Alarm processing knowledge base construction method based on large model and alarm processing method

Abstract

The invention provides an alarm processing knowledge base construction method based on a large model and an alarm processing method, and relates to the technical field of power grid information processing. The alarm processing knowledge base construction method comprises the steps of obtaining alarm data of a power grid system; according to the alarm data and a pre-trained knowledge extraction model, first attribute information corresponding to the alarm data is determined, and the knowledge extraction model is constructed based on a large model; and constructing an alarm processing knowledge base of the power grid system according to the first attribute information. According to the method, the strong learning ability and generalization ability of a large model are utilized, the attribute information corresponding to the alarm data of the power grid system is extracted, in-depth mining and reasoning of potential rules and features in the alarm data are achieved, the alarm processing knowledge base is constructed based on the attribute information, the accuracy and coverage of the alarm processing knowledge base are improved, and in addition, the method has good application prospects. The knowledge base construction efficiency is improved through an automatic mode, effective and reliable decision support is provided for operation and maintenance personnel, and the reliability and stability of power grid system operation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid information processing, and in particular to an alarm processing knowledge base construction method based on a large model and an alarm processing method. BACKGROUND

[0002] In the operation process of many complex systems, such as industrial production, network operation and maintenance, and power grid systems, the generation of alarms is a common phenomenon. Alarms often contain key information about the system operation state, and accurate monitoring and efficient processing of alarms are crucial to ensuring stable system operation. With the emergence of the concept of alarm processing knowledge base, it can provide timely and accurate guidance for operation and maintenance personnel to handle alarms generated in daily production activities, greatly improving the work efficiency of operation and maintenance personnel. Therefore, how to effectively construct the alarm processing knowledge base becomes a key link.

[0003] In related technologies, different alarm processing suggestions are collected and sorted by artificial means, and the alarms and their processing suggestions are stored correspondingly to form an alarm processing knowledge base. However, there are problems such as low construction efficiency, incompleteness, and low accuracy. SUMMARY

[0004] The present application provides an alarm processing knowledge base construction method based on a large model and an alarm processing method to solve the problems of low construction efficiency, incomplete knowledge base, and low accuracy in related technologies.

[0005] In a first aspect, the present application provides an alarm processing knowledge base construction method based on a large model, comprising:

[0006] obtaining alarm data of a power grid system;

[0007] determining first attribute information corresponding to the alarm data according to the alarm data and a pre-trained knowledge extraction model, the first attribute information including alarm categories, alarm causes, impact ranges and processing suggestions, and the knowledge extraction model being constructed based on a large model;

[0008] constructing an alarm processing knowledge base of the power grid system according to the first attribute information.

[0009] In a possible implementation, constructing the alarm processing knowledge base of the power grid system according to the first attribute information comprises:

[0010] integrating the first attribute information using a graph retrieval-augmented generation (GraphRAG) technology to obtain an alarm processing vector corresponding to the alarm data;

[0011] constructing the alarm processing knowledge base of the power grid system according to the alarm processing vector.

[0012] In a possible implementation, the knowledge extraction model is obtained by the following manner:

[0013] obtain historical alarm data of a power grid system and auxiliary data corresponding to the historical alarm data;

[0014] perform preprocessing on the historical alarm data and the auxiliary data to obtain processed alarm data, the preprocessing including data cleaning and normalization;

[0015] perform iterative training on a large model according to the processed alarm data and a preset expert rule to obtain the knowledge extraction model, the large model including a Transformer model.

[0016] In a possible implementation, the method for constructing the alarm processing knowledge base further includes:

[0017] extract target attribute information corresponding to target alarm data from the knowledge extraction model, the target alarm data being a set of alarm data that has not occurred in the power grid system;

[0018] update the alarm processing knowledge base according to the target attribute information.

[0019] In a second aspect, the application provides an alarm processing method, including:

[0020] obtain real-time alarm data of a power grid system;

[0021] perform vector conversion on the real-time alarm data to obtain a real-time alarm vector;

[0022] perform vector matching on the real-time alarm vector in an alarm processing knowledge base, the alarm processing knowledge base being constructed by the method of the first aspect;

[0023] if a target vector corresponding to the real-time alarm vector exists in the alarm processing knowledge base, determine and output an alarm processing suggestion corresponding to the real-time alarm data according to the target vector, to guide relevant personnel to perform alarm processing based on the alarm processing suggestion.

[0024] In a possible implementation, the alarm processing method further includes: if a target vector corresponding to the real-time alarm vector does not exist in the alarm processing knowledge base, extract second attribute information corresponding to the real-time alarm data based on the knowledge extraction model; and determine a processing suggestion contained in the second attribute information as the alarm processing suggestion corresponding to the real-time alarm data.

[0025] In a third aspect, the application provides a device for constructing an alarm processing knowledge base based on a large model, including:

[0026] an obtaining module configured to obtain alarm data of a power grid system;

[0027] determining a first attribute information corresponding to the alarm data according to the alarm data and a pre-trained knowledge extraction model, the first attribute information including an alarm category, an alarm cause, an influence range and a processing suggestion, and the knowledge extraction model being constructed based on a large model;

[0028] constructing a power grid system alarm processing knowledge base according to the first attribute information.

[0029] In a possible implementation, the constructing module is specifically configured to: integrate the first attribute information by using a GraphRAG technology to obtain an alarm processing vector corresponding to the alarm data; and construct the power grid system alarm processing knowledge base according to the alarm processing vector.

[0030] In a possible implementation, the knowledge extraction model is obtained by: obtaining historical alarm data of the power grid system and auxiliary data corresponding to the historical alarm data; pre-processing the historical alarm data and the auxiliary data to obtain processed alarm data, the pre-processing including data cleaning and normalization; and iteratively training a large model according to the processed alarm data and a preset expert rule to obtain the knowledge extraction model, the large model including a Transformer model.

[0031] In a possible implementation, the power grid system alarm processing knowledge base construction apparatus based on the large model further includes an updating module, the updating module being configured to: extract target attribute information corresponding to target alarm data according to the knowledge extraction model, the target alarm data being a set of alarm data that has not occurred in the power grid system; and update the alarm processing knowledge base according to the target attribute information.

[0032] In a fourth aspect, the present application provides an alarm processing apparatus, including:

[0033] an obtaining module configured to obtain real-time alarm data of a power grid system;

[0034] a converting module configured to perform vector conversion on the real-time alarm data to obtain a real-time alarm vector;

[0035] a matching module configured to perform vector matching on the real-time alarm vector in an alarm processing knowledge base, the alarm processing knowledge base being constructed by the method of the first aspect;

[0036] a determining module configured to, when a target vector corresponding to the real-time alarm vector exists in the alarm processing knowledge base, determine and output an alarm processing suggestion corresponding to the real-time alarm data according to the target vector, to guide relevant personnel to perform alarm processing based on the alarm processing suggestion.

[0037] In a possible implementation, the determining module is further configured to: if the target vector corresponding to the real-time alarm vector does not exist in the alarm processing knowledge base, extract second attribute information corresponding to the real-time alarm data based on the knowledge extraction model; and determine the processing suggestion contained in the second attribute information as the alarm processing suggestion corresponding to the real-time alarm data.

[0038] In a fifth aspect, the present application provides an electronic device, comprising: a memory, a processor;

[0039] The memory stores computer execution instructions.

[0040] The processor executes the computer execution instructions stored in the memory, so that the processor executes the alarm processing knowledge base construction method of the first aspect, and / or executes the alarm processing method of the second aspect.

[0041] In a sixth aspect, the present application provides a computer readable storage medium, the computer readable storage medium stores computer execution instructions, and the computer execution instructions are used to implement the alarm processing knowledge base construction method of the first aspect, and / or implement the alarm processing method of the second aspect when executed by a processor.

[0042] In a seventh aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the alarm processing knowledge base construction method of the first aspect, and / or implements the alarm processing method of the second aspect.

[0043] The alarm processing knowledge base construction method and the alarm processing method based on a large model provided by the present application obtain alarm data of a power grid system; determine first attribute information corresponding to the alarm data according to the alarm data and a pre-trained knowledge extraction model, the first attribute information including an alarm category, an alarm reason, an influence range and a processing suggestion, and the knowledge extraction model is constructed based on a large model; and construct an alarm processing knowledge base of the power grid system according to the first attribute information. The present application uses the powerful learning and generalization capabilities of the large model to extract attribute information corresponding to the alarm data of the power grid system, realizes in-depth mining and reasoning of potential laws and characteristics in the alarm data, constructs the alarm processing knowledge base based on the attribute information, thereby improving the accuracy and coverage of the alarm processing knowledge base, and in addition, the automatic construction of the knowledge base improves the efficiency of the knowledge base construction, provides effective and reliable decision support for operation and maintenance personnel, and further improves the reliability and stability of the power grid system operation, reduces the fault processing cost and risk. BRIEF DESCRIPTION OF DRAWINGS

[0044] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate an embodiment consistent with the present application and, together with the description, serve to explain the principles of the application.

[0045] Figure 1 A schematic diagram illustrating a scenario for the alarm processing knowledge base construction method based on a large model provided in this application embodiment;

[0046] Figure 2 A flowchart illustrating the alarm processing knowledge base construction method based on a large model provided in this application embodiment. Figure 1 ;

[0047] Figure 3 A flowchart illustrating the alarm processing method provided in this application embodiment;

[0048] Figure 4 A flowchart illustrating the alarm processing knowledge base construction method based on a large model provided in this application embodiment. Figure 2 ;

[0049] Figure 5 Schematic diagram of the structure of the alarm processing knowledge base construction device based on a large model provided in the embodiments of this application. Figure 1 ;

[0050] Figure 6 Schematic diagram of the structure of the alarm processing knowledge base construction device based on a large model provided in the embodiments of this application. Figure 2 ;

[0051] Figure 7 This is a schematic diagram of the alarm processing device provided in the embodiments of this application;

[0052] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0053] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0054] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0055] The terms "first", "second", and the like in the description and in the claims of this application are used for distinguishing between similar objects and are not necessarily used to describe a particular sequential or chronological order. It is to be understood that the use of the term data herein is not to be construed as limiting in any way, and that the data can be interchanged, where appropriate, to describe the embodiments of the application described herein in terms of the other data. Furthermore, the terms "comprise", "comprising", "has", "having", "include", "including", "contain", "containing", and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises, has, includes, contains a list of steps or elements does not necessarily comprise, have, include, or contain only those steps or elements but can include additional steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

[0056] In the daily production activities of the power grid, a large amount of maintenance experience and professional knowledge is accumulated for the processing of various types of alarms. In the related art, these experiences and knowledge are mainly collected and sorted manually to construct an alarm processing knowledge base. However, when facing a large amount of alarms, maintenance experience and knowledge, the manual method is often inefficient, the collected and sorted information is not comprehensive, and the knowledge base accuracy may be low due to human errors.

[0057] To solve the above technical problems, the application provides a large model-based alarm processing knowledge base construction method. The large model's powerful learning and generalization capabilities are used to extract attribute information contained in the alarm data of the power grid system, realize in-depth mining and reasoning of the potential laws and characteristics in the alarm data, and construct an alarm processing knowledge base based on the attribute information. The accuracy and coverage of the alarm processing knowledge base can be improved, and in addition, the automated method improves the knowledge base construction efficiency.

[0058] The technical solutions of the application and how the technical solutions of the application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the application will be described below with reference to the drawings.

[0059] Figure 1 The scenario diagram of the large model-based alarm processing knowledge base construction method provided by the embodiments of the application is provided. The specific application scenarios of the application can include a power grid system 11 and a computing device 12. The computing device 12 is connected to the power grid system 11. For example, the computing device 12 is in communication connection with the alarm management platform of the power grid system 11. The computing device 12 can obtain the alarm data of the power grid system 11 through the alarm management platform. The large model-based alarm processing knowledge base construction method and the pre-trained knowledge extraction model are deployed on the computing device 12. The alarm processing knowledge base construction method provided by the embodiments of the application is executed on the computing device 12.

[0060] It should be noted that,Figure 1 The application scenario shown is only illustrative, the computing device 12 is a device with a certain computing power, which can be a laptop, a desktop computer, a workstation, a personal digital assistant, etc. The computing device 12 can also be a server, a virtual resource or a server cluster. In addition, the number of computing devices 12 is not limited in the embodiments of the present application, and multiple computing devices can cooperatively complete the alarm processing knowledge base construction method provided by the embodiments of the present application.

[0061] The application scenario of Figure 1 , the alarm processing knowledge base construction method based on a large model provided by the embodiments of the present application will be described with reference to Figure 2 . The alarm processing knowledge base construction method based on a large model provided by the embodiments of the present application is executed by the computing device 12 in Figure 1 . It should be noted that the above-mentioned application scenario is only shown for the purpose of facilitating the understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited by the application scenario shown in Figure 1 .

[0062] Figure 2 The flow of the alarm processing knowledge base construction method based on a large model provided by the embodiments of the present application is shown in Figure 1 , as shown in Figure 2 , the alarm processing knowledge base construction method comprises:

[0063] S201, obtaining alarm data of a power grid system.

[0064] The alarm data is not limited to including the generation time, source, type, level and content description of the alarm signal.

[0065] For example, the computing device 12 interfaces with an alarm management platform in the power grid system, which uniformly manages alarm signals of different subsystems in the power grid system. The computing device 12 obtains the alarm data of the power grid system through the alarm management platform. When obtaining, it can also be based on certain filtering conditions, such as alarm data within a certain time period, alarm data of a certain type, alarm data of a certain subsystem, alarm data of a certain time period and a certain type, etc.

[0066] S202, determining the first attribute information corresponding to the alarm data according to the alarm data and the pre-trained knowledge extraction model, the first attribute information including the alarm category, the alarm reason, the influence range and the processing suggestion, and the knowledge extraction model is constructed based on a large model.

[0067] By way of example, the alarm data is input into the knowledge extraction model, and after deep extraction processing by the knowledge extraction model, first attribute information corresponding to the alarm data is output. The large model can be a large language model suitable for processing text, time series data, and other multi-modal information, such as GPT-4, LLaMA, PaLM, Qwen, and other large models. Compared with traditional machine learning models, large models have higher reasoning and generalization capabilities.

[0068] Taking one set of alarm data as an example, the input of the knowledge extraction model can be:

[0069] {

[0070] "Time": "2024-03-20 10:15:23",

[0071] "Device ID": "SUB_STATION_A_35kV_Transformer_02",

[0072] "Device Type": "35kV Oil-Immersed Transformer",

[0073] "Original Alarm Information": [

[0074] "High-voltage side A-phase current suddenly jumps to 1200A (rated value 630A)",

[0075] "Body oil temperature rises to 95°C (alarm threshold 85°C)",

[0076] "Gas relay acts, light gas signal triggers",

[0077] "Pressure release valve sprays oil"

[0078] }

[0079] The output of the knowledge extraction model can be:

[0080] {

[0081] "Alarm Category": "Device Serious Failure (Transformer Internal Short Circuit)",

[0082] "Alarm Reason": "Insulation aging causes A-phase winding inter-turn short circuit, triggering overcurrent and abnormal oil temperature rise",

[0083] "Influence Scope": {

[0084] "Region": "SUB_STATION_A powered industrial park (about 1200 enterprises)",

[0085] "Device": "Adjacent transformer overload risk increases, may trigger cascading failures",

[0086] "System": "35kV bus voltage fluctuation, affecting the stability of surrounding distributed photovoltaic grid-connected"

[0087] },

[0088] "Processing suggestion": {

[0089] "Emergency operation": [

[0090] "Immediately isolate the faulty transformer (open the 35kV side disconnector)",

[0091] "Start the standby transformer (SUB_STATION_A_Transformer_01)",

[0092] "Notify the dispatch center to adjust the regional load distribution"

[0093] }

[0094] S203、According to the first attribute information, an alarm processing knowledge base of the power grid system is constructed.

[0095] It can be understood that for multiple groups of alarm data, multiple groups of attribute information can be obtained at the same time. The alarm processing knowledge base can be regarded as a database that centrally stores alarm data and its solutions.

[0096] For example, for each group of alarm data, the alarm data and the first attribute information corresponding to the alarm data are associated and mapped, and the unstructured text (i.e. the association and mapping result) is stored to form an alarm processing knowledge base.

[0097] Optionally, multiple keywords of the alarm data can also be extracted, and the same method can be applied to extract multiple keywords contained in the first attribute information. In the form of keyword mapping, the alarm data and the first attribute information are associated and stored.

[0098] Optionally, the first attribute information can also be vectorized to construct a vectorized alarm processing knowledge base.

[0099] In the embodiments of the present application, the powerful learning and generalization capabilities of the large model are used to extract the attribute information contained in the alarm data of the power grid system, to realize in-depth mining and reasoning of the potential rules and characteristics in the alarm data, and to construct an alarm processing knowledge base based on the attribute information, thereby improving the accuracy and coverage of the alarm processing knowledge base. In addition, the automated way improves the efficiency of knowledge base construction, provides effective and reliable decision support for operation and maintenance personnel, and further improves the reliability and stability of the power grid system operation, reduces the cost and risk of fault processing.

[0100] In some embodiments, according to the first attribute information, an alarm processing knowledge base of the power grid system is constructed, including: using the GraphRAG technology to integrate the first attribute information to obtain an alarm processing vector corresponding to the alarm data; and constructing the alarm processing knowledge base of the power grid system according to the alarm processing vector.

[0101] For example, the GraphRAG is used to integrate the first attribute information, the association relationship between the knowledge elements is determined, and the alarm processing knowledge base is formed. The alarm processing knowledge base records the alarm processing vector.

[0102] The GraphRAG is used to convert the first attribute information extracted by the knowledge extraction model into a graph structure of nodes (entities) and edges (relationships), and the nodes and edges are vectorized, so that vector matching can be performed through semantic retrieval in the subsequent process. For example, the node types are devices, faults, events, etc., and the edge types are devices and faults, faults and protection actions, and protection actions and devices. When vectorization (vector embedding) is performed, text data is converted into a fixed-dimensional semantic vector representation (i.e., an alarm processing vector) by using the Sentence-BERT technology and the InstructorEmbedding technology. Further, the alarm processing vector is stored in a vector database (such as Milvus and Faiss) to obtain the alarm processing knowledge base.

[0103] In the embodiments of the application, the GraphRAG technology is used to convert unstructured text data into an unstructured vector knowledge base. The vector knowledge base supports fast similarity retrieval, improves retrieval efficiency, is suitable for large-scale knowledge bases, and is especially suitable for power grid system alarm processing scenarios with high real-time requirements.

[0104] Optionally, the graph structure can also be stored in a graph database (such as Neo4j), and the graph database (such as Neo4j) and the vector database (such as Milvus) are combined into the alarm processing knowledge base. This facilitates mixed queries when alarm data matching is performed in the alarm processing knowledge base. That is, the graph database stores the graph structure and supports relationship traversal, the vector database stores the vectors of the nodes / edges and supports similarity retrieval, and when alarm data matching is performed, the graph structure and the vector similarity are combined, sorted, and then the processing suggestions corresponding to the alarm data are output, so that the effect of optimizing the relevance of the results is achieved.

[0105] Optionally, the alarm processing knowledge base also supports user browsing, searching, and editing of the knowledge base content. At the same time, a visual display interface is provided to facilitate the user to intuitively understand the knowledge system of the alarm.

[0106] In some embodiments, the knowledge extraction model is obtained by: obtaining historical alarm data of a power grid system and auxiliary data corresponding to the historical alarm data; preprocessing the historical alarm data and the auxiliary data to obtain processed alarm data, the preprocessing including data cleaning and normalization; and iteratively training a large model according to the processed alarm data and a preset expert rule to obtain the knowledge extraction model, the large model including a Transformer model.

[0107] The auxiliary data is not limited to including alarm-related system operation parameters (such as voltage, current, power, load, temperature, etc.), log files. The historical alarm data and the auxiliary data can be generally classified as text data and time series data.

[0108] For example, when preprocessing data, data cleaning for text data can include removing irrelevant content, removing duplicates, filtering stop words, and unifying encoding, and normalization can include standardizing formats such as unifying case, normalizing numbers, and expanding abbreviations. Data cleaning for time series data can include missing value processing and outlier detection, and normalization can use Min-Max, Z-Score, etc. By preprocessing the historical alarm data and the auxiliary data, more accurate sample data (i.e., processed alarm data) is obtained.

[0109] The preset expert rule can be understood as expert knowledge, i.e., the processing suggestions provided by experts corresponding to the alarm. When training the model, the analysis experience, disposal suggestions, and other knowledge of the experts on the alarm are integrated into the model training process, for example, before training, during training, and after training. Taking the pre-training integration as an example, synthetic data can be generated according to the expert rule, the training set is expanded, the expanded training set is applied, the hyperparameters of the model are iteratively adjusted, so that the model can learn the feature representation of the alarm data, the generation rule, the association with the system operation state, and the processing suggestions.

[0110] Optionally, when the model is trained, when the experts correct the prediction errors of the model, the corrected samples can also be added to the training set and the model can be fine-tuned.

[0111] In the embodiments of the present application, expert knowledge is introduced for supervised learning during model training, the analysis experience, disposal suggestions, and other knowledge of the experts on the alarm signal are integrated into the model training, the understanding and judgment ability of the knowledge extraction model on the alarm signal is improved, the performance of the model in complex alarm scenarios can be significantly improved, and the model is particularly suitable for scenarios in the power field that require high safety and interpretability. In addition, compared with the training of traditional machine learning models, the training of large models requires less feature labeling, thereby reducing the workload of manual labeling.

[0112] In practical applications, as the power grid system is continuously upgraded, the operating environment changes, and new technologies are applied, new alarm types are constantly emerging, and the alarm processing knowledge base needs to be updated in a timely manner. In some embodiments, the alarm processing knowledge base construction method further includes: based on the knowledge extraction model, extracting target attribute information corresponding to the target alarm data, the target alarm data being a set of alarm data that has not been generated in the power grid system; and updating the alarm processing knowledge base according to the target attribute information.

[0113] For example, the target alarm data is input into the knowledge extraction model, and after deep extraction processing by the knowledge extraction model, the target attribute information corresponding to the target alarm data is output. The target attribute information is integrated using the GraphRAG technology to clearly define the association relationship between the knowledge elements, and the target attribute information is converted into an alarm processing vector in a vector form and added to the alarm processing knowledge base to complete the update of the alarm processing knowledge base.

[0114] In an implementation manner, the update of the alarm processing knowledge base is triggered manually, that is, when it is detected that a user requests to update the alarm processing knowledge base, the target alarm data is used to update the alarm processing knowledge base. The target alarm data can be a set of alarm data of a new alarm type that is manually sorted, or a set of alarm data of a new alarm type that is obtained based on logical judgment by the computing device 12.

[0115] Optionally, when the target attribute information corresponding to the target alarm data is extracted based on the knowledge extraction model, the knowledge extraction model can also output the confidence (such as 0-1 score) of each attribute, and when the confidence is lower than a threshold, manual review is triggered, and the alarm processing knowledge base is updated based on the target attribute information modified by the manual review.

[0116] In another implementation manner, when the computing device 12 detects that a new alarm type appears, the update of the alarm processing knowledge base is automatically triggered.

[0117] In the embodiments of the present application, the alarm processing knowledge base is dynamically updated to ensure that the alarm processing knowledge base can timely and effectively integrate new knowledge, provide effective, timely, and comprehensive alarm processing guidance suggestions for the operation and maintenance personnel, and thus provide a solid foundation for improving the reliability and stability of system operation and reducing the cost and risk of fault processing.

[0118] The traditional alarm processing method mainly relies on manual experience and simple rule matching, and it is difficult for manual operation to quickly and accurately judge and classify a large number of alarm signals of different types and different sources, thereby affecting the timely response and processing of system faults. In some embodiments, for example, Figure 3 As shown in the flowchart of the alarm processing method provided by the embodiments of the present application, the alarm processing method includes:

[0119] S301, acquire real-time alarm data of the power grid system.

[0120] S302, perform vector conversion on the real-time alarm data to obtain a real-time alarm vector.

[0121] S303, perform vector matching on the real-time alarm vector in the alarm processing knowledge base.

[0122] S304, if there is a target vector corresponding to the real-time alarm vector in the alarm processing knowledge base, determine and output an alarm processing suggestion corresponding to the real-time alarm data according to the target vector, to guide relevant personnel to process the alarm based on the alarm processing suggestion.

[0123] For example, still referring to Figure 1 , the alarm processing method is also deployed in the computing device 12, the computing device 12 acquires real-time alarm data from the alarm management platform in the power grid system 11 in real time, converts the real-time alarm data into a fixed-dimension, continuous vector representation using the Embedding technology, and the purpose of converting the real-time alarm data into a vector is to quantify the features of the alarm, facilitate subsequent similarity calculation and matching, and the Embedding can also automatically capture semantic associations.

[0124] For similarity measurement, cosine similarity, Euclidean distance, inner product, etc. can be used to quickly find the most similar historical alarm data to the real-time alarm data in the alarm processing knowledge base. Specifically, the top K alarm vectors corresponding to the similarity threshold can be set as target vectors by setting a similarity threshold. The target vector is converted into text again using the Embedding technology, and the alarm processing suggestion associated with the target vector can be obtained.

[0125] Optionally, after obtaining the alarm processing suggestion corresponding to the real-time alarm data, relevant operation and maintenance personnel can be notified through email, SMS or system message, and the notification content at least includes the real-time alarm data and its alarm processing suggestion.

[0126] In the embodiments of the present application, the newly converted real-time alarm vector is compared with the existing alarm vectors in the alarm processing knowledge base for similarity calculation or matching to find the most similar historical alarm and its processing suggestion, improve the search efficiency of the alarm processing suggestion, provide timely and effective decision support for system maintenance personnel, reduce the dependence on manual experience, improve the fault processing efficiency, reduce the time and impact range of service interruption, and thus improve customer satisfaction.

[0127] In some embodiments, the alarm processing method further comprises: if the target vector corresponding to the real-time alarm vector does not exist in the alarm processing knowledge base, extracting second attribute information corresponding to the real-time alarm data based on the knowledge extraction model; and determining the processing suggestion contained in the second attribute information as the alarm processing suggestion corresponding to the real-time alarm data.

[0128] For example, if the alarm vector satisfying the similarity threshold does not exist in the alarm processing knowledge base, it is determined that the real-time alarm data is a new type of alarm that has never occurred in history or has never been integrated into the alarm processing knowledge base. At this time, the second attribute information corresponding to the real-time alarm data is extracted through the knowledge extraction model, the alarm category, alarm reason, impact range and processing suggestion are contained in the second attribute information, and the processing suggestion is determined as the alarm processing suggestion corresponding to the real-time alarm data.

[0129] In the embodiments of the present application, when it is determined that the target vector corresponding to the real-time alarm vector does not exist in the alarm processing knowledge base, the pre-trained knowledge extraction model with strong reasoning ability and generalization ability is used to predict the second attribute information based on the real-time alarm data, and then the alarm processing suggestion is obtained, thereby solving the problem of insufficient coverage of the alarm processing knowledge base. The system can automatically generate a temporary processing suggestion, improve the robustness of the system, improve the response speed to new alarms, and further improve the intelligent level, adaptability and efficiency of alarm processing.

[0130] Figure 4 Flowchart of the alarm processing knowledge base construction method based on a large model provided by the embodiments of the present application Figure 2 As shown in Figure 4 , the alarm processing knowledge base construction method generally involves data collection and preprocessing, large model training and optimization, knowledge extraction and integration, and knowledge base updating mechanism. Specifically, it includes:

[0131] 1. Obtain historical alarm data of a power grid system and auxiliary data corresponding to the historical alarm data.

[0132] The alarm signal data and auxiliary data are collected in Figure 4 .

[0133] 2. Preprocess the historical alarm data and auxiliary data to obtain processed alarm data, and the preprocessing includes data cleaning and normalization.

[0134] The data cleaning and normalization processing in Figure 4 .

[0135] 3. According to the processed alarm data and the preset expert rules, iteratively train the large model to obtain a knowledge extraction model.

[0136] A large model architecture suitable for processing multi-modal information such as text and time series data is selected as a base model, such as a variant of the Transformer architecture. The large model is trained using processed alarm data, and by adjusting the model's hyperparameters, optimization algorithms, and other means, the model can learn the characteristic representation, generation rules, and association with system operating status of the alarm data. During the training process, expert knowledge is introduced for supervised learning, and the knowledge of the domain experts, such as analysis experience and treatment suggestions for alarm signals, is integrated into the model training to improve the model's understanding and judgment of alarm signals. Ultimately, the model can learn the characteristic representation, generation rules, association with system operating status, and treatment suggestions of the alarm data.

[0137] The process involves steps 3-6 in Figure 4 .

[0138] 4. Obtain alarm data of the power grid system.

[0139] For example, alarm data within a set time period, alarm data of a set type, alarm data of a set subsystem, alarm data of a set type and a set time period, and the like.

[0140] 5. Determine the first attribute information corresponding to the alarm data according to the alarm data and the pre-trained knowledge extraction model.

[0141] That is, the knowledge extraction model is used to extract knowledge from the alarm data, and key information (first attribute information) such as the category, reason, impact range, and treatment method of the alarm is extracted.

[0142] 6. Construct an alarm handling knowledge base of the power grid system according to the first attribute information.

[0143] Integrate the first attribute information using GraphRAG to clarify the association between knowledge elements and form an alarm handling knowledge base that records alarm handling vectors.

[0144] 7. Extract target attribute information corresponding to target alarm data based on the knowledge extraction model, where the target alarm data is a set of alarm data that has not occurred in the power grid system.

[0145] For example, when the computing device 12 detects the occurrence of a new alarm type, it automatically triggers an update to the alarm handling knowledge base. During the update, the target alarm data is input into the knowledge extraction model, which performs deep extraction processing and outputs the target attribute information corresponding to the target alarm data.

[0146] The process involves steps 10-11 in Figure 4 .

[0147] 8. Update the alarm handling knowledge base according to the target attribute information.

[0148] The target attribute information is integrated by using the GraphRAG technology, the association relationship between the knowledge elements is determined, the target attribute information is converted into an alarm processing vector in a vector form, and is added to the alarm processing knowledge base to complete updating of the alarm processing knowledge base.

[0149] In actual application, the performance of the alarm processing knowledge base construction system can be evaluated and optimized regularly, including performance tuning of the large model, structure optimization of the alarm processing knowledge base, and communication optimization between modules, etc., to ensure that the system always maintains high efficiency and stability during long-term operation. At the same time, according to the needs of business development, the content of the alarm processing knowledge base is continuously expanded and improved, and the processing capability of the system for new alarm signals is improved.

[0150] Next, the hardware and software environment required for executing the large model-based alarm processing knowledge base construction method and the alarm processing method provided in the present application will be described. It should be noted that a set of hardware environment commonly used for implementing the alarm processing knowledge base construction method and the alarm processing method is taken as an example for description.

[0151] For the construction of the hardware environment, a server cluster with high-performance computing capability is usually selected as the basic hardware platform for system operation to ensure the efficiency of large model training and real-time data processing. The server configuration should meet the memory, storage and computing resource requirements of the large model, and be equipped with high-speed network equipment to ensure the stability and low delay of data transmission.

[0152] For the deployment of the software system, three parts are involved, including basic software, data collection and knowledge graph construction module, and large model and GraphRAG integration. For the basic software, the operating system, graph database, such as Apache HugeGraph (distributed architecture, supporting trillion-level edge storage) or Neo4j Enterprise (suitable for small and medium-sized graphs), and large model development framework PyTorch + DeepSpeed (distributed training optimization) and other related software environments are installed on the server, and GraphRAG is installed to support community detection (Leiden algorithm) and summary generation. For the data collection and knowledge graph construction module, it includes multi-source alarm data access, knowledge extraction and graph construction, entity relationship extraction (calling LLM (such as GPT-4 or Qwen) to perform triple extraction (entity-relation-entity), supporting dynamic attributes (such as device state, alarm level)), hierarchical clustering (text block (default 1200 characters / block), extracting entities and relationships), graph storage optimization (embedding entity, community summary, and original text block into a vector database (such as Faiss) to realize “vector + graph” dual-path retrieval), and running the Leiden community detection algorithm to generate sub-communities and automatically generate summaries (such as “power equipment failure community” containing transformer, circuit breaker and other entities and their associated rules). For large model and GraphRAG integration, it involves model deployment (using NVIDIA inference server to deploy quantized LLM); developing API interfaces to support two query modes, global query: calling community summary to answer macroscopic questions (such as “system recent main alarm type distribution”), local query: generating answers based on entity multi-hop retrieval (such as “transformer A associated fault history”); prompt engineering: designing structured prompt templates, injecting subgraph context and reasoning path.

[0153] Figure 5 Structure diagram of the alarm processing knowledge base construction device based on a large model provided by the embodiment of the present application Figure 1 As shown in Figure 5 The alarm processing knowledge base construction device 50 based on a large model provided by the embodiment of the present application comprises an acquisition module 51, a determination module 52, and a construction module 53. Among them:

[0154] The acquisition module 51 is configured to acquire alarm data of a power grid system.

[0155] The determination module 52 is configured to determine first attribute information corresponding to the alarm data according to the alarm data and a pre-trained knowledge extraction model, wherein the first attribute information includes alarm category, alarm reason, impact range, and processing suggestion, and the knowledge extraction model is constructed based on a large model.

[0156] The construction module 53 is configured to construct an alarm processing knowledge base of the power grid system according to the first attribute information.

[0157] In one possible implementation, the construction module 53 is specifically used to: integrate attribute information using GraphRAG technology to obtain alarm processing vectors corresponding to alarm data; and construct an alarm processing knowledge base for the power grid system based on the alarm processing vectors.

[0158] In one possible implementation, the knowledge extraction model is obtained by: acquiring historical alarm data of the power grid system and corresponding auxiliary data; preprocessing the historical alarm data and auxiliary data to obtain processed alarm data, the preprocessing including data cleaning and normalization; and iteratively training a large model based on the processed alarm data and preset expert rules to obtain the knowledge extraction model, the large model including the Transformer model.

[0159] like Figure 6 As shown, in one possible implementation, the alarm processing knowledge base construction device 50 based on a large model further includes an update module 54, which is used to: extract target attribute information corresponding to the target alarm data based on the knowledge extraction model, wherein the target alarm data is a collection of alarm data that has not been generated in the power grid system; and update the alarm processing knowledge base according to the target attribute information.

[0160] Figure 7 This is a schematic diagram of the alarm processing device provided in the embodiments of this application, as shown below. Figure 7 As shown, the alarm processing device 70 includes: an acquisition module 71, a conversion module 72, a matching module 73, and a determination module 74. Wherein:

[0161] Module 71 is used to acquire real-time alarm data of the power grid system;

[0162] The conversion module 72 is used to perform vector conversion on the real-time alarm data to obtain the real-time alarm vector.

[0163] The matching module 73 is used to perform vector matching for real-time alarm vectors in the alarm processing knowledge base, which is constructed through the embodiments in the above-mentioned alarm processing knowledge base construction method based on a large model.

[0164] The determination module 74 is used to determine and output alarm processing suggestions corresponding to real-time alarm data based on the target vector when there is a target vector corresponding to the real-time alarm vector in the alarm processing knowledge base, so as to guide relevant personnel to perform alarm processing based on the alarm processing suggestions.

[0165] In a possible implementation, the determining module 74 is further configured to: if the target vector corresponding to the real-time alarm vector does not exist in the alarm processing knowledge base, extract second attribute information corresponding to the real-time alarm data based on the knowledge extraction model; and determine the processing suggestion contained in the second attribute information as the alarm processing suggestion corresponding to the real-time alarm data.

[0166] The alarm processing apparatus provided in this embodiment can execute the method provided in the method embodiments, and has similar implementation principles and technical effects. Details are not described herein again.

[0167] Figure 8 A structural schematic diagram of an electronic device provided in this embodiment is shown in FIG. 8. As shown in FIG. 8, the electronic device 80 provided in this embodiment includes at least one processor 801 and a memory 802. Optionally, the electronic device 80 further includes a communication component 803. The processor 801, the memory 802 and the communication component 803 are connected through a bus 804. Figure 8

[0168] In the implementation process, the at least one processor 801 executes the computer-executed instructions stored in the memory 802, so that the at least one processor 801 executes the method described above.

[0169] The specific implementation process of the processor 801 can refer to the method embodiments described above, and has similar implementation principles and technical effects. Details are not described herein again.

[0170] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the application can be directly embodied as execution completed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0171] The memory can include a random access memory (RAM), and can also include a non-volatile memory (NVM), for example, at least one disk memory.

[0172] ​The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application does not limit to only one bus or one type of bus.

[0173] The embodiment of the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described above.

[0174] The embodiment of the present application further provides a computer readable storage medium, which stores computer execution instructions, and when a processor executes the computer execution instructions, the method described above is implemented.

[0175] The readable storage medium described above can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0176] An exemplary readable storage medium is coupled to the processor, so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the device.

[0177] The division of units is only a logical functional division, and in actual implementation, there can be another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0178] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e. may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0179] In addition, each functional unit in various embodiments of the application can be integrated into one processing unit, or each unit can exist physically, or two or more units can be integrated into one unit.

[0180] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the application essentially or the part of the prior art that contributes to the technical solutions or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the various embodiment methods of the application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0181] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The aforementioned program can be stored in a computer readable storage medium. The program executes the steps including the above-mentioned method embodiments when executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, and various media that can store program codes.

[0182] Finally, it should be noted that those skilled in the art, after considering the specification and practicing the application disclosed herein, will easily think of other embodiments of the application. The application is intended to cover any variations, uses or adaptations of the application that follow the general principles of the application and include common knowledge or conventional technical means in the art that are not disclosed in the application, and is not limited to the precise structure described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the application is only limited by the appended claims.

Claims

1. A method for constructing an alarm processing knowledge base based on a large model, characterized by, The method comprises the following steps: obtaining alarm data of a power grid system; determining first attribute information corresponding to the alarm data according to the alarm data and a pre-trained knowledge extraction model, wherein the first attribute information comprises alarm categories, alarm causes, influence ranges and processing suggestions, and the knowledge extraction model is constructed based on a large model; constructing an alarm processing knowledge base of the power grid system according to the first attribute information.

2. The method of claim 1, wherein the constructing of the alarm processing knowledge base comprises: The step of constructing the alarm processing knowledge base of the power grid system according to the first attribute information comprises the following steps: integrating the first attribute information by using a graph retrieval augmented generation (GraphRAG) technology to obtain alarm processing vectors corresponding to the alarm data; constructing the alarm processing knowledge base of the power grid system according to the alarm processing vectors.

3. The method of claim 1, wherein the constructing of the alarm processing knowledge base comprises: The knowledge extraction model is obtained by the following method: obtaining historical alarm data of a power grid system and auxiliary data corresponding to the historical alarm data; preprocessing the historical alarm data and the auxiliary data to obtain processed alarm data, wherein the preprocessing comprises data cleaning and normalization; iteratively training the large model according to the processed alarm data and a preset expert rule to obtain the knowledge extraction model, wherein the large model comprises a Transformer model.

4. The alarm processing knowledge base construction method according to claim 2 or 3, characterized by, The method further comprises the following steps: extracting target attribute information corresponding to target alarm data based on the knowledge extraction model, wherein the target alarm data is a set of alarm data that has not occurred in the power grid system; updating the alarm processing knowledge base according to the target attribute information.

5. An alarm processing method characterized by comprising: The method comprises the following steps: obtaining real-time alarm data of a power grid system; performing vector conversion on the real-time alarm data to obtain real-time alarm vectors; performing vector matching on the real-time alarm vectors in an alarm processing knowledge base, wherein the alarm processing knowledge base is constructed by the method according to any one of claims 1 to 4; if there is a target vector corresponding to the real-time alarm vectors in the alarm processing knowledge base, determining and outputting alarm processing suggestions corresponding to the real-time alarm data according to the target vector to guide relevant personnel to process the alarm based on the alarm processing suggestions.

6. The alert processing method of claim 5, wherein, The method further comprises the following steps: if there is no target vector corresponding to the real-time alarm vectors in the alarm processing knowledge base, extracting second attribute information corresponding to the real-time alarm data based on the knowledge extraction model; determining processing suggestions contained in the second attribute information as alarm processing suggestions corresponding to the real-time alarm data. 7.A device for constructing a large model-based alarm processing knowledge base, characterized by comprising: The method comprises the following steps: an obtaining module configured to obtain alarm data of a power grid system; a determining module configured to determine first attribute information corresponding to the alarm data according to the alarm data and a pre-trained knowledge extraction model, wherein the first attribute information comprises alarm categories, alarm causes, influence ranges and processing suggestions, and the knowledge extraction model is constructed based on a large model; a constructing module configured to construct an alarm processing knowledge base of the power grid system according to the first attribute information.

8. An alarm processing apparatus characterized by comprising: The method comprises the following steps: an obtaining module configured to obtain real-time alarm data of a power grid system; a converting module configured to perform vector conversion on the real-time alarm data to obtain real-time alarm vectors; The matching module is configured to perform vector matching on the real-time alarm vector in an alarm processing knowledge base, wherein the alarm processing knowledge base is constructed by the method in any one of claims 1 to 4. The determining module is configured to, when the target vector corresponding to the real-time alarm vector exists in the alarm processing knowledge base, determine and output an alarm processing suggestion corresponding to the real-time alarm data according to the target vector, so as to guide a relevant person to perform alarm processing based on the alarm processing suggestion.

9. An electronic device, comprising: Comprise: A memory, a processor; The memory stores computer execution instructions; The processor executes the computer execution instructions stored in the memory, so that the processor executes the method in any one of claims 1 to 4, and / or realizes the method in claim 5 or 6.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed to realize the method in any one of claims 1 to 4, and / or realize the method in claim 5 or 6.

Images