Transformer equipment fault diagnosis method, system, equipment and medium

Abstract

The invention discloses a transformer equipment fault diagnosis method and system, equipment and a medium, and relates to the technical field of power equipment detection.The method comprises the steps that various operation fault cases of transformer equipment are integrated, and an initial question-answer pair sample set is constructed and input into a preset large model for training; a preset dynamic knowledge extractor is introduced in the training process to carry out text entity extraction on the initial question and answer pair sample set, and an equipment fault map is generated; converting an equipment fault map vector, and inputting the converted equipment fault map vector into an attention layer reconstructed by rotating position coding to capture an equipment coupling relationship; performing parameter fine tuning optimization on an output result of the attention layer, and iterating a training process to obtain a trained fault diagnosis large model; and acquiring operation data of transformer equipment in real time, inputting the operation data into the large fault diagnosis model, and generating a high-precision fault diagnosis result and a maintenance strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment detection, and in particular to a transformer equipment fault diagnosis method, system, equipment and medium. Background Art

[0002] Transformers are the core hub and stable pillar of the power system. Their operating status directly determines the grid's safety margins, power quality, and energy transmission efficiency. In the context of new power system construction, transformers, as key hubs for energy conversion and voltage regulation, have become a crucial technical defense for ensuring the safety and resilience of large power grids.

[0003] Traditional fault diagnosis methods rely on manual experience, which is not only time-consuming and labor-intensive, but also inefficient and highly subjective. This leads to large errors in the detection results, making it difficult to formulate effective and reasonable maintenance strategies, thus affecting the stable operation of power equipment.

[0004] It can be seen that how to effectively diagnose transformer equipment and improve the accuracy of fault judgment has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0005] The present invention provides a transformer equipment fault diagnosis method, system, equipment and medium, which solves the problem of how to improve the accuracy of fault diagnosis by optimizing the training process of a large model.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a transformer equipment fault diagnosis method, comprising: Integrate and analyze various operational failure cases of transformer equipment in the target power system to construct an initial set of question-answer pair samples; The initial question-answer pair sample set is input into a preset large model for training. During the training process, a preset dynamic knowledge extractor is introduced to extract text entities from the initial question-answer pair sample set to generate a device fault map; the dynamic knowledge extractor is composed of a distillation mechanism and a graph neural network fusion; The device fault map vector is transformed and input into the attention layer reconstructed by the rotation position encoding to capture the device coupling relationship; The output of the attention layer is fine-tuned and optimized, and the training process is iterated to obtain a trained fault diagnosis model. The operating data of the transformer equipment is collected in real time and input into the fault diagnosis model to generate fault diagnosis results and maintenance strategies.

[0007] Furthermore, the integration and analysis of various operational failure cases of transformer equipment in the target power system constructs an initial set of question-answer pair samples, including: Collect transformer operating procedures, transformer test specification data, and historical transformer failure data, annotate them with questions and answers, and integrate them into operating failure cases; A generative adversarial network is used to perform sample enhancement on the running failure cases to construct the initial question-answer pair sample set.

[0008] Furthermore, the process of generating the equipment fault map includes: Each question-answer sample in the initial question-answer pair sample set is converted into a sequence and then input into a preset large model. A distillation mechanism is introduced to dynamically adjust the weights of transformer equipment operating indicators to generate a fault knowledge vector; The driven graph neural network extracts the text entities of the fault knowledge vector to construct node features and edge features, and generates the equipment fault map.

[0009] Furthermore, the device fault map vector is transformed and input into the attention layer reconstructed by the rotational position encoding to capture the device coupling relationship, including: Input the vector representation of the device fault map into the attention layer and perform feature normalization; The normalization processing result is mapped to a rotation angle using a learnable frequency parameter; wherein the rotation angle is designed to dynamically modify the rotation angle weight using a gated network under a preset indicator threshold condition; Perform weighted analysis on the rotation angles to generate a rotation matrix; During attention calculation, the rotation matrix is used to perform encoding reconstruction on the query matrix and the key-value matrix to generate a reconstructed attention matrix to capture the device coupling relationship.

[0010] Furthermore, the process of parameter fine-tuning and optimization includes: Deploy an expert routing network to make probability predictions for each eigenvector in the reconstructed attention matrix, and use eigenvectors with probability values exceeding the preset conditions as activation experts; The DoRA fine-tuning mechanism is introduced to decompose the original weights of the activation experts into amplitude and direction components; A low-rank update is performed on the direction component, and the updated direction component is merged with the magnitude component to optimize training.

[0011] Furthermore, the process of fine-tuning the parameters of the output of the attention layer and decoding and iteratively training the results also includes: During the iterative training process, the greedy algorithm is integrated into the output layer of the preset large model to select the result corresponding to the maximum prediction probability from the output probability distribution as the target output result.

[0012] Furthermore, after collecting the operating data of the transformer equipment in real time and inputting it into the fault diagnosis model to generate the fault diagnosis results and maintenance strategy, the method further includes: Feedback the execution status of the maintenance strategy to the monitoring terminal; The equipment fault map is updated according to the feedback results, and the fault diagnosis model is iteratively optimized.

[0013] Another embodiment of the present invention provides a transformer equipment fault diagnosis system, comprising: The training sample construction module is used to integrate and analyze various operational failure cases of transformer equipment in the target power system and construct an initial set of question-answer pair samples; A text entity extraction module is used to input the initial question-answer pair sample set into a preset large model for training. During the training process, a preset dynamic knowledge extractor is introduced to extract text entities from the initial question-answer pair sample set to generate an equipment fault map; the dynamic knowledge extractor is composed of a distillation mechanism and a graph neural network fusion; An attention reconstruction module is used to transform the device fault map vector and input it into the attention layer reconstructed by the rotational position encoding to capture the device coupling relationship; The iterative training module is used to fine-tune and optimize the parameters of the output results of the attention layer and iterate the training process to obtain a trained fault diagnosis model; The fault diagnosis module is used to collect the operating data of the transformer equipment in real time and input it into the fault diagnosis model to generate fault diagnosis results and maintenance strategies.

[0014] Yet another embodiment of the present invention provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the transformer equipment fault diagnosis method described above when executing the computer program.

[0015] Yet another embodiment of the present invention provides a computer-readable storage medium storing a computer program. When a device containing the computer-readable storage medium executes the computer program, the transformer equipment fault diagnosis method described above is implemented.

[0016] Compared with the prior art, the embodiments of the present invention have the following advantages: The present invention introduces a generative adversarial network to generate rare fault samples, constructing a question-and-answer sample database containing a large amount of transformer fault diagnosis expertise, which can more comprehensively cover the fault types of transformer equipment and provide sufficient data input for model training. In the process of training the model, a dynamic knowledge distillation mechanism is adopted to construct a domain knowledge extractor based on a graph neural network, which improves the efficiency of knowledge transfer while ensuring the accuracy of model recognition. The hybrid optimization mechanism of multi-head self-attention, rotation encoding and expert routing + fine-tuning is introduced to greatly reduce the computational complexity and false alarm rate of model training. Finally, the trained fault diagnosis model is used to diagnose the transformer equipment, thereby formulating accurate maintenance strategies to maintain the stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic flow chart of a transformer equipment fault diagnosis method in one embodiment of the present invention; Figure 2 Schematic diagram of the encoding and reconstruction architecture of the attention mechanism in one embodiment of the present invention; Figure 3 is a schematic diagram of a parameter fine-tuning optimization architecture in one embodiment of the present invention; Figure 4 is a schematic diagram of a component decomposition architecture in parameter fine-tuning optimization in one embodiment of the present invention; Figure 5 is a schematic diagram of the evaluation results of the model maintenance strategy generation capability in one embodiment of the present invention; Figure 6 It is a schematic structural diagram of a transformer equipment fault diagnosis system in one embodiment of the present invention; Figure 7 A structural block diagram of a preferred embodiment of a computer device provided by the present invention; Figure numerals: training sample construction module M1, text entity extraction module M2, attention reconstruction module M3, iterative training module M4, fault diagnosis module M5. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0019] In the description of this application, the terms "first," "second," "third," etc. are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first," "second," "third," etc. may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0020] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meanings as those commonly understood by those skilled in the art. The terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. Those of ordinary skill in the art will understand the specific meanings of the above terms in this application in specific circumstances.

[0021] An embodiment of the present invention provides a method for diagnosing transformer equipment faults. For details, see Figure 1 , Figure 1 In one embodiment of the present invention, the following steps are included: S1. Integrate and analyze various operational failure cases of transformer equipment in the target power system to construct an initial set of question-answer pair samples.

[0022] Transformer operating procedure data, transformer test specification data, and historical transformer failure data from the target power system are collected and annotated as question-answer pairs. These pairs are then integrated into operational failure cases required for model training. Specifically, this transformer operating procedure data, transformer test specification data, and historical transformer failure data can include records of transformer defects and abnormal parameter values from the past three years, or oil chromatography reports. In other words, the integrated operational failure cases can describe the type, cause, and status of transformer failures.

[0023] For example, in some embodiments of the present invention, the number of characters collected from the transformer operation and maintenance manual is 1.4× The data on transformer operating procedures, such as daily operation, monitoring, maintenance and overhaul of transformers, were collected; transformer test specification data were collected from the IEC 60076 series of standards or the GB 1094 test specification series of power transformer standards; 2,318 typical transformer failure cases and defect records were selected from the power systems of power supply bureaus in multiple locations, including historical transformer failure data such as 521 bushing failures, 732 body failures, 496 protection device failures, 276 voltage regulating device failures and 251 cooling device failures.

[0024] Bushing failures may include: water seepage and contamination caused by moisture or other contaminants entering the inside or surface of the bushing, or insulation breakdown caused by the formation of a conductive path due to failure of the insulating material; body failures include damage to the casing caused by mechanical damage or corrosion to the oil tank; protection device failures commonly include malfunction or misoperation caused by the inability of the relay to operate correctly; voltage regulating device failures include: mechanical failures caused by failure of the mechanical parts of the tap changer, such as the switch rod and drive mechanism; cooling device failures include failure of the oil pump in the oil cooling system, resulting in increased oil temperature.

[0025] At the same time, this embodiment also provides a specific example of the following question-answer pair, i.e., a QA pair: Q: What standards should be followed for the maintenance of transformer on-load tap-changers? A: Maintenance of transformer on-load tap-changers should be performed in accordance with the manufacturer's specifications. If no manufacturer's specifications are available, maintenance can be performed in accordance with the relevant standard DUT574. Maintenance work includes removing oxide films and oil contamination, measuring the DC resistance of the windings, and recording the tap change status. By performing maintenance in accordance with the standard, the normal operation and use of the transformer tap-changer can be ensured.

[0026] A generative adversarial network (CGAN) is used to enhance the operational failure cases to construct an initial set of question-answer pairs. This step actually uses the GAN to synthesize rare failure samples to expand the question-answer pairs. For example, in this embodiment, the number of question-answer samples for cooling device failures is the smallest, so a GAN can be used to enhance these data. The GAN generator and discriminator include the following formulas: Where, is the positive slope coefficient, x is the discriminator input sample, y is the clear transformer fault type condition label, z is the noise vector, is the Sigmoid function, W k is the weight matrix of the kth layer, is the normalization function, Concat is the concatenation function, G is the mathematical expression formula of the generator, and D is the mathematical expression formula of the discriminator.

[0027] Furthermore, CGAN will introduce conditional constraints on this basis, namely the objective function V, which is expressed as follows: Where, P data is the real data of transformer fault, P z is the noise distribution data, is the expectation of the real data distribution, is the expectation of the noise distribution.

[0028] In this embodiment, the conditional information "y" introduced by the CGAN can be a clear transformer fault type label, such as a bushing fault, a transformer body fault, a cooling device fault, a voltage regulator fault, or a protective device fault. During the generation phase, the generator G concatenates random noise z with y to generate fault text, such as a simulated cooling system fault sample: "A fan failed to start during operation in a 220kV transformer, causing the oil temperature to rise by 5°C, triggering a cooling system fault alarm." The discriminator D then evaluates whether the input sample meets the specified conditional label, thereby training the generator to produce high-fidelity, fault-specific question-and-answer data.

[0029] S2: The initial set of question-answer pairs is fed into a pre-set large model for training. During the training process, a pre-set dynamic knowledge extractor is introduced to extract text entities from the initial set of question-answer pairs, generating a device fault map. The dynamic knowledge extractor is composed of a fusion of a distillation mechanism and a graph neural network.

[0030] First, each question and answer sample in the generated initial question and answer sample set needs to be preprocessed, such as data cleaning, and converted into a text vector format. Then, it is embedded in a preset large model for training. The preferred large model in this embodiment is the Qwen-7B-Chat open source large model.

[0031] A dynamic knowledge extractor, consisting of a distillation mechanism and a graph neural network, is first introduced within the large model to optimize the efficiency of knowledge transfer during training. Specifically, this embodiment uses a distillation mechanism to dynamically adjust the weights of the transformer equipment operating indicators reflected by the vector to generate a fault knowledge vector.

[0032] For example, three types of teacher models trained using this distillation mechanism are processed in parallel: an electrical characteristics model, a mechanical state model, and a chemical analysis model. In this example, the electrical characteristics model outputs the probability distribution of transformer discharge / overheating, the mechanical state model outputs the degradation score of transformer components, and the chemical analysis model outputs the correlation of oil cracking products. A gating network then dynamically assigns weights to the teacher models based on text keywords. For example, the weight of the chemistry teacher increases to 0.82 when "acetylene" appears. Finally, these weighted combinations are combined to form a fault knowledge vector.

[0033] It should be understood that in the process of adjusting the weights, this embodiment also uses a cosine annealing strategy to adjust the loss weights generated by the distillation mechanism. The specific formula is as follows: In the formula, is the distillation loss weight, t is the current training step, and T is the total number of steps. This adjustment strategy can enable the model to gradually transition to autonomous optimization after relying on supervision in the early stage.

[0034] Next, the graph neural network extracts textual entities from the fault knowledge vector, constructs node and edge features, and generates a device fault map. For example, from the sentence "Fan stall causes oil temperature to rise," two entities, "cooling fan" and "oil temperature," are extracted. The relationship between them is "cause of rise," forming a knowledge graph structure centered on "component - fault type - causal relationship." The graph neural network uses these entities and relationships to construct a node-edge graph structure. Node representations are aggregated through multiple layers to obtain contextual features, which are ultimately embedded into vector form and fed into the attention mechanism module of the larger model.

[0035] S3. The device fault map vector is transformed and input into the attention layer reconstructed by the rotation position encoding to capture the device coupling relationship.

[0036] This step aims to introduce a rotational position encoding mechanism, called Trans-RoPE, into the multi-head self-attention layer of the large model. This mechanism captures the multi-dimensional coupling relationships between various indicators, such as transformer oil temperature, load, and operating time, to generate the final attention output. Specifically, the attention layer inputs a vector representation of the device fault map, performs feature normalization, and maps the normalized results to rotation angles using a learnable frequency parameter.

[0037] For example, the equipment fault map is represented by vector (dimension 16×1024×4096), and the vector is accompanied by M dynamic indicators / features, such as oil temperature t i , load l i , running time h i , normalize these three indicators, if the learnable frequency parameter defined by each dimension of the vector is , then the process of mapping the normalized index to the rotation angle using the learnable frequency parameter can be expressed as follows: Where, is the rotation angle, reflecting the dynamic characteristics of indicator d at position i; is the normalized result of the indicator.

[0038] In this embodiment, the rotation angle is designed to dynamically modify the rotation angle weight using a gated network under a preset indicator threshold condition. For example, the frequency parameter can be learned. =0.35 maps the oil temperature rotation angle θ=1.7×0.35≈0.6 radians. When the gated network detects that the oil temperature change rate is 8°C / h, which exceeds the threshold (the preset oil temperature threshold is 2°C / h), the oil temperature weight α is increased from 0.2 to 0.6, thereby enhancing the large model's attention to this oil temperature indicator.

[0039] Furthermore, Trans-RoPE can dynamically assign location information weights to each indicator at different time points. For example, “before a sudden failure, the weight of oil temperature increases and the load decreases.”

[0040] Next, the rotation angle weights are analyzed to generate a rotation matrix. During attention calculation, this rotation matrix is used to reconstruct the query matrix and key-value matrix in the attention matrix to generate a reconstructed attention matrix that captures the device coupling relationship. For example, it can be captured that in the oil temperature sudden change scenario, the value of the 128th dimension (oil temperature-load coupling coefficient) in the output feature jumps from 0.45 to 0.91. For details, please refer to Figure 2 As shown in the figure, it can be seen that the example illustrates the operation of a 750kV transformer experiencing a heavy gas fault. The reconstruction process of the attention mechanism in this embodiment actually involves performing a Hadamard product operation on each dimension of the query matrix and the key-value matrix with the rotation matrix. This operation enables the large model to accurately capture coupling effects such as "at a load factor of 90%, every 1°C increase in oil temperature corresponds to a 0.8% acceleration in insulation aging," providing explicit physical perception features for subsequent diagnosis.

[0041] S4. Fine-tune and optimize the parameters of the output results of the attention layer and iterate the training process to obtain a trained fault diagnosis model.

[0042] In this step, the deep features output by the attention layer will enter the hybrid fine-tuning decoding stage. Specifically, the MoE-DoRA hybrid fine-tuning framework, i.e., the expert-fine-tuning hybrid framework, is adopted in the decoding layer of the large model.

[0043] In the MoE (Expert Routing Network) component, an expert routing network is deployed to perform probabilistic predictions on each eigenvector in the reconstructed attention matrix output by the attention system. Eigenvectors whose probabilities exceed a preset condition are activated as experts. Specifically, the routing network M generates expert selection probabilities based on the input eigenvector x and performs weighted aggregation using a gating function.

[0044] In this process, for example, the probability prediction result of the discharge fault feature is 1.1, which exceeds the preset threshold condition 1.0. At this time, the routing network M will activate the discharge fault expert E3, and its activation weight is 0.73. In general, when the large model predicts the probability of various fault features as follows: "Casing fault: 0.83, cooling system abnormality: 0.12, protection device fault: 0.05", when the predicted probability of a certain type of fault is higher than the threshold (such as 0.70), the fault judgment is triggered, and the corresponding maintenance strategy is generated by the currently activated expert sub-network. For details, please refer to Figure 3 As shown on the far right, it can be seen that the routing component FFNN selects a specific activation expert FFNN1 by outputting the probability.

[0045] At the same time, in the decoding layer, the DoRA fine-tuning mechanism is introduced to work in parallel with the expert routing network. In this embodiment, each expert network corresponds to an independent DoRA low-rank adapter, which is responsible for decomposing the original weights of the activated expert into amplitude components and direction components. The DoRA fine-tuning mechanism performs a low-rank update on the direction component and merges the updated direction component with the amplitude component of the original weight to optimize training.

[0046] For example, combined Figure 3 The middle frame and Figure 4 As shown, the DoRA low-rank adapter decomposes the weight matrix W of the input activation expert to obtain the magnitude component of the weight: ||W||c∈R 1×n and the directional component of the weight: V∈R m×n .

[0047] In the process of low-rank update, the direction component V is adjusted using low-rank matrices A and B to generate an updated ΔV = A × B. The adapted direction is V + ΔV. The adapted weight is merged with the amplitude component to obtain a new merged weight W', which is expressed as follows: In the collaborative parallel training process, the routing network and the adapter jointly optimize the parameter loss, where the loss function L includes the task loss L task With expert load balancing loss L balance , which is expressed as follows: The second term in the formula is L balance The representation of p i is the historical call frequency of expert routing network i, λ is the balance coefficient, and N is the number of expert routing networks.

[0048] Fully implement the above entity extraction, attention reconstruction and parameter fine-tuning process in this step, and iteratively train the large model until the error meets the requirements. Figure 3 As can be seen on the far left, in this embodiment, the output layer of the large model is configured as a combination of a normalization layer and a fully connected layer. During iterative training, a greedy algorithm is preferably integrated into the output layer of the large model to select the result corresponding to the maximum predicted probability from the output probability distribution as the target output. Specifically, the large model converts the output result into a probability distribution through a function, and then uses the greedy algorithm to select the word with the highest probability from the probability distribution as the final prediction result.

[0049] For example, the preferred normalization layer in this embodiment is RMS Norm, while the fully connected layer introduces a gating mechanism by adding an activation function, SwiGLU, to enable the large model to select different activation modes in different input regions. Finally, a linear layer extracts features and outputs the prediction results. It is worth noting that verification and evaluation are required after each round of iterative training. The results are fed back to the CGAN rare fault generation stage and the parameter fine-tuning stage to correct and optimize the large model parameters and improve the model output accuracy.

[0050] Specifically, this invention will provide a specific large model training example to describe the above steps in detail: 1. The number of characters collected in this embodiment is 1.4× The number of question-answer pairs for transformer operating procedure data was 17,526; the number of characters in transformer test specification data was 9.9×106, and 13,148 pairs of question-answer pairs were constructed; the number of characters in transformer fault cases and defect records was 1.3×107, and 15,974 pairs of question-answer pairs were constructed. The above question-answer pairs were randomly split into a training set with a ratio of 70%, a validation set with a ratio of 15%, and a test set with a ratio of 15%. The training set was converted into an input sequence index (QA question-answer pair sample) with a maximum sampling data of 100,000 and a training data set length truncated to 1024 after data cleaning and generative adversarial network preprocessing. 2. The input sequence index undergoes data cleaning and preprocessing. A dynamic knowledge distillation mechanism using a cosine annealing strategy is employed, and the distillation loss weights are adjusted. A domain knowledge extractor based on a graph neural network is constructed to automatically extract fault entities from Q&A pairs. The output sequence is processed through an embedding layer into integer identifiers of varying lengths. The embedding matrix then converts these integer identifiers into a vector sequence as the initial input for the large model. In this example, the Qwen-7B-Chat open-source large model is selected for improvement and training.

[0051] 3. The vector sequence constructed in step 2 is trained using multi-level optimized Trans-RoPE rotation encoding. Rotational transformations are introduced into the query and key vectors to capture positional information during attention calculations. Backpropagation is used to jointly optimize parameters, allowing the large model to adaptively adjust the importance of different dimensions. This captures complex relationships such as the coupling effect of oil temperature and load and the cumulative impact of run time. The relative positional relationships between elements in the sequence are then injected into the vector input of the large model's decoding layer. 4. The decoding layer adds a low-rank adapter layer with DoRA rank 8 between the fully connected layer and the attention layer, freezes all parameters of the pre-trained model, and sets the learning rate to 5e -5 ,The computational model is fp16, and the parameters of the adapter layer added for training,fault information is extracted from unstructured text data and continuously updated.,The MoE-DoRA framework realizes fine-grained adaptation of fault scenarios through a dynamic routing mechanism,forming knowledge-driven, and the gradient accumulation is 16; 5. Repeat steps 1-4. After completing one round of training, evaluate the model's performance on the validation set to prevent overfitting and determine whether to stop training. After 10 rounds of training, continuously update the model's weights and biases. 6. Test the trained large model on the test set. Specifically, for ease of evaluation, three evaluation metrics were introduced: precision, recall, and F1 value. The model's transformer equipment fault type and fault status diagnosis results were evaluated. BLEU-4 (a machine translation evaluation metric) and ROUGE-1, ROUGE-2, and ROUGE-L (recall-based text summarization evaluation metrics) were also introduced to evaluate the transformer equipment fault maintenance strategy generation results.

[0052] To verify whether the trained large model, fine-tuned through MoE-DoRA, has formed more knowledge links and effectively improved vertical domain expertise question answering, this example selected 70 different transformer equipment failure scenarios and used a scoring system to evaluate the trained large model's ability to generate transformer fault maintenance strategies. In this example, the improved and trained large model is named TransQwen.

[0053] Test results see Figure 5 The TransQwen model can generate accurate maintenance strategies for transformer equipment faults, reducing the ambiguity in operation. The generated maintenance strategies are comprehensive, detailed, safe, and systematic. Figure 5As can be seen, the TransQwen model achieved BLEU-4, ROUGE-1, ROUGE-2, and ROUGE-L metrics of 79.86, 86.18, 81.92, and 84.58, respectively. The total score (SUM) and average score (AVG) were 89.94 and 88.65, respectively. This demonstrates that the TransQwen prediction model trained in this example significantly outperforms other models (the traditional Qwen-7B-Chat and a pre-trained general Qwen model), enabling relatively accurate diagnosis of transformer equipment faults.

[0054] S5. Collect the operating data of the transformer equipment in real time and input it into the fault diagnosis model to generate fault diagnosis results and maintenance strategies.

[0055] After the trained large model is deployed, multi-source data from transformer equipment is collected in real time, such as structured monitoring data: oil chromatography gas concentration, infrared temperature measurement maps, load curves, etc., as well as unstructured text records: inspection descriptions, abnormal sounds, etc. This data is standardized to generate a unified vector representation, which is input into the trained fault diagnosis model for diagnosis. The model goes through various layers of operations and ultimately outputs the fault diagnosis results and corresponding maintenance strategies.

[0056] An example fault diagnosis includes the following: Fault Type: Insulating oil is flowing downward around the upper edge of the porcelain bushing on the high-voltage side of Phase A of the No. 1 main transformer in a 110kV substation. Component Attributes: The No. 1 main transformer is an oil-immersed, self-cooled transformer. The 110kV side bushing is a high-voltage oil-paper capacitor type with a capacitance of 308pF, a creepage distance of 3906mm, and a weight of approximately 80kg.

[0057] The corresponding maintenance strategy could be: 1. Power outage: Immediately notify the relevant authorities and obtain power outage permission to ensure the transformer is completely de-energized to ensure the safety of maintenance personnel. 2. On-site isolation: Identify the leak location and promptly isolate the faulty area. 3. Oil draining: Drain the leaked oil and replace the insulating oil. 4. Bushing removal: Check for loose connections between the oil conservator and the porcelain bushing flange on the phase A bushing. Remove the bushing and confirm whether the connection between the aluminum cable tube and the tapered porcelain bushing at the bottom of the bushing is broken. 5. Replace the high-voltage bushing: Ensure all components are tightly connected and use an extended high-voltage side cable guide for installation. 6. Equipment testing: Conduct infrared temperature retests to ensure all components are operating normally and there are no signs of overheating or leakage. Conduct comprehensive infrared temperature checks on transformers of the same model. Also, test the dielectric strength, moisture content, and other indicators of the insulating oil to ensure oil quality is acceptable, and conduct a power-on test. 7. Regular inspection and maintenance: Conduct infrared testing in accordance with the "Condition-Based Inspection and Testing Procedures for Power Transmission and Transformation Equipment," especially increasing the frequency of testing during high-temperature and high-load operation. Pay attention to checking the bushing oil level. Strengthen follow-up monitoring of oil chromatography and high-voltage testing for equipment with family-specific defects, and promptly overhaul or replace equipment with serious problems. 8. Training and Improvement: Strengthen professional training for maintenance personnel, update maintenance manuals and operating procedures, and ensure the standardization and effectiveness of maintenance work. At the same time, strengthen management of the entire process of transformer selection, ordering, acceptance, and commissioning to ensure the use of reliable products.

[0058] It should be understood that in actual applications, the execution status of maintenance strategies can be further fed back to the monitoring terminal for staff to understand. This feedback can then be used to update the equipment fault map and iteratively optimize the fault diagnosis model. For example, if the fault type diagnosed is a cooling oil line blockage, the backup oil pump is started and cleaned, and the feedback result shows that the flow rate has recovered to 95% after cleaning, the knowledge map can be updated accordingly to speed up the handling of the same fault the next time it occurs.

[0059] In summary, the embodiment of the present invention constructs a fault diagnosis model based on a large language model, integrates historical transformer fault cases and uses CGAN adversarial generation technology to expand rare samples to construct an initial QA sample set, and uses a dynamic knowledge distillation mechanism combined with a graph neural network to extract equipment fault ontology from unstructured text and generate a knowledge graph; introduces a Trans-RoPE rotation position encoding layer after the model embedding layer to accurately capture the equipment state coupling relationship; adopts the MoE-DoRA hybrid fine-tuning framework to achieve efficient parameter updating; finally, the fault diagnosis results and customized maintenance strategies are output through the normalization and fully connected layers, forming a complete intelligent power equipment operation and maintenance system from diagnosis to strategy, thereby improving the fault detection efficiency and accuracy of power equipment.

[0060] An embodiment of the present invention provides a transformer equipment fault diagnosis system. For details, see Figure 6 , Figure 6 The figure shows a schematic diagram of the structure of a transformer equipment fault diagnosis system in one embodiment of the present invention, including: The training sample construction module M1 is used to integrate and analyze various operational fault cases of transformer equipment in the target power system and construct an initial set of question-answer pair samples; The text entity extraction module M2 is used to input the initial question-answer pair sample set into a preset large model for training. During the training process, a preset dynamic knowledge extractor is introduced to extract text entities from the initial question-answer pair sample set to generate an equipment fault map; the dynamic knowledge extractor is composed of a distillation mechanism and a graph neural network fusion; An attention reconstruction module M3 is used to transform the device fault map vector and input it into the attention layer reconstructed by the rotational position encoding to capture the device coupling relationship; Iterative training module M4 is used to fine-tune and optimize the parameters of the output results of the attention layer and iterate the training process to obtain a trained fault diagnosis model; The fault diagnosis module M5 is used to collect the operating data of the transformer equipment in real time and input the data into the fault diagnosis model to generate fault diagnosis results and maintenance strategies.

[0061] like Figure 7 As shown, an embodiment of the present invention further provides a computer device, Figure 7 This is a structural block diagram of a preferred embodiment of a computer device provided by the present invention, wherein the computer device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the above-mentioned method when executing the computer program.

[0062] Preferably, the computer program can be divided into one or more modules / units (e.g., computer program 1, computer program 2, ...). These one or more modules / units are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the computer device.

[0063] 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. The general-purpose processor may be a microprocessor, or the processor may be any conventional processor. The processor is the control center of the terminal device, and various parts of the terminal device are connected using various interfaces and lines.

[0064] The memory primarily includes a program storage area and a data storage area. The program storage area can store an operating system, at least one application required for a function, and the data storage area can store related data. Furthermore, the memory can be a high-speed random access memory or a non-volatile memory, such as a plug-in hard drive, a SmartMedia Card (SMC), a Secure Digital (SD) card, or a Flash Card. Alternatively, the memory can be other volatile solid-state storage devices.

[0065] It should be noted that the above terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will understand that Figure 7 The block diagram is merely an example of a terminal device and does not limit the terminal device. The terminal device may include more or fewer components than shown, or a combination of certain components, or different components. Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program may be stored in a computer-readable storage medium. When executed, the program may include the processes of the above-described method embodiments. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0066] Accordingly, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to perform the steps in the method of the above embodiment, for example Figure 1 Steps S1 to S5 described in .

[0067] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A transformer equipment fault diagnosis method, characterized in that: include: Integrate and analyze various operational failure cases of transformer equipment in the target power system to construct an initial set of question-answer pair samples; The initial question-answer pair sample set is input into a preset large model for training. During the training process, a preset dynamic knowledge extractor is introduced to extract text entities from the initial question-answer pair sample set to generate a device fault map; the dynamic knowledge extractor is composed of a distillation mechanism and a graph neural network fusion; The device fault map vector is transformed and input into the attention layer reconstructed by the rotation position encoding to capture the device coupling relationship; The output of the attention layer is fine-tuned and optimized, and the training process is iterated to obtain a trained fault diagnosis model. The operating data of the transformer equipment is collected in real time and input into the fault diagnosis model to generate fault diagnosis results and maintenance strategies.

2. The transformer equipment fault diagnosis method according to claim 1, characterized in that: The integration and analysis of various operational failure cases of transformer equipment in the target power system constructs an initial set of question-answer pair samples, including: Collect transformer operating procedures, transformer test specification data, and historical transformer failure data, annotate them with questions and answers, and integrate them into operating failure cases; A generative adversarial network is used to perform sample enhancement on the running failure cases to construct the initial question-answer pair sample set.

3. The transformer equipment fault diagnosis method according to claim 1, characterized in that: The process of generating the equipment fault map includes: Each question-answer sample in the initial question-answer pair sample set is converted into a sequence and then input into a preset large model. A distillation mechanism is introduced to dynamically adjust the weights of transformer equipment operating indicators to generate a fault knowledge vector; The driven graph neural network extracts the text entities of the fault knowledge vector to construct node features and edge features, and generates the equipment fault map.

4. The transformer equipment fault diagnosis method according to claim 1, characterized in that: The device fault map vector is transformed and input into the attention layer reconstructed by the rotation position encoding to capture the device coupling relationship, including: Input the vector representation of the device fault map into the attention layer and perform feature normalization; The normalization processing result is mapped to a rotation angle using a learnable frequency parameter; wherein the rotation angle is designed to dynamically modify the rotation angle weight using a gated network under a preset indicator threshold condition; Perform weighted analysis on the rotation angles to generate a rotation matrix; During attention calculation, the rotation matrix is used to perform encoding reconstruction on the query matrix and the key-value matrix to generate a reconstructed attention matrix to capture the device coupling relationship.

5. The transformer equipment fault diagnosis method according to claim 1, characterized in that: The parameter fine-tuning and optimization process includes: Deploy an expert routing network to make probability predictions for each eigenvector in the reconstructed attention matrix, and use eigenvectors with probability values exceeding the preset conditions as activation experts; The DoRA fine-tuning mechanism is introduced to decompose the original weights of the activation experts into amplitude and direction components; A low-rank update is performed on the direction component, and the updated direction component is merged with the magnitude component to optimize training.

6. The transformer equipment fault diagnosis method according to claim 1, characterized in that: The process of fine-tuning the parameters of the output of the attention layer and decoding and iteratively training the results also includes: During the iterative training process, the greedy algorithm is integrated into the output layer of the preset large model to select the result corresponding to the maximum prediction probability from the output probability distribution as the target output result.

7. The transformer equipment fault diagnosis method according to claim 1, characterized in that: After the real-time collection of transformer equipment operation data and input into the fault diagnosis model to generate fault diagnosis results and maintenance strategies, the method further includes: Feedback the execution status of the maintenance strategy to the monitoring terminal; The equipment fault map is updated according to the feedback results, and the fault diagnosis model is iteratively optimized.

8. A transformer equipment fault diagnosis system, characterized in that: include: The training sample construction module is used to integrate and analyze various operational failure cases of transformer equipment in the target power system and construct an initial set of question-answer pair samples; A text entity extraction module is used to input the initial question-answer pair sample set into a preset large model for training. During the training process, a preset dynamic knowledge extractor is introduced to extract text entities from the initial question-answer pair sample set to generate an equipment fault map; the dynamic knowledge extractor is composed of a distillation mechanism and a graph neural network fusion; An attention reconstruction module is used to transform the device fault map vector and input it into the attention layer reconstructed by the rotational position encoding to capture the device coupling relationship; The iterative training module is used to fine-tune and optimize the parameters of the output results of the attention layer and iterate the training process to obtain a trained fault diagnosis model; The fault diagnosis module is used to collect the operating data of the transformer equipment in real time and input it into the fault diagnosis model to generate fault diagnosis results and maintenance strategies.

9. A computer device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the transformer equipment fault diagnosis method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the transformer equipment fault diagnosis method according to any one of claims 1 to 7 is implemented.

Images