CN-122001075-A - Large-model-based power grid monitoring alarm event handling method and system

CN122001075ACN 122001075 ACN122001075 ACN 122001075ACN-122001075-A

Abstract

The invention discloses a method and a system for handling a power grid monitoring alarm event based on a large model, which relate to the technical field of power systems and comprise the steps of obtaining a power grid monitoring alarm event history sample and preprocessing; the method comprises the steps of constructing a balance training sample set based on a sample enhancement method combining a first large language model and resampling, utilizing the balance training sample set to conduct fine tuning training based on a pre-training language model to construct a power grid monitoring alarm event diagnosis model, constructing a power grid monitoring alarm event disposal model based on a second large language model, enabling the disposal model to be used for combining event type diagnosis results output by the diagnosis model, retrieving reference information from a historical disposal knowledge base, generating auxiliary disposal suggestions for input alarm event texts, and integrating the diagnosis model and the disposal model to form a power grid monitoring alarm event disposal framework. The intelligent diagnosis and auxiliary treatment of the power grid monitoring alarm are realized, and the diagnosis accuracy and the treatment efficiency are improved.

Inventors

CONG LEYAO
XU QIANG
WANG ZHENG
SU HE
WANG JIANING
MIN JIE
CAI YIJUN
YANG YEFEI

Assignees

国网江苏省电力有限公司无锡供电分公司

Dates

Publication Date: 20260508
Application Date: 20251211
Priority Date: 20250930

Claims (10)

1. A method for handling a grid monitoring alarm event based on a large model, comprising: Acquiring a power grid monitoring alarm event history sample and preprocessing; Based on a sample enhancement method combining a first large language model and resampling, processing the preprocessed historical samples to construct a balance training sample set; Based on a pre-training language model, performing fine tuning training by using the balance training sample set, constructing a power grid monitoring alarm event diagnosis model, and outputting an event type diagnosis result according to an input alarm event text; based on a second large language model, constructing a power grid monitoring alarm event handling model, wherein the power grid monitoring alarm event handling model is used for combining event type diagnosis results output by the power grid monitoring alarm event diagnosis model, retrieving reference information from a historical handling knowledge base and generating auxiliary handling suggestions for input alarm event texts; wherein the first large language model and the second large language model adopt the same or different large language models.
2. A method for handling grid monitoring alarm events based on a large model as claimed in claim 1, wherein: The constructing a balance training sample set includes: Determining a reference number according to the distribution of the number of various events in the preprocessed historical samples, and setting a target number interval after the enhancement of each type of event samples by taking the reference number as a reference; identifying a minority class of events with the number of samples lower than the lower limit of the target number interval and a majority class of events with the number of samples higher than the upper limit of the target number interval; generating an extended sample based on an original historical sample of the class of events by using a first large language model for each class of events in the minority class of events; After generating the expanded samples, for each minority event, if the current total sample number of each minority event still does not reach the lower limit of the target number interval, randomly copying the current samples in the minority event to enable the sample number of each minority event to reach the lower limit of the target number interval; For most events, the number of the samples is reduced to be within the target number interval by randomly deleting part of the samples.
3. A method for handling grid monitoring alarm events based on a large model as claimed in claim 2, wherein: The setting of the target number interval after the enhancement of each type of event sample comprises the following steps: calculating the ratio of the number of samples of various events to the maximum number of samples; selecting a preset proportion of the maximum sample number as a reference number according to the concentrated distribution condition of the ratio; A target number interval is determined based on the reference number.
4. A method for handling grid monitoring alarm events based on a large model as claimed in claim 2, wherein: The generating the extended samples includes: Constructing a structured prompt word, wherein the structured prompt word comprises an information class element for explaining the background of a data enhancement task, a structure class element for prescribing that an output sample is required to be consistent with the original format and key information, and a control class element for indicating the generation of multiple groups of similar samples; the original historical samples of the few types of events are input into a first large language model, and the first large language model is guided to generate an expansion sample based on the structured prompt words.
5. A method for handling grid monitoring alarm events based on a large model as claimed in claim 1, wherein: the construction of the power grid monitoring alarm event diagnosis model comprises the following steps: performing word segmentation and stop word removal processing on the alarm event text in the balance training sample set to form a model input sequence; dividing the model input sequence into a training set, a verification set and a test set, and inputting the training set, the verification set and the test set into a pre-training language model for fine-tuning training; And optimizing parameters of the pre-training language model through fine tuning training to obtain the power grid monitoring alarm event diagnosis model.
6. A method for handling grid monitoring alarm events based on a large model as claimed in claim 1, wherein: the constructing the power grid monitoring alarm event handling model based on the second large language model comprises the following steps: the method comprises the steps of collecting and arranging historical alarm event handling cases, classifying and storing according to event types, and structuring each case into an information unit containing alarm event description, reporting scheduling content and handling process; Vectorizing text content in a history handling knowledge base and storing the vectorized text content in a vector database; Configuring a global prompt word, wherein the global prompt word is used for defining that a power grid monitoring alarm event disposal model needs to search related disposal cases from the vector database based on input event types, and generating auxiliary disposal suggestions comprising specific equipment actions and disposal steps by combining current input alarm event texts; And integrating the vector database and the global prompt word into a second large language model to construct a power grid monitoring alarm event handling model.
7. A method for handling grid monitoring alarm events based on a large model as claimed in claim 1, wherein: The operation flow of the treatment frame comprises the following steps: Receiving a power grid monitoring alarm event text; calling a power grid monitoring alarm event diagnosis model to diagnose a power grid monitoring alarm event text to obtain an event type; inputting the text of the power grid monitoring alarm event and the diagnosed event type into a power grid monitoring alarm event disposal model; the power grid monitoring alarm event disposal model retrieves relevant disposal case information from a historical disposal knowledge base according to the event type, and generates auxiliary disposal suggestions by combining the power grid monitoring alarm event text; The event type and corresponding auxiliary treatment advice are output.
8. A large model based grid monitoring alarm event handling system for implementing a large model based grid monitoring alarm event handling method as set forth in any of claims 1-7, comprising: the system comprises a data preprocessing and enhancing module, an alarm event diagnosis module, an alarm event handling module and a system integration module, wherein: The data preprocessing and enhancing module is used for acquiring and preprocessing a historical sample of the power grid monitoring alarm event, and processing the preprocessed historical sample based on a sample enhancing method combining the first large language model and resampling to construct a balance training sample set; the alarm event diagnosis module is internally provided with a pre-training language model which is obtained based on the balance training sample set through fine tuning training and is used for receiving alarm event text and outputting event types after diagnosis; The alarm event handling module is internally provided with a second large language model which integrates the configuration of a history handling knowledge base and a global prompt word, and is used for receiving alarm texts and event types thereof, and retrieving reference information from the history handling knowledge base by combining the event types output by the diagnosis module to generate auxiliary handling suggestions; The system integration module is used for integrating and scheduling the diagnosis module and the treatment module, receiving input power grid monitoring alarm event text, sequentially calling the diagnosis module and the treatment module for processing, and outputting a final event type and a corresponding auxiliary treatment suggestion.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the computer program when loaded to the processor implements a large model based grid monitoring alarm event handling method according to any of claims 1-7.
10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements a large model based grid monitoring alarm event handling method according to any of claims 1-7.

Description

Large-model-based power grid monitoring alarm event handling method and system Technical Field The invention relates to the technical field of power systems, in particular to a method and a system for handling a power grid monitoring alarm event based on a large model. Background When the monitoring alarm event occurs to the power grid, the power operation and maintenance center can receive a large number of monitoring alarm signals at the same time, and the monitoring alarm signals form a power grid monitoring alarm event together. The power operation staff usually needs to judge specific power grid monitoring alarm event types according to the alarm signals, so that effective monitoring alarm event treatment is carried out, fault elements are isolated, and the stable operation of the power grid is recovered. The traditional diagnosis and treatment of monitoring alarm events depend on the knowledge and long-term working experience of power grid operation and maintenance personnel, and related manuals are required to be consulted if necessary, so that the diagnosis and treatment efficiency is low, and misjudgment, missed judgment and the like are easy to occur. The unbalance of the power grid monitoring alarm event samples also restricts the accuracy of the artificial intelligent diagnosis method, so that the model can obtain a better result by evaluating indexes on the whole, but in particular, on each event type, the event type evaluation accuracy with fewer part of samples possibly exists is lower. Common data enhancement methods include simple data enhancement methods, resampling methods, and the like. However, the samples added by the method have higher similarity with the original samples, and are easy to copy a large amount of sample noise, so that the model training result is influenced. In the aspect of power grid monitoring alarm event handling, the current power grid monitoring alarm handling still depends on the manual experience of operation and maintenance personnel and related instruction manuals, and the existing research mainly provides auxiliary decisions for power grid monitoring alarm event handling by constructing a power grid fault handling knowledge graph. However, knowledge graph construction requires knowledge extraction, knowledge connection and other works on power grid topology and other data, and the construction process is complicated. In the prior art, an alarm research and judgment treatment scheme based on a large model exists, and the alarm research and judgment treatment scheme generally drives the large safety model to generate a corresponding treatment decision tree by collecting alarm information and constructing analysis prompt words so as to guide response. However, when the general scheme is applied to a power grid monitoring scene, the general scheme still has obvious limitations that firstly, the advanced adaptation of the knowledge, terms and business processes in the power system professional field is lacking, the expertise and accuracy of a model are not enough, secondly, the problem of unbalanced categories of alarm event samples is not considered, an effective data enhancement mechanism is lacking to optimize a model training basis, thirdly, the whole process of research, judgment and treatment is processed by adopting a single model, and a cooperative mechanism is not constructed to give consideration to the diagnosis precision and the efficiency of invoking the treatment knowledge. Therefore, how to realize the intelligent alarm treatment with high precision, high efficiency and close to the operation and maintenance requirements of the power grid is still a technical problem to be solved. Disclosure of Invention In order to solve the defects existing in the prior art, the invention provides a large-model-based power grid monitoring alarm event handling method and a large-model-based power grid monitoring alarm event handling system. The invention adopts the following technical scheme. The first aspect of the invention provides a large-model-based power grid monitoring alarm event handling method, which comprises the following steps: Acquiring a power grid monitoring alarm event history sample and preprocessing; Based on a sample enhancement method combining a first large language model and resampling, processing the preprocessed historical samples to construct a balance training sample set; Based on a pre-training language model, performing fine tuning training by using the balance training sample set, constructing a power grid monitoring alarm event diagnosis model, and outputting an event type diagnosis result according to an input alarm event text; based on a second large language model, constructing a power grid monitoring alarm event handling model, wherein the power grid monitoring alarm event handling model is used for combining event type diagnosis results output by the power grid monitoring alarm event diagnosis model, retrieving reference informat