Transformer state diagnosis method and system, and computer program product

Abstract

The invention belongs to a diagnosis method based on a large model, and provides a transformer state diagnosis method, a transformer state diagnosis system and a computer program product to solve the problems that when a knowledge graph and the large model are combined at present, data integration is difficult, feature fusion is insufficient, and old knowledge is easily forgotten by the model. Transformer real-time multi-mode time sequence signal data and a text problem are input into the transformer state diagnosis multi-mode large model, the transformer state is diagnosed by means of the transformer state diagnosis multi-mode large model, and the transformer state diagnosis multi-mode large model comprises a large language model and a mode encoder. The knowledge source of the large language model is a knowledge graph based on transformer state diagnosis, the modal encoder is used for extracting time sequence features from transformer real-time multi-modal time sequence signal data, and the transformer state diagnosis multi-modal large model is used for processing text problems and extracting key features through multi-modal time sequence signals. The transformer state diagnosis multi-mode large model is a model adjusted based on a graph neural network.

Description

Technical Field

[0001] The present application relates to a large model-based diagnostic method, and specifically to a transformer status diagnostic method, system, and computer program product. Background Art

[0002] In modern industrial systems, transformers are key power equipment, and operating status monitoring and fault diagnosis are crucial to ensuring the safety and stability of power systems. However, traditional transformer fault diagnosis methods often rely on a single type of data or rules based on expert experience, making it difficult to fully capture multi-dimensional information under complex working conditions. In addition, with the diversification of data sources, including the emergence of a large number of multimodal time series signals such as current, voltage, oil temperature, and oil chromatography, how to effectively integrate and analyze these data has become a major challenge.

[0003] In recent years, knowledge graph technology has shown great potential in processing complex information and semantic reasoning. It can organize knowledge from multiple heterogeneous data sources in the form of graph structures to facilitate subsequent query and analysis. At the same time, with the development of large-scale pre-trained language models, these models have performed well in tasks such as natural language processing, data analysis, and intelligent diagnosis, especially when combined with multimodal data, which can achieve more accurate prediction and judgment.

[0004] However, when combining knowledge graphs with large models, existing technologies still have problems such as difficulty in data integration, insufficient feature fusion, and the model easily forgetting old knowledge. Summary of the invention

[0005] This application provides a transformer status diagnosis method, system, and computer program product to address the technical problems of difficulty in data integration, insufficient feature fusion, and the model's tendency to forget old knowledge when currently combining knowledge graphs with large models.

[0006] In order to achieve the above objectives, this application adopts the following technical solutions: In a first aspect, the present application proposes a transformer status diagnosis method, comprising: Obtain real-time multi-modal time series signal data of transformer; Input the transformer real-time multimodal time series signal data and textual questions into the transformer state diagnosis multimodal large model; the transformer state diagnosis multimodal large model includes a large language model and a modal encoder, the knowledge source of the large language model is a knowledge graph based on transformer state diagnosis, and the modal encoder is used to extract time series features from the transformer real-time multimodal time series signal data; the textual questions are used to inquire about the transformer state; the transformer state diagnosis multimodal large model is a model adjusted based on a graph neural network, and can perform spatial fusion of features of different modalities; The transformer state diagnosis multi-modal large model is enabled to output transformer state diagnosis results.

[0007] Furthermore, the method for constructing a knowledge graph based on transformer status diagnosis includes: Collect knowledge sources related to transformer status diagnosis, and extract knowledge points related to transformer faults from the knowledge sources; Organizing and storing knowledge points in a standardized manner to form knowledge representation; the knowledge representation includes entities, attributes and relationships, where entities are used to represent concepts or objects in the transformer, attributes are used to represent the characteristics of entities, and relationships are used to represent the associations between entities; The entities and relations in the knowledge representation are aligned, and ambiguities in the entities and relations are eliminated to obtain a processed knowledge graph.

[0008] Furthermore, after obtaining the processed knowledge graph, the method further includes: The processed knowledge graph is stored in the Neo4j graph database.

[0009] Furthermore, the large language model adopts a QWEN2 large model, and the modal encoder adopts a Transformer model.

[0010] Furthermore, the training method of the large language model includes: Map the entities and relations in the knowledge graph based on transformer condition diagnosis into low-dimensional vector representations to obtain an embedding vector; generating an input instruction associated with a transformer status diagnosis task in combination with the embedded vector; Conditional instructions are set based on the relationships in the knowledge graph based on transformer status diagnosis to guide the large language model on how to process the input under corresponding conditions; Setting an output instruction set, wherein the output instruction set is used to regulate the output of the large language model to conform to known knowledge in the knowledge graph based on transformer state diagnosis; Set the true value of the transformer fault category to enable the large language model to output content in a preset format; The large language model is trained by inputting instructions, conditional instructions, and output instruction sets combined with true values.

[0011] Furthermore, mapping entities and relationships in the knowledge graph based on transformer status diagnosis into low-dimensional vector representations includes: A graph embedding algorithm is used to map the entities and relations in the knowledge graph based on transformer condition diagnosis into low-dimensional vector representations.

[0012] Furthermore, the adjustment based on graph neural network includes: Constructing a graph structure, in which each node represents a feature vector of a modality, and edges between the nodes represent similarities or correlations between feature vectors of different modalities; By calculating the cosine similarity or Euclidean distance between feature vectors, the edge weights are generated to obtain the graph structure of multimodal features; Use graph neural networks to propagate features of the graph structure of multimodal features and generate unified feature representations; The large multimodal model for transformer condition diagnosis is adjusted through unified feature representation.

[0013] Furthermore, the graph neural network-based adjustment also includes: introducing elastic weight consolidation technology.

[0014] In a second aspect, the present application proposes a transformer status diagnosis system, comprising: An acquisition module is used to acquire real-time multi-modal time series signal data of the transformer; An input module is used to input transformer real-time multimodal time series signal data and textual questions into a transformer state diagnosis multimodal large model; the transformer state diagnosis multimodal large model includes a large language model and a modal encoder, the knowledge source of the large language model is a knowledge graph based on transformer state diagnosis, and the modal encoder is used to extract time series features from the transformer real-time multimodal time series signal data; the textual questions are used to inquire about the transformer state; the transformer state diagnosis multimodal large model is a model adjusted based on a graph neural network, and can perform spatial fusion of features of different modalities; The output module is used to enable the transformer state diagnosis multi-modal large model to output the transformer state diagnosis result.

[0015] In a third aspect, the present application proposes a computer program product, wherein the computer program product comprises instructions, and when the instructions are executed by a processor, the transformer status diagnosis method as described above is implemented.

[0016] Compared with the prior art, this application has the following beneficial effects: The present application proposes a transformer state diagnosis method, which inputs the transformer real-time multimodal time series signal data and textual problems into the transformer state diagnosis multimodal large model, and diagnoses the transformer state with the help of the transformer state diagnosis multimodal large model, wherein the transformer state diagnosis multimodal large model includes a large language model and a modal encoder, the knowledge source of the large language model is a knowledge graph based on transformer state diagnosis, the modal encoder is used to extract time series features from the transformer real-time multimodal time series signal data, the large language model integrates knowledge from multiple sources to construct a corresponding knowledge graph, and uses the transformer state diagnosis multimodal large model to process textual problems and multimodal time series signals to extract key features. At the same time, the transformer state diagnosis multimodal large model is a model adjusted based on the graph neural network, which realizes the fusion and propagation of multimodal features, and provides a more efficient, accurate and stable transformer fault diagnosis method, which can efficiently and fully realize data integration.

[0017] The present application also proposes a transformer state diagnosis system, a computer program product, which has all the advantages of the above transformer state diagnosis method. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A first flow chart of a transformer status diagnosis method for this application; Figure 2 A second flow chart of a transformer status diagnosis method for this application; Figure 3 A schematic diagram of a transformer status diagnosis system according to the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0022] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0023] In the description of the embodiments of the present application, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings, or the orientation or position relationship in which the invented product is usually placed when used. It is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0024] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0025] In the description of the embodiments of the present application, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0026] Transformers play a vital role in the power system. Transformers are key equipment that connect power grids of different voltage levels and are responsible for the transmission, distribution and voltage conversion of electric energy. Therefore, ensuring the safe and stable operation of transformers is of great significance to the reliability and stability of the entire power system. As an important part of transformer maintenance and management, operating status monitoring and fault diagnosis are of self-evident importance. They can not only help operators to detect and deal with potential faults in a timely manner, reduce power outages and economic losses caused by faults, but also improve the service life and operating efficiency of transformers. In addition, with the development of intelligent technology, the monitoring and diagnostic methods of transformers are also constantly upgraded, providing a more powerful guarantee for the safety and stability of the power system.

[0027] The combination of knowledge graph technology and large-scale pre-trained language models has shown unprecedented potential in processing complex information and semantic reasoning, but it also faces a series of technical challenges. Knowledge graphs can organize knowledge from multiple heterogeneous data sources in the form of graph structures. This structured representation is easy for computers to understand and process, providing strong support for subsequent queries and analysis. Through the entities, relationships and attributes in the knowledge graph, complex reasoning rules can be constructed to achieve knowledge-based intelligent reasoning. This reasoning ability is particularly important when dealing with complex problems, such as intelligent question answering, recommendation systems and other fields. In addition, large-scale pre-trained language models have made significant progress in natural language processing tasks. Large-scale pre-trained language models can understand and generate natural language texts, providing powerful text processing capabilities for the construction and application of knowledge graphs. Although knowledge graphs and multimodal pre-trained models perform well in processing data in their respective fields, when combining the two, how to effectively integrate data from different sources and different types is still a difficult problem. Differences in the quality and format of different data may lead to information loss or errors in the integration process. There may be a mismatch between the structured features of the knowledge graph and the distributed features of the pre-trained model during the fusion process. How to effectively integrate the characteristics of the two to give full play to their respective advantages is an urgent problem to be solved.

[0028] Based on the above situation, the present application proposes a transformer status diagnosis method, system, and computer program product. The present application is described in detail below in conjunction with embodiments and drawings.

[0029] like Figure 1 As shown, it is a first flow chart of a transformer status diagnosis method of the present application, which may include: S101, obtaining real-time multi-modal time series signal data of the transformer.

[0030] In practical applications, data can be collected in real time from a variety of sensors installed on the transformer. These sensors may include temperature sensors, vibration sensors, current and voltage sensors, etc., which can be selected according to the diagnostic objectives. These sensors can capture various physical and chemical state changes of the transformer during operation. These collected data are multimodal, that is, they contain multiple types of information, such as temperature, vibration frequency, current and voltage values, etc., and these data are time-series signals that change with time, which is crucial for analyzing the dynamic behavior of the transformer.

[0031] S102, input the transformer real-time multi-modal time series signal data and textual questions into the transformer state diagnosis multi-modal large model.

[0032] After obtaining the real-time multi-modal time series signal data of the transformer, these data will be input into the transformer status diagnosis multi-modal large model for diagnosis. At the same time, a text question will also be input, which is usually asked by the operation and maintenance personnel or the automation system to inquire about the current status of the transformer or diagnose a specific problem.

[0033] The transformer state diagnosis multimodal large model includes a large language model and a modal encoder. The knowledge source of the large language model is a knowledge graph based on transformer state diagnosis. The modal encoder is used to extract time series features from the transformer real-time multimodal time series signal data. The large language model is responsible for processing and understanding the input textual questions. Its knowledge source is the knowledge graph based on transformer state diagnosis. This knowledge graph usually contains a lot of information about transformer structure, working principle, common faults and treatment methods. The modal encoder is responsible for extracting useful time series features from the input transformer real-time multimodal time series signal data. These features can reflect the operating status and potential problems of the transformer. These time series features may include temperature change trends, vibration spectrum characteristics, current and voltage fluctuations, etc., which are crucial for diagnosing the health of the transformer.

[0034] The textual question is used to inquire about the transformer status. The transformer status diagnosis multimodal large model is a model adjusted based on the graph neural network, which can spatially fuse the features of different modes. The multimodal large model adopts the technical architecture of the graph neural network, which is a neural network model specially used to process graph structure data. Through the graph neural network, the large model can effectively spatially fuse the features from different modes (such as temperature, vibration, current and voltage, etc.), that is, represent and correlate these features in a unified space, so as to more accurately diagnose the status of the transformer.

[0035] S103, enabling the transformer state diagnosis multi-modal large model to output a transformer state diagnosis result.

[0036] After inputting the transformer's real-time multimodal time series signal data and textual questions, the multimodal big model will use its internal knowledge graph, modal encoder, graph neural network and other technical components to conduct in-depth analysis and processing of the input data. Ultimately, it will output a diagnosis result on the transformer's status, which may include transformer health status assessment, fault warning, maintenance suggestions, etc., to provide decision support for operation and maintenance personnel.

[0037] like Figure 2 As shown, it is a second flow chart of a transformer status diagnosis method of the present application, which may include: S201, build a knowledge graph.

[0038] Knowledge graph is a structured semantic knowledge base, which is mainly used to describe entities and their relationships in the real world, describe knowledge resources and their carriers through visualization technology, and mine, analyze, construct, draw and display knowledge and their connections. When used, it can reveal the dynamic development laws of the knowledge field through data mining, information processing, knowledge measurement and graphics drawing.

[0039] In this embodiment, the following method can be used to construct a knowledge graph: (1) Knowledge acquisition. The main purpose is to collect knowledge sources related to the target field, such as transformer status diagnosis. For example, literature, expert knowledge, historical data, technical manuals, etc. Through these knowledge sources, knowledge points related to transformer faults are extracted, such as fault type, monitoring indicators, fault causes, etc.

[0040] (2) Knowledge representation. The collected knowledge points are organized and stored in a standardized way for subsequent graph construction. Common forms of knowledge representation include: Entity (node): In the knowledge graph, an entity represents a concept or object in the real world. For the transformer fault knowledge graph, entities can include transformer components (such as windings, insulation materials), fault types (such as short circuit, overheating), monitoring indicators (such as temperature, current), fault causes (such as overload, aging), etc.

[0041] Attributes (properties of a node): Each entity can have multiple attributes, for example, a transformer winding can have attributes such as rated voltage, rated current, material, etc.

[0042] Relationships (edges): Relationships describe the association between entities. For example, "Winding" and "Short Circuit" can be connected through the "Fault Type" relationship, and "Overheating" and "Temperature Rise" can be connected through the "Symptom-Indicator" relationship.

[0043] (3) Data integration and fusion.

[0044] The goal of data integration is to unify data from different sources into the same graph structure, eliminate redundant information, and ensure data consistency. This can be achieved in the following ways: (3.1) Entity alignment Entities from different data sources may represent the same concept or object, and these entities need to be aligned and merged. For example, the concept of "winding" from different documents should be merged into one entity in the graph.

[0045] (3.2) Relationship alignment: Different data sources may describe similar or identical relationships, which also need to be merged and deduplicated in the graph.

[0046] (3.3) Disambiguation: Different expressions of the same entity or relationship need to be disambiguated through rules or algorithms to ensure the clarity and consistency of the graph structure.

[0047] (4) Knowledge graph storage. Knowledge graphs are usually stored in graph databases, such as Neo4j, ArangoDB, JanusGraph, etc. These databases are designed specifically for processing graph structured data and support efficient query and reasoning. In this embodiment, Neo4j is used to store the database.

[0048] S202, selection of large language model and modal encoder.

[0049] In this embodiment, the large model QWEN2 is used as the pre-trained large language model in the multimodal large model. It should be noted that the large model QWEN2 is the second-generation large language model developed by Alibaba Cloud. It is based on the Transformer architecture and can support multi-language processing. The large model QWEN2 uses optimization technologies such as Grouped Query Attention and DualChunk Attention to enhance the reasoning speed and long text processing capabilities. The open source nature of the large model QWEN2 provides extensive support for research and application.

[0050] Both attention mechanisms are technologies used in natural language processing and deep learning to optimize model performance and computational efficiency. Grouped query attention improves inference speed by grouping query heads while maintaining model accuracy. Dual-block attention processes long texts in blocks, reducing computation and memory usage and improving processing efficiency. Both technologies help improve model performance and user experience in practical applications.

[0051] The large model QWEN2 includes dense models and mixed expert models, with parameter sizes ranging from 500 million to 7.2 billion. Pre-training is completed on high-quality multilingual datasets, making it perform well in tasks such as language understanding, generation, and mathematical reasoning.

[0052] In this embodiment, the transformer real-time multimodal time series signal data includes multimodal time series signal data such as current and voltage time series signals, oil temperature time series signals, and oil chromatogram time series signals. The modal encoder can use Transformer to extract time series features and capture dynamic changes in multimodal time series signal data. By capturing dynamic changes, high-quality initial input can be provided for subsequent feature fusion.

[0053] S203, training the large language model.

[0054] (1) Knowledge graph embedding.

[0055] Convert entities and relationships in the knowledge graph into vector representations, that is, convert them into corresponding data for integration with large language models or use in other machine learning tasks. Graph embedding algorithms such as TransE, DistMult, ComplEx, etc. can be used to map entities and relationships in the knowledge graph into low-dimensional vector representations, which retain the semantic information and structural relationships in the knowledge graph. In this embodiment, the TransE algorithm can be used to implement knowledge graph embedding.

[0056] It should be noted that knowledge graph embedding is also called knowledge representation learning. Mapping entities and relationships in the knowledge graph into a low-dimensional continuous vector space helps capture the semantic information between entities and relationships, and thus supports various knowledge graph-related applications. The main purpose of knowledge graph embedding is to represent the triples (head entity, relationship, tail entity) in the knowledge graph with vectors so that these vectors can reflect the semantic relationship between entities and relationships. Through embedding, the originally discrete and sparse knowledge graph data is converted into a dense and continuous vector representation, which is convenient for subsequent calculations and reasoning.

[0057] (2) Fine-tune instruction set design.

[0058] (2.1) Input instructions: Combine the embedding vectors of the knowledge graph with the text input of the large language model to generate task-related input instructions. For example, provide monitoring data to the QWEN2 model, such as "the winding temperature is 85°C and the current is 120A". And add instructions: "According to the knowledge graph, check whether the temperature and current of the winding are abnormal, and diagnose with reference to the 'overheating' fault type and the corresponding 'temperature increase' relationship." (2.2) Conditional rules: Set conditional instructions based on the relationships and rules in the knowledge graph to guide the large language model on how to process input in specific situations. For example, add a rule: "If the "temperature" in the monitoring data exceeds 80°C, and there is a direct relationship between "winding" and "overheating" in the knowledge graph, the model should give priority to the "overheating" fault." (2.3) Output constraints: Set the output instruction set to ensure that the output of the large language model conforms to the known knowledge in the knowledge graph. For example: "If the diagnosis result is "winding overheating", then output the fault type "overheating" and prompt the duration of monitoring "temperature increase"." (2.4) Output truth value: Set the truth value of the fault category. For example, for the input of "winding temperature is 85°C, current is 120A", the corresponding fault category is "overheating".

[0059] S204, fine-tuning the large multimodal model for transformer state diagnosis based on graph neural network.

[0060] After training, the large language model only has certain knowledge and can perform language question and answer. It cannot make diagnoses based on the dynamically changing data features in the captured signal. It needs to be fine-tuned first and cannot directly process multimodal data.

[0061] After extracting text features (also a modality) and features of other modalities (features extracted based on time series signal data), the multimodal large model then builds a graph structure based on these features. Each node in the graph structure represents a feature vector of a modality, and the edges between nodes represent the similarity or correlation between features of different modalities. By calculating the cosine similarity or Euclidean distance between these feature vectors, the weights of the edges are generated, thus forming a graph structure of multimodal features. This graph structure allows information to be shared effectively between different modalities, thereby enhancing the model's ability to understand complex data.

[0062] After the graph structure is established, a Graph Neural Network (GNN) can be used for feature propagation. The role of GNN is to update the feature representation of each node through information transmission between nodes and edges. During the GNN processing, each node will not only retain its original features, but also integrate the information transmitted from neighboring nodes to generate a richer feature representation. Specifically, the feature propagation process of GNN is usually divided into the following steps: (1) Information aggregation: Each node receives information from its neighboring nodes. This information includes the characteristics of the neighboring nodes and the weights of the edges connecting these nodes. The information can be aggregated in a simple weighted sum or in a more complex nonlinear transformation.

[0063] (2) Feature update: The aggregated information is combined with the features of the current node, and the feature representation of the node is updated after being processed by a nonlinear activation function. This process ensures that the features of each node not only depend on its own data, but also reflect the modal relationship in its neighborhood.

[0064] (3) Repeated propagation: GNN usually updates node features through multiple layers of propagation. As the number of propagation layers increases, each node can integrate information from nodes farther away, and the generated feature representation will be more comprehensive.

[0065] Through these steps, GNN can effectively capture and fuse the feature information of different modalities to generate a unified feature representation. In the fine-tuning process of the multimodal large model, this unified feature representation can fully utilize the potential capabilities of the multimodal large model, making it more flexible and efficient in processing multimodal data. This method not only improves the performance of the multimodal large model, but also ensures that the deep associations between different modalities are fully expressed, making the multimodal large model more adaptable and accurate in practical applications.

[0066] In the fine-tuning process, in order to prevent the multimodal large model from forgetting the previously learned knowledge, the elastic weight consolidation (EWC) technology can also be introduced. When fine-tuning the multimodal large model, EWC imposes constraints on the important parameters of the multimodal large model by adding regularization terms to the loss function, thereby preventing these parameters from changing significantly. This method ensures that the multimodal large model does not forget important knowledge in old tasks when adapting to new tasks, and maintains its overall performance in multiple tasks.

[0067] S205, diagnose the transformer status through a multi-modal large model.

[0068] Based on the existing technology, the application of transformer status diagnosis mainly faces the problems of strong data dependence, difficulty in feature extraction, weak ability to identify new faults, lack of domain knowledge integration, and insufficient model interpretability. This application integrates knowledge from multiple sources to build a standardized knowledge graph, and uses the domestic large model QWEN2 to process multimodal time series signals and extract key features. At the same time, the graph embedding algorithm and graph neural network are used to realize the fusion and propagation of multimodal features, and combined with the elastic weight consolidation technology, it ensures that the large model maintains the memory of old knowledge during the fine-tuning process, thereby enhancing the understanding and diagnosis capabilities of complex data.

[0069] Based on the above transformer status diagnosis method, such as Figure 3 As shown, it is a schematic diagram of a transformer status diagnosis system of the present application, which may include: An acquisition module is used to acquire real-time multi-modal time series signal data of the transformer; An input module is used to input transformer real-time multimodal time series signal data and textual questions into a transformer state diagnosis multimodal large model; the transformer state diagnosis multimodal large model includes a large language model and a modal encoder, the knowledge source of the large language model is a knowledge graph based on transformer state diagnosis, and the modal encoder is used to extract time series features from the transformer real-time multimodal time series signal data; the textual questions are used to inquire about the transformer state; the transformer state diagnosis multimodal large model is a model adjusted based on a graph neural network, and can perform spatial fusion of features of different modalities; The output module is used to enable the transformer state diagnosis multi-modal large model to output the transformer state diagnosis result.

[0070] It should be noted that in the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic. For example, the division of each module is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another device, or some features can be ignored or not executed. The module described as a separate component may or may not be physically separated. The component displayed as a module may be a physical unit or multiple physical units, that is, it may be located in one place, or it may be distributed in multiple different places. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0071] In addition, each module in each embodiment of the present invention may be integrated into a processing unit, each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0072] The present application also proposes a computer program product, which includes instructions, and when the instructions are executed by a processor, the steps in the above transformer state diagnosis method embodiment are implemented. Alternatively, when the processor executes the computer program, the functions of each module / unit in the above system embodiment are implemented.

[0073] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in a memory and executed by the processor to accomplish the present invention.

[0074] The carrier for implementing the above-mentioned computer program product may be a computer device, or the computer program product may be stored in a computer storage medium.

[0075] The computer device may be a desktop computer, a notebook computer, a palm computer, a cloud server, etc. The computer device may include, but is not limited to, a processor and a memory.

[0076] 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, discrete gate or transistor logic devices, discrete hardware components, etc.

[0077] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0078] If the module / unit integrated in the computer 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, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. 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 capable of carrying 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), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0079] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A transformer status diagnosis method, characterized in that: include: Obtain real-time multi-modal time series signal data of transformer; Input the transformer real-time multi-modal time series signal data and textual problems into the transformer status diagnosis multi-modal large model; The transformer state diagnosis multimodal large model includes a large language model and a modal encoder. The knowledge source of the large language model is a knowledge graph based on transformer state diagnosis. The modal encoder is used to extract time series features from the real-time multimodal time series signal data of the transformer. The textual question is used to inquire about the transformer state. The transformer state diagnosis multimodal large model is a model adjusted based on a graph neural network, which can spatially fuse features of different modalities. The transformer state diagnosis multi-modal large model is enabled to output transformer state diagnosis results.

2. A transformer status diagnosis method according to claim 1, characterized in that: The method for constructing a knowledge graph based on transformer status diagnosis includes: Collect knowledge sources related to transformer status diagnosis, and extract knowledge points related to transformer faults from the knowledge sources; Organizing and storing knowledge points in a standardized manner to form knowledge representation; the knowledge representation includes entities, attributes and relationships, where entities are used to represent concepts or objects in the transformer, attributes are used to represent the characteristics of entities, and relationships are used to represent the associations between entities; The entities and relations in the knowledge representation are aligned, and ambiguities in the entities and relations are eliminated to obtain a processed knowledge graph.

3. A transformer status diagnosis method according to claim 2, characterized in that: After obtaining the processed knowledge graph, the method further includes: The processed knowledge graph is stored in the Neo4j graph database.

4. A transformer status diagnosis method according to claim 3, characterized in that: The large language model adopts the QWEN2 large model, and the modal encoder adopts the Transformer model.

5. A transformer status diagnosis method according to claim 3, characterized in that: The training method of the large language model includes: Map the entities and relations in the knowledge graph based on transformer condition diagnosis into low-dimensional vector representations to obtain an embedding vector; generating an input instruction associated with a transformer status diagnosis task in combination with the embedded vector; Conditional instructions are set based on the relationships in the knowledge graph based on transformer status diagnosis to guide the large language model on how to process the input under corresponding conditions; Setting an output instruction set, wherein the output instruction set is used to regulate the output of the large language model to conform to known knowledge in the knowledge graph based on transformer state diagnosis; Set the true value of the transformer fault category to enable the large language model to output content in a preset format; The large language model is trained by inputting instructions, conditional instructions, and output instruction sets combined with true values.

6. A transformer status diagnosis method according to claim 5, characterized in that: The mapping of entities and relationships in the knowledge graph based on transformer status diagnosis into low-dimensional vector representations includes: A graph embedding algorithm is used to map the entities and relations in the knowledge graph based on transformer condition diagnosis into low-dimensional vector representations.

7. A transformer status diagnosis method according to claim 6, characterized in that: The graph neural network-based adjustment includes: Constructing a graph structure, in which each node represents a feature vector of a modality, and edges between the nodes represent similarities or correlations between feature vectors of different modalities; By calculating the cosine similarity or Euclidean distance between feature vectors, the edge weights are generated to obtain the graph structure of multimodal features; Use graph neural networks to propagate features of the graph structure of multimodal features and generate unified feature representations; The large multimodal model for transformer condition diagnosis is adjusted through unified feature representation.

8. A transformer status diagnosis method according to claim 7, characterized in that: The graph neural network-based adjustment also includes: introducing elastic weight consolidation technology.

9. A transformer status diagnosis system, characterized in that: include: An acquisition module is used to acquire real-time multi-modal time series signal data of the transformer; An input module is used to input transformer real-time multimodal time series signal data and textual questions into a transformer state diagnosis multimodal large model; the transformer state diagnosis multimodal large model includes a large language model and a modal encoder, the knowledge source of the large language model is a knowledge graph based on transformer state diagnosis, and the modal encoder is used to extract time series features from the transformer real-time multimodal time series signal data; The textual question is used to inquire about the transformer status; the transformer status diagnosis multimodal large model is a model adjusted based on the graph neural network, which can spatially fuse the features of different modes; The output module is used to enable the transformer state diagnosis multi-modal large model to output the transformer state diagnosis result.

10. A computer program product, characterized in that: The computer program product comprises instructions, and when the instructions are executed by a processor, the transformer state diagnosis method according to any one of claims 1 to 8 is implemented.

Images