage}; select a result from the above brackets to answer; answer example: operation stage; the following is the input information:".

6. The method for monitoring power grid violations according to claim 1, characterized in that: When building a large model, set the prompt for the large model's prediction of accident types as: "Based on the input information, the output scenario is required to be of what type of accident. The accident types are {GIS, switchgear accident, transformer damage accident, personal injury accident, ..., transmission line accident}; select a result from the above brackets to answer; answer example: power cable damage accident; the following is the input information:".

7. The method for monitoring power grid violations according to claim 1, characterized in that: The large model uses internlm2-chat-7b.

8. A power grid violation monitoring terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 7 when executing the computer program.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method as claimed in any one of claims 1 to 7 are implemented.

Images