Power report generation method, system and equipment based on adaptive learning and medium

Abstract

The invention discloses a power report generation method, system and equipment based on adaptive learning and a medium, which are applied to the field of power reports, and comprise the following steps: converting professional power knowledge into prompt lexical elements by utilizing a prompt technology driven by power knowledge, and taking the prompt lexical elements as context information; through a cross-modal feature enhancement mechanism, searching historical report fragments similar to the current equipment state from a historical fault report database, and fusing the historical report fragments with the preprocessed power operation data to generate enhanced feature representation; the enhanced feature representation and prompt lexical elements are input into a report generation model, a self-adaptive reflection learning loss function is adopted, the loss weight is dynamically adjusted according to the learning state of the data category in the training process, and the report generation model is optimized; and generating a power field equipment inspection report according to the optimized model. According to the invention, the professionality and the accuracy of report contents are improved, and the intelligent level and the practicability of the power report generation system are enhanced.

Description

Technical Field

[0001] This invention relates to the field of power reporting technology, and in particular to a method, system, device and medium for generating power reports based on adaptive learning. Background Technology

[0002] With the intelligent and digital transformation of power systems, the amount of operational data from power equipment is growing exponentially, placing higher demands on the efficiency and accuracy of power report generation. Traditional power report generation methods mainly rely on manual writing or automated tools based on fixed templates. These methods have many limitations when dealing with complex and ever-changing power operation data. For example, manual report writing is time-consuming and labor-intensive, and prone to subjective bias and errors; while automated tools based on fixed templates lack flexibility and are difficult to adapt to dynamically changing data and diverse reporting needs.

[0003] Existing technologies struggle to effectively integrate multi-source data in power report generation, such as equipment operation data, historical fault records, and professional knowledge, resulting in reports with limited content, depth, and breadth. Secondly, existing technologies perform poorly when processing dynamic data, failing to reflect the latest operating status of the power system in real time. Furthermore, traditional methods lack a deep understanding and application of power industry expertise when generating reports, leading to insufficient professionalism and accuracy in the report content. These problems severely impact the practicality and reliability of power reports, limiting the level of intelligent management of power systems.

[0004] Therefore, how to achieve intelligent and automated generation of power reports, improve the accuracy and efficiency of reports, and ensure the professionalism and timeliness of report content has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a method, system, device, and medium for generating electricity reports based on adaptive learning, in order to improve the effectiveness of report generation.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a power report generation method based on adaptive learning, comprising: Collect power system operation data and preprocess it to obtain preprocessed power operation data; By utilizing power knowledge-driven prompting technology, power knowledge is transformed into prompting words, and these prompting words are used as contextual information. By using a cross-modal feature enhancement mechanism, historical report fragments similar to the current equipment status are retrieved from the historical fault report database and fused with preprocessed power operation data to generate an enhanced feature representation. The enhanced feature representations and cue words are input into the report generation model. An adaptive reflective learning loss function is used to dynamically adjust the loss weights according to the learning state of the data category during training, thereby optimizing the report generation model. Power equipment inspection reports are generated based on the optimized model.

[0007] Furthermore, the process of collecting and preprocessing power system operation data to obtain preprocessed power operation data includes: Raw data on the operating status of the power system are collected through power quality monitoring devices; The collected raw data is cleaned, denoised, and normalized to obtain preprocessed power operation data.

[0008] Furthermore, the power knowledge-driven prompting technology transforms power-related professional knowledge into prompting terms, including: Retrieve equipment fault classification information from the power system database; Transform equipment fault classification information into structured prompt words; The prompt words are used as contextual information for subsequent report generation.

[0009] Furthermore, the cross-modal feature enhancement mechanism retrieves historical report fragments similar to the current equipment state from the historical fault report database and fuses them with preprocessed power operation data to generate an enhanced feature representation, including: Using a pre-trained multimodal retrieval model, retrieve the top-k historical report fragments that are similar to the current device status from the historical fault report database; The retrieved historical report fragment features are dynamically aggregated and fused with preprocessed power operation data to generate enhanced feature representations.

[0010] Furthermore, the adaptive reflective learning loss function includes: The loss weights are dynamically adjusted based on the learning status of each data type during the training process. In the optimization process, a category prior distribution is introduced to enhance the learning sensitivity to low-frequency power equipment categories; By dynamically adjusting the loss weights, the model's ability to identify rare fault types can be improved.

[0011] Furthermore, the generation of power equipment inspection reports based on the optimized model includes: In the report generation model, an encoder-decoder architecture is used, where the encoder extracts features from the power operation data and the decoder combines cue words and enhanced feature representations to generate report content. In the decoder, cue words are used as contextual information to guide the report generation process and ensure that the generated report content conforms to power industry expertise.

[0012] Another embodiment of the present invention provides an adaptive learning-based power report generation system, comprising: The data processing module is used to collect power system operation data and perform preprocessing to obtain preprocessed power operation data. The lexical conversion module is used to convert electrical knowledge into prompt lexical units using power knowledge-driven prompting technology, and to use the prompt lexical units as contextual information. The feature enhancement module is used to retrieve historical report fragments similar to the current equipment status from the historical fault report database through a cross-modal feature enhancement mechanism, and fuse them with preprocessed power operation data to generate an enhanced feature representation; The model optimization module is used to input the enhanced feature representations and prompt lexical units into the report generation model. It adopts an adaptive reflective learning loss function to dynamically adjust the loss weights according to the learning state of the data category during the training process, thereby optimizing the report generation model. The report generation module is used to generate equipment inspection reports in the power sector based on the optimized model.

[0013] Furthermore, the lexical conversion module is used for: Retrieve equipment fault classification information from the power system database; Transform equipment fault classification information into structured prompt words; The prompt words are used as contextual information for subsequent report generation.

[0014] Another embodiment of the present invention provides a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the adaptive learning-based power report generation method as described above.

[0015] In another embodiment of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the power report generation method based on adaptive learning as described above.

[0016] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: By introducing power knowledge-driven prompting technology, power expertise is transformed into prompt terms and integrated into the report generation process, significantly improving the professionalism and accuracy of the report content. This method ensures that the generated reports not only comply with power industry standards and specifications but also accurately reflect the actual operating status of equipment, providing maintenance personnel with more valuable information and thus improving the scientific rigor and reliability of decision-making. A cross-modal feature enhancement mechanism is employed to retrieve historical report fragments similar to the current equipment status from a historical fault report database and fuse them with preprocessed power operation data. This process effectively integrates multi-source information, enriching the details and background content of the report, making it more in-depth and comprehensive. This helps maintenance personnel fully understand the historical problems and potential risks of the equipment, thereby improving fault diagnosis and prevention capabilities. An adaptive reflective learning loss function is used to dynamically adjust the loss weights based on the learning state of the data category during training, optimizing the report generation model. This adaptive optimization mechanism enables the model to better handle the problem of data class imbalance, improve the ability to identify rare fault types, ensure the accuracy and robustness of report generation, and further enhance the intelligence and practicality of the power report generation system. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of an adaptive learning-based power report generation method in one embodiment of the present invention. Figure 2 This is a structural block diagram of an adaptive learning-based power report generation system according to one embodiment of the present invention; Figure 3 A structural diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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 those of ordinary skill in the art without creative effort are 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 construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0020] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0021] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0022] One embodiment of the present invention provides a power report generation method based on adaptive learning. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The flowchart shown is a step diagram of an adaptive learning-based power report generation method according to one embodiment of the present invention, including steps S11-S15: S11. Collect power system operation data and perform preprocessing to obtain preprocessed power operation data.

[0023] In the daily operation monitoring of power systems, power quality monitoring devices play a crucial role. These devices can collect raw data on the operating status of the power system in real time, including but not limited to key indicators such as voltage, current, frequency, and harmonics. This raw data is a direct reflection of the operating status of the power system, but it often contains noise and outliers, and the data volume is huge and the formats are diverse, making it inconvenient to use it directly for analysis and processing.

[0024] To ensure the accuracy and effectiveness of subsequent analysis, the collected raw data first needs to be cleaned, denoised, and normalized. Data cleaning primarily removes errors, missing values, and outliers to ensure data integrity and accuracy. Denoising utilizes signal processing techniques, such as filtering algorithms, to remove noise components from the data, making it clearer. Normalization converts data of different dimensions and magnitudes to a uniform scale, facilitating subsequent analysis and comparison. Commonly used methods include min-max normalization and Z-score normalization.

[0025] S12. Utilize power knowledge-driven prompting technology to transform power knowledge into prompt words and use these prompt words as contextual information.

[0026] Power system databases are a crucial source of equipment fault classification information, storing vast amounts of data related to equipment faults, including fault descriptions, fault types, and the time and location of occurrence. This information is typically stored in structured or semi-structured formats within relational databases. For instance, the `description` attribute in equipment fault repair information records the specific fault description, while the `conclusion` and `advice` attributes record the repair conclusions and recommendations, respectively. By extracting relevant data from these databases, comprehensive and detailed equipment fault classification information can be obtained, providing fundamental data support for subsequent fault analysis and report generation.

[0027] Structured processing of equipment fault classification information is a crucial step in achieving efficient fault diagnosis and report generation. First, the acquired fault classification information needs preprocessing, including data cleaning, deduplication, handling missing values, and text standardization. Then, natural language processing techniques, such as word segmentation and text vectorization, are used to transform the unstructured fault description text into structured cue words. For example, a pre-trained word vector library can be used to encode the fault description, generating corresponding word embeddings and sentence embeddings. These are then further processed by a text classification model to obtain feature vectors reflecting the distinguishability of fault types. These structured cue words clearly express the characteristics and categories of the fault, facilitating subsequent analysis and processing.

[0028] Using structured cue words as contextual information provides rich background knowledge and key information for the report generation process. During report generation, these cue words can serve as input, helping to generate accurate and detailed fault reports. For example, by inputting cue words into a fault classification model, the model can automatically classify faults and generate corresponding fault classification labels. These labels can serve as key information in the report, helping technicians quickly understand the type and severity of the fault. Furthermore, combining historical fault records and industry expertise can further optimize the report content, making it more targeted and practical.

[0029] S13. Through a cross-modal feature enhancement mechanism, retrieve historical report fragments similar to the current equipment status from the historical fault report database, and fuse them with the preprocessed power operation data to generate an enhanced feature representation.

[0030] In power system fault diagnosis and analysis, historical fault report databases are a valuable resource, recording detailed information about past equipment failures, including fault symptoms, diagnostic processes, and solutions. These historical reports provide important references for analyzing the current equipment status. To efficiently find segments similar to the current equipment status from massive historical reports, pre-trained multimodal retrieval models can be used. These models can simultaneously process data of multiple modalities, such as text and images, and achieve rapid retrieval of historical reports by learning features and patterns in the data.

[0031] Specifically, the multimodal retrieval model encodes the text descriptions and related images (such as pictures of the equipment failure site) in historical fault reports to generate a unified feature representation. Then, by calculating the similarity between the feature representation of the current equipment state and the feature representations of historical reports, the top-k historical report fragments most similar to the current equipment state are retrieved. These fragments may contain fault phenomena, diagnostic methods, and solutions similar to the current fault, providing rich background information for subsequent fault analysis.

[0032] While retrieved historical report fragments provide important reference information, they need to be combined with the actual operating data of the current equipment to more accurately reflect its current status. Preprocessed power operation data includes real-time operating parameters of the equipment, such as voltage, current, and frequency, providing detailed information about the current equipment operation. Dynamically aggregating and fusing the features of the retrieved historical report fragments with the preprocessed power operation data can generate a more comprehensive and accurate feature representation.

[0033] The dynamic aggregation and fusion process can be achieved through various methods. For example, attention mechanisms can be used to dynamically adjust the weights of historical report fragment features and power operation data, assigning higher weights to more relevant features based on the importance of the current equipment status. Furthermore, neural network models, such as fully connected networks or Long Short-Term Memory (LSTM) networks, can be used to deeply fuse the two types of features, generating enhanced feature representations. These enhanced feature representations not only incorporate empirical information from historical faults but also combine the actual operating status of the current equipment, providing stronger support for fault diagnosis and prediction.

[0034] S14. Input the enhanced feature representation and prompt lexical units into the report generation model, and use an adaptive reflective learning loss function to dynamically adjust the loss weight according to the learning state of the data category during the training process to optimize the report generation model.

[0035] During the training of a power equipment fault diagnosis model, different data types (such as equipment operation data, historical fault report fragments, etc.) may have different learning progress and difficulty levels. To optimize model performance, a dynamic weight optimization strategy can be adopted, dynamically adjusting the loss weights based on the learning status of each data type during training. For example, if the loss function of a certain data type decreases rapidly in the early stages of training, it indicates that the learning difficulty of that data type is low, and its weight can be appropriately reduced; conversely, if the loss function of a certain data type decreases slowly, it indicates that its learning difficulty is high, and its weight needs to be increased to promote further learning of that data type by the model.

[0036] Power equipment fault data may exhibit class imbalance, where the number of samples for certain fault types (low-frequency categories) is significantly less than that for other fault types (high-frequency categories). This imbalance can cause the model to favor high-frequency categories during training, thus reducing its ability to identify low-frequency categories. To address this issue, a prior class distribution can be introduced during optimization to enhance the learning sensitivity to low-frequency categories. Specifically, initial weights can be set based on the reciprocal of the number of samples for each category or the reciprocal of the frequency of occurrence of each category, allowing the model to pay more attention to low-frequency categories from the early stages of training. As training progresses, the weights can be dynamically adjusted based on the learning progress of each category, further optimizing the model's ability to identify low-frequency categories.

[0037] Rare fault types, due to their limited sample size, are easily overlooked by the model during training, leading to insufficient recognition ability. Dynamically adjusting the loss weights can effectively improve the model's ability to identify rare fault types. Specific methods include: periodically calculating the accuracy of each fault type during training; increasing the loss weight for rare fault types with low accuracy, and decreasing the loss weight for common fault types with high accuracy. Furthermore, the weight adjustment strategy can be further optimized by incorporating prior class distributions to ensure the model pays balanced attention to all fault types during training. In this way, the model can better learn the characteristics of rare fault types, thereby improving its ability to identify these types.

[0038] S15. Generate an inspection report for power equipment based on the optimized model.

[0039] In power system fault report generation tasks, models employing an encoder-decoder architecture can efficiently transform power operation data into meaningful fault reports. The encoder is responsible for deep feature extraction from the power operation data, including real-time operating parameters of equipment (such as voltage, current, and power) and equipment status information (such as temperature and vibration). Through multi-layer neural network processing, the encoder can transform the raw data into high-dimensional feature vectors, which can effectively capture the core information of the power equipment's operating status.

[0040] The decoder uses the feature vectors extracted by the encoder, combined with preprocessed cue words and enhanced feature representations, to generate report content. The cue words are structured fault classification information, providing clear fault types and key information for report generation. The enhanced feature representations are obtained by dynamically aggregating and fusing features from historical report segments with power operation data; they contain rich background knowledge and historical experience. The decoder integrates this information to progressively generate fault report content that conforms to power industry expertise.

[0041] In the report generation process of the decoder, cue words play a crucial role. As contextual information, cue words provide the decoder with clear fault types and key information, guiding the report generation process in the right direction. For example, when a cue word indicates a fault type of "transformer overheating," the decoder will use this information, combined with electrical expertise, to generate a report related to transformer overheating, including the cause of the fault (such as cooling system failure, overload, etc.), detection methods (such as abnormal temperature sensor data), and handling suggestions (such as checking the cooling system, adjusting the load, etc.).

[0042] To ensure that the generated report content conforms to power industry expertise, the decoder references a large amount of power industry knowledge base and historical fault reports during the generation process. These knowledge bases and historical reports provide the decoder with rich background information and language templates, enabling it to generate accurate, professional, and readable fault reports. In this way, the decoder effectively utilizes cue words as contextual information to guide the report generation process, ensuring that the generated report content not only accurately reflects the current fault status of the equipment but also conforms to the professional standards and specifications of the power industry.

[0043] To more accurately disclose the technical content of the solution, another embodiment is provided to describe the solution in detail: First, this embodiment designs a power knowledge-driven prompting technology. This framework guides the text generation process by introducing a power knowledge-aware prompting mechanism, aiming to improve the accuracy and completeness of the generated reports. Specifically, this embodiment adopts a mainstream encoder-decoder architecture and adds knowledge such as equipment fault classification. During the generation of power-related reports, the output of this knowledge is transformed into structured prompting terms, serving as contextual information to explicitly guide the decoder in generating more targeted and professional fault descriptions. Second, to further improve the quality of report generation, this embodiment designs a cross-modal feature enhancement mechanism. This mechanism utilizes a pre-trained multimodal retrieval model to retrieve the top-k historical report fragments most similar to the current image from a historical fault report database. These semantically rich text features are dynamically aggregated and fused with visual features for use in power-related report generation, simulating the engineer's thinking process of "referencing historical cases" for judgment. Finally, addressing the problem of imbalanced distribution of power equipment data categories, this embodiment introduces an adaptive reflective learning loss function. This loss function dynamically adjusts the weight allocation of each data type according to its learning state during training, significantly improving the learning effect on rare data categories while ensuring the recognition ability of common categories. This strategy overcomes the technical bottleneck of traditional text decoders being unable to actively adjust the generation probability of different data types, and achieves collaborative optimization of report generation.

[0044] This invention will employ a mainstream encoder-decoder architecture for domain-related knowledge representation learning, with the encoder... Mainly used for extracting images in the power industry. Visual features, decoder Generating domain reports through the combined effect of visual features and knowledge-driven cues in the power sector. Specifically, its visual feature extraction process can be represented as follows: in, Represents feature map blocks, For feature dimension, A set of feature map patches, It is the input image. The total number of tiles, It is an encoder function. Therefore, report. It can be defined as a set of word sequences , , For vocabulary list, The input to the encoder is a preprocessed image of the power sector, divided into several patches. Each patch is mapped to a fixed-dimensional feature vector through an embedding layer, and positional encoding is added to preserve spatial information. The encoder extracts high-dimensional semantic features through a multi-layer Transformer encoding structure, outputting a contextual feature representation containing both global and local information. Therefore, the decoding process can be formally represented as: in, For time step Unpredictable word units to for This invention provides knowledge-driven prompts in the power sector. The decoder takes visual features output from the encoder and prompts from a power knowledge base as input. Internally, it employs a self-attention mechanism to capture sequence dependencies and fuses visual features and knowledge prompts through a cross-modal attention mechanism. Finally, it predicts the probability distribution of the current word unit through linear transformation and a softmax function, generating structured and semantically accurate power report content word by word. Therefore, this invention uses language modeling loss as the main optimization objective for the report generation task, as shown below: Among them, the report A set of word sequence , , For vocabulary list, For report length, To predict the probability distribution of word origins.

[0045] In power sector report generation tasks (using equipment fault inspection report generation as an example below), generating text consistent with the fault diagnosis results of power equipment is crucial. This is because in power systems, inspection reports not only need to provide a detailed description of the equipment status but also accurately reflect the current operating condition and potential fault risks. If the generated report has biases in fault judgment, it may lead to incorrect decisions by maintenance personnel, thereby affecting the safe and stable operation of the power grid and even causing serious accidents. However, existing models still face challenges in generating inspection reports with high fault diagnosis consistency and practical application value. Specifically, this invention trained a basic fault classification model on a substation image training set provided by a power company and compared it with mainstream methods on a test set. The results show that the current model still has significant shortcomings in the consistency between fault identification and description.

[0046] Therefore, this invention proposes a power knowledge-driven prompting technology (such as common knowledge about power equipment faults), which guides the text decoder to generate more engineering-practical inspection reports by conveying diagnostic results from fault classification. The classification takes average pooled visual features enhanced across modalities as input (see the next subsection) and outputs fault prediction results for each equipment type through L classification heads, where L is the predefined number of fault types. Each classification head performs a four-class judgment task through a fully connected layer: "normal," "abnormal," "pending," and "not detected." Classification labels can be obtained through structured parsing of the original inspection text, such as extracting keywords or using a pre-trained NLP model to complete entity recognition and classification tasks. The standard cross-entropy loss function is used to optimize classification performance during training.

[0047] During the inference phase, the power equipment fault classification results are converted into prompt tokens, each corresponding to a specific equipment fault state. To this end, this invention introduces four new word segments into the vocabulary: [NOR] for "normal," [ABN] for "abnormal," [UND] for "pending," and [NA] for "not detected." These word segments are injected into the language model as additional contextual information, guiding it to prioritize semantic content related to the current equipment state when generating reports. Through this explicit prompting mechanism, the decoder can more accurately combine visual features with fault diagnosis information, thereby generating inspection reports with greater engineering guidance.

[0048] Furthermore, to further enhance the model's ability to understand complex fault modes, this invention also includes a visualization analysis module to demonstrate how prompt words influence the selection of key sentence structures and terms during generation through an attention weighting mechanism. Notably, this invention discovered that the inspection reports in the training data contain far more fault types than the preset L-class common problems, such as abnormal transformer oil temperature, circuit breaker malfunction, and insulator discharge traces. This additional information is of significant value in improving the model's diagnostic capabilities. Therefore, leveraging the capabilities of a large model, this invention successfully obtained four types of auxiliary fault labeling results, including: 1) Transformer malfunctions: such as excessively high oil temperature, overheating windings, abnormal oil pressure, etc. 2) Malfunction of the isolator switch: such as poor contact, mechanical jamming, or burnt contacts; 3) Surge arrester malfunctions: such as excessive leakage current, aging of the arrester itself, or poor grounding; 4) Abnormalities in cable termination: such as partial discharge, loose joints, or damaged sheath.

[0049] In actual training, this invention uses the Vicuna-13B model as an auxiliary annotation tool. By designing prompt templates for different equipment types, the original inspection report is input into the model to obtain the annotation results of the target faults. In this way, this invention effectively expands the semantic coverage of the training data, improves the model's generalization ability and diagnostic accuracy in real-world scenarios, and provides a foundation for subsequent training.

[0050] In the process of generating equipment fault inspection reports in the power sector, relying solely on equipment status images for judgment may have certain limitations. This is because, in actual inspections and fault analysis, power engineers typically combine historical operating data, past maintenance records, and relevant fault case databases for comprehensive analysis to improve the accuracy and reliability of diagnosis. Inspired by this, in the framework proposed in this invention, in addition to using a visual feature extraction module to obtain key features representing the status from power equipment images, this invention further introduces cross-modal information based on a historical report database and designs a cross-modal feature enhancement module.

[0051] in, For the retrieved Top-k report features, For average pooling of visual features, ⊕ represents the concatenation operation. This is a feature aggregation function. It's used to implement visual features. Guided dynamic feature extraction: This invention performs cross-modal model pre-training on a large-scale power fault report corpus to achieve dynamic feature extraction of input images. Cross-modal retrieval. Specifically, given an image of the status of power equipment, the model can retrieve the top-k most similar fault report features from a historical report database. These report features contain key semantic information such as fault descriptions, maintenance suggestions, and anomaly types that are potentially relevant to the current image.

[0052] Subsequently, to more effectively integrate this information from the text modality, this invention designs a Dynamic Aggregation module to integrate the retrieved Top-k report features into a representative embedding vector. The core idea of ​​the Dynamic Aggregation module is to achieve weighted aggregation through an attention mechanism, allowing different report features to receive different weights based on their relevance to the current image. Specifically, the retrieved report features are first subjected to self-attention processing to capture their contextual relationships; then, their output is used as key-value pairs input to a cross-attention layer, using image features as the query vector, to calculate the relevance weights between each report feature and the current image. Therefore, the above aggregation process can be defined as follows: in, Embedded features for the retrieved Top-k reports, The visual features are averaged pooled. It's important to note that the parameters in the dynamic aggregation module are learnable throughout the training process, while the pre-trained model used for cross-modal retrieval remains frozen to ensure the stability and generalization ability of the retrieval process. Therefore, the design concept of the cross-modal feature enhancement module is highly consistent with the actual workflow. When faced with a new equipment status image, users often refer to the handling records and reports of similar past faults to make an accurate judgment. The method proposed in this invention adopts this "human-like" reasoning process, and by integrating visual information with historical textual information, it achieves a more interpretable and robust power industry report generation capability.

[0053] In the fault inspection of power equipment, the probability of different types of equipment faults occurring in the actual operating environment varies significantly. Some common faults (such as insulation aging and poor contact) appear frequently in historical data, while some rare faults (such as partial discharge mutations and short circuits in extreme environments) are rarely recorded. This inherent imbalance in the distribution of categories in power equipment leads to the model learning well for high-frequency faults during training, but poorly for low-frequency faults, thus affecting the robustness and generalization ability of the overall diagnostic system.

[0054] Although existing research has explored various technical approaches to address the aforementioned problems, no method has yet effectively mitigated the learning bias caused by class imbalance. This is primarily because traditional text decoders rely solely on linguistic probability modeling when generating inspection reports, without explicitly distinguishing between different fault types. Therefore, it is difficult to directly control the generation tendency of different categories through the decoding process. To overcome this limitation, a dynamically adjustable adaptive reflective learning mechanism is proposed to achieve balanced learning across different fault types and further guide the text generation module to output more discriminative fault descriptions. Specifically, this invention designs an adaptive reflective learning algorithm that dynamically adjusts the loss weights based on the learning status of each fault type in each training round, thereby improving the model's ability to identify low-frequency categories. To this end, this invention designs an improved Logit Adjustment Loss (LA Loss) function. Its core idea is to enhance the learning sensitivity to low-frequency power equipment categories by introducing a prior distribution of categories during the optimization process. For a given power equipment fault type... Its positive sample labels Corresponding log-adjusted loss function Defined as follows: in, For the input sample, This indicates the true label category of the current sample. Indicates the input sample category The logarithm of Indicates device type Category distribution, Indicates device type The prior distribution is in logarithmic form. This loss function is introduced by... This feature enables automatic adjustment of the weights of different features, allowing the model to focus more on fault types with small sample sizes and easy to ignore during the optimization process, thereby improving its recognition accuracy.

[0055] However, logarithmic adjustment methods based on a fixed class distribution cannot reflect the dynamic learning state of the model during training. In reality, different fault types differ not only in the number of samples but also in their learning difficulty—for example, some rare faults, although with few samples, have obvious features and are easy to identify; while other common faults may be easily confused due to their high similarity to other types. Inspired by existing work, this invention proposes using the model's average prediction score on the validation set to evaluate the learning state of each fault type: if a fault type maintains a low prediction score after multiple training epochs, it indicates greater learning difficulty and should be given a higher loss weight. Therefore, this invention sets the initial class distribution π as the statistical frequency of the training data and adaptively updates it during training using the following formula: in, Indicates the first After the training cycle ends, the validation set is at the end of the first training cycle. The average prediction score for each fault type. In this way, the present invention can dynamically assign a reasonable loss weight to each fault, taking into account both the original data distribution and the changes in the model's mastery of it during training.

[0056] Finally, this invention combines adaptive reflective learning loss with language modeling loss to form the total loss function for training the model, which is defined as follows: in, This represents the language modeling loss in the report text generation section. This indicates adaptive reflective learning loss. It is a hyperparameter that balances the two types of losses.

[0057] Therefore, the adaptive reflective learning mechanism proposed in this invention not only solves the long-standing class imbalance problem in power equipment fault diagnosis, but also improves the model's ability to identify complex and rare faults by dynamically adjusting loss weights. More importantly, this mechanism can be naturally integrated with a prompt-driven report generation process, significantly enhancing the semantic accuracy and structural integrity of the generated text content, thereby better serving the practical application needs of power sector report generation tasks in business scenarios.

[0058] This invention introduces an adaptive learning-based power report generation method that employs power knowledge-driven prompting technology. Power expertise is transformed into prompt terms and integrated into the report generation process, significantly improving the professionalism and accuracy of the report content. This method ensures that the generated reports not only comply with power industry standards and specifications but also accurately reflect the actual operating status of equipment, providing maintenance personnel with more valuable information and thus improving the scientific rigor and reliability of decision-making. A cross-modal feature enhancement mechanism is employed to retrieve historical report fragments similar to the current equipment status from a historical fault report database and fuse them with preprocessed power operation data. This process effectively integrates multi-source information, enriching the details and background content of the report, making it more in-depth and comprehensive. This helps maintenance personnel fully understand the historical problems and potential risks of the equipment, thereby improving fault diagnosis and prevention capabilities. An adaptive reflective learning loss function is used to dynamically adjust the loss weights based on the learning state of the data category during training, optimizing the report generation model. This adaptive optimization mechanism enables the model to better handle the problem of data category imbalance, improve the ability to identify rare fault types, ensure the accuracy and robustness of report generation, and further enhance the intelligence and practicality of the power report generation system.

[0059] This invention also provides an adaptive learning-based power report generation system for executing the adaptive learning-based power report generation method described above. Figure 2 This is a block diagram of an adaptive learning-based power report generation system according to an embodiment of the present invention. The device includes: The data processing module 21 is used to collect power system operation data and perform preprocessing to obtain preprocessed power operation data; The word conversion module 22 is used to convert power knowledge into prompt words using power knowledge-driven prompting technology, and use the prompt words as contextual information. Feature enhancement module 23 is used to retrieve historical report fragments similar to the current equipment status from the historical fault report database through a cross-modal feature enhancement mechanism, and fuse them with preprocessed power operation data to generate enhanced feature representations; The model optimization module 24 is used to input the enhanced feature representation and prompt lexical units into the report generation model, and adopts an adaptive reflective learning loss function to dynamically adjust the loss weight according to the learning state of the data category during the training process to optimize the report generation model. The report generation module 25 is used to generate equipment inspection reports in the power sector based on the optimized model.

[0060] The lexical conversion module is used for: Retrieve equipment fault classification information from the power system database; Transform equipment fault classification information into structured prompt words; The prompt words are used as contextual information for subsequent report generation.

[0061] The technical features and effects of the system proposed in the embodiments of the present invention are the same as those of the method proposed in the embodiments of the present invention, and will not be repeated here. Each module in the above system can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0062] See Figure 3 This is a structural block diagram of a computer device provided in an embodiment of the present invention. The computer device provided in this embodiment includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps as described in the above embodiment of the power report generation method based on adaptive learning. Figure 1 Steps S11 to S15 as described above; or, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments, such as modules 21 to 25 of the power report generation method system based on adaptive learning.

[0063] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer device.

[0064] The computer device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the schematic diagram is merely an example of a computer device and does not constitute a limitation on the computer device. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, buses, etc.

[0065] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the computer device, connecting various parts of the computer device via various interfaces and lines.

[0066] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0067] If the modules integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0068] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0069] Accordingly, embodiments of the present invention provide a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform steps in the adaptive learning-based power report generation method of the above embodiments, for example... Figure 1 Steps S11 to S15 as described above.

[0070] In summary, compared with the prior art, the power report generation method, apparatus, computer device, and computer-readable storage medium provided by the embodiments of the present invention have the following beneficial effects: By introducing power knowledge-driven prompting technology, power expertise is transformed into prompt terms and integrated into the report generation process, significantly improving the professionalism and accuracy of the report content. This method ensures that the generated reports not only comply with power industry standards and specifications but also accurately reflect the actual operating status of equipment, providing maintenance personnel with more valuable information and thus improving the scientific rigor and reliability of decision-making. A cross-modal feature enhancement mechanism is employed to retrieve historical report fragments similar to the current equipment status from a historical fault report database and fuse them with preprocessed power operation data. This process effectively integrates multi-source information, enriching the details and background content of the report, making it more in-depth and comprehensive. This helps maintenance personnel fully understand the historical problems and potential risks of the equipment, thereby improving fault diagnosis and prevention capabilities. An adaptive reflective learning loss function is used to dynamically adjust the loss weights based on the learning state of the data category during training, optimizing the report generation model. This adaptive optimization mechanism enables the model to better handle the problem of data class imbalance, improve the ability to identify rare fault types, ensure the accuracy and robustness of report generation, and further enhance the intelligence and practicality of the power report generation system.

[0071] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A power report generation method based on adaptive learning, characterized in that, include: Collect power system operation data and preprocess it to obtain preprocessed power operation data; By utilizing power knowledge-driven prompting technology, power knowledge is transformed into prompting words, and these prompting words are used as contextual information. By using a cross-modal feature enhancement mechanism, historical report fragments similar to the current equipment status are retrieved from the historical fault report database and fused with preprocessed power operation data to generate an enhanced feature representation. The enhanced feature representations and cue words are input into the report generation model. An adaptive reflective learning loss function is used to dynamically adjust the loss weights according to the learning state of the data category during training, thereby optimizing the report generation model. Power equipment inspection reports are generated based on the optimized model.

2. The power report generation method based on adaptive learning as described in claim 1, characterized in that, The process of collecting and preprocessing power system operation data to obtain preprocessed power operation data includes: Raw data on the operating status of the power system are collected through power quality monitoring devices; The collected raw data is cleaned, denoised, and normalized to obtain preprocessed power operation data.

3. The power report generation method based on adaptive learning as described in claim 1, characterized in that, The aforementioned prompting technology, driven by electricity knowledge, transforms electricity-related professional knowledge into prompting terms, including: Retrieve equipment fault classification information from the power system database; Transform equipment fault classification information into structured prompt words; The prompt words are used as contextual information for subsequent report generation.

4. The power report generation method based on adaptive learning as described in claim 1, characterized in that, The method employs a cross-modal feature enhancement mechanism to retrieve historical report fragments similar to the current equipment status from the historical fault report database, and fuses them with preprocessed power operation data to generate an enhanced feature representation, including: Using a pre-trained multimodal retrieval model, retrieve the top-k historical report fragments that are similar to the current device status from the historical fault report database; The retrieved historical report fragment features are dynamically aggregated and fused with preprocessed power operation data to generate enhanced feature representations.

5. The power report generation method based on adaptive learning as described in claim 1, characterized in that, The adaptive reflective learning loss function includes: The loss weights are dynamically adjusted based on the learning status of each data type during the training process. In the optimization process, a category prior distribution is introduced to enhance the learning sensitivity to low-frequency power equipment categories; By dynamically adjusting the loss weights, the model's ability to identify rare fault types can be improved.

6. The power report generation method based on adaptive learning as described in claim 1, characterized in that, The process of generating power equipment inspection reports based on the optimized model includes: In the report generation model, an encoder-decoder architecture is used, where the encoder extracts features from the power operation data and the decoder combines cue words and enhanced feature representations to generate report content. In the decoder, cue words are used as contextual information to guide the report generation process and ensure that the generated report content conforms to power industry expertise.

7. A power report generation system based on adaptive learning, characterized in that, include: The data processing module is used to collect power system operation data and perform preprocessing to obtain preprocessed power operation data. The lexical conversion module is used to convert electrical knowledge into prompt lexical units using power knowledge-driven prompting technology, and to use the prompt lexical units as contextual information. The feature enhancement module is used to retrieve historical report fragments similar to the current equipment status from the historical fault report database through a cross-modal feature enhancement mechanism, and fuse them with preprocessed power operation data to generate an enhanced feature representation; The model optimization module is used to input the enhanced feature representations and prompt lexical units into the report generation model. It adopts an adaptive reflective learning loss function to dynamically adjust the loss weights according to the learning state of the data category during the training process, thereby optimizing the report generation model. The report generation module is used to generate equipment inspection reports in the power sector based on the optimized model.

8. The power report generation system based on adaptive learning as described in claim 7, characterized in that, The word conversion module is used for: Retrieve equipment fault classification information from the power system database; Transform equipment fault classification information into structured prompt words; The prompt words are used as contextual information for subsequent report generation.

9. A computer device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the adaptive learning-based power report generation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the power report generation method based on adaptive learning as described in any one of claims 1 to 7.

Images