Information-enhanced named entity recognition system and equipment for power operation and maintenance text

Abstract

This invention provides an information-enhanced named entity recognition system and device for power equipment operation and maintenance text. The system includes: a dataset construction module for correcting characters in power equipment operation and maintenance work orders, determining entity categories, and labeling entities; a word embedding module for converting words from discrete symbolic forms into continuous vector representations; a location-aware global attention mechanism module for performing location encoding and attention score calculation, and obtaining correlation information between different samples; a fine-grained enhancement module for capturing local information and more accurately identifying entities in operation and maintenance work orders; a conditional random field module for constructing a probabilistic framework and capturing dependencies between labels; and a visualization, interaction, and storage module for providing visual interaction for named entity recognition in power equipment operation and maintenance text. This invention helps promote standardized equipment management, reduce operation and maintenance costs, and optimize decision support.

Description

Technical Field

[0001] This invention relates to the field of power operation and maintenance technology, and in particular to an information-enhanced named entity recognition system and device for power operation and maintenance text. Background Technology

[0002] With the continuous increase in electricity demand, the workload of power equipment operation and maintenance is also increasing. The operation and maintenance data generated during this process is crucial for analyzing equipment operating status and building a smart grid. Power equipment operation and maintenance work orders are the main source of this data, primarily recording specific information about the work performed by maintenance personnel, including equipment name, line, damaged parts, and specific maintenance content. However, most existing power equipment operation and maintenance work orders are unstructured text, making accurate information extraction difficult and time-consuming. Therefore, how to efficiently extract information from a large amount of unstructured power equipment operation and maintenance text has become a critical problem that urgently needs to be solved. However, extracting entity information from power equipment operation and maintenance work orders is challenging because they often contain a large number of technical terms and industry-specific abbreviations; secondly, the work orders are recorded by different maintenance personnel, resulting in significant differences in writing styles and frequent issues such as non-standard expression and incomplete descriptions. Traditional general-purpose natural language processing methods often fail to achieve ideal results when processing these specialized domain texts, failing to effectively identify entity boundaries and providing insufficient global information extraction from the dataset. Therefore, developing an information-enhanced named entity recognition system and equipment for power operation and maintenance texts can effectively overcome the shortcomings of the aforementioned related technologies, and has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0003] To address the aforementioned problems in the existing technology, embodiments of the present invention provide an information-enhanced named entity recognition system and device for power operation and maintenance text.

[0004] In a first aspect, embodiments of the present invention provide an information-enhanced named entity recognition system for power operation and maintenance text, comprising: a dataset construction module for correcting characters in power equipment operation and maintenance work orders, determining entity categories, and labeling entities; a word embedding module for converting words from discrete symbolic forms into continuous vector representations; a location-aware global attention mechanism module for performing location encoding and attention score calculation, and obtaining association information between different samples; a fine-grained enhancement module for capturing local information and more accurately identifying entities in operation and maintenance work orders; a conditional random field module for constructing a probabilistic framework and capturing dependencies between labels; and a visualization interaction and storage module for performing visual interaction of named entity recognition in power operation and maintenance text.

[0005] Based on the above system embodiments, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes a dataset construction module comprising: a preprocessing module for deduplicating the acquired data, removing content containing useless characters, missing information, and erroneous statements; an entity category determination module for determining the equipment name, voltage level, line, substation, damaged part, specific operation and maintenance content, and time point; and an entity annotation module for annotating the text data using the BMES annotation system, where "B" represents the beginning of an entity, "M" represents the middle part of an entity, "E" represents the end of an entity, and "S" represents a single-word entity.

[0006] Based on the above system embodiments, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention converts words from discrete symbolic forms to continuous vector representations, including: using BERT-wwm-ext to embed characters into the input sequence, enhancing the word embedding of the Chinese corpus, consisting of 12 Transformer layers, each with 768 hidden units and 12 attention heads.

[0007] Based on the above system embodiments, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes the following steps: performing position encoding and attention score calculation, and obtaining association information between different samples.

[0008]

[0009]

[0010]

[0011]

[0012]

[0013]

[0014]

[0015] Where Attn is the position-aware global attention mechanism; softmax is the softmax function; The positional awareness score is the value matrix between positions t and j. This is global information; The query vector for the token at position t; Let j be the key vector of the token at position j; is the relative position embedding vector between two tokens; u is the first learnable bias vector, which adjusts the attention based on the inherent properties of the tokens; v is the second learnable bias vector, which adjusts the attention based on the direction and distance between the two tokens, further enhancing the model's ability to capture spatial and sequential relationships; T is the matrix transpose symbol; t is the current position; j is the position to be focused on; i is the frequency index of the position encoding, with a value range of [0, d]. k / 2]; Let be the dimension of the key vector; L-DNorm is a learnable dual normalization strategy; Q is the query matrix; K is the key matrix; , These are two globally shared learnable memory units, where, This helps calculate similarity to generate attention weights. By combining with attention weights, the input features are updated, thereby affecting the final output; X is the input sequence; , , It is a parameter matrix used to map the input X to different spaces.

[0016] Based on the above system embodiments, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes a learnable dual normalization strategy, comprising:

[0017]

[0018]

[0019]

[0020] in, Let be the intermediate matrix element obtained through matrix multiplication, representing the initial associated value in the i-th row and j-th column; These are the normalized values ​​of the column; The learnable parameter for adjusting the strength of column normalization; exp is the exponential function; This represents the original association value between the nth row and the jth column; n is the row and column dimension index. This is the final output value after double normalization; This represents the value in the i-th row and n-th column of the intermediate matrix after column normalization. Learnable parameters for adjusting row normalization strength.

[0021] Based on the above system embodiments, the information enhancement named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes a fine-grained enhancement module comprising:

[0022]

[0023]

[0024]

[0025] in, This is the result of fusing the nonlinear features of the second convolutional layer with the features of the first convolutional layer as the main focus, capturing the local details in the input that are focused by different convolutional kernels; Norm is a layer normalization operation; This is the first one-dimensional convolutional layer; This is the second one-dimensional convolutional layer; Hadamard product symbol; GELU is the activation function; This is the result of fusing the nonlinear features of the first convolutional layer with the features of the second convolutional layer as the main driver, thus mining potential fine-grained semantic differences from another perspective; Out is the final output of the fine-grained enhancement module; This is the third one-dimensional convolutional layer.

[0026] Based on the above system embodiments, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes the following steps: Constructing a probabilistic framework and capturing the dependencies between tags.

[0027]

[0028] Where P(Y|X) is the conditional probability; A quantification symbol for compatibility between adjacent tags; Assign a label to position i The appropriateness evaluation symbol; Y represents a complete label sequence; The set of all possible labeled sequences corresponding to the observed sequence X; It is the quantization symbol for the compatibility between the label at position i and the label at position i+1 in all possible label sequences in the denominator; The evaluation symbol for assigning a specific label at position i in all possible label sequences in the denominator.

[0029] Based on the above system embodiments, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes the following visual interaction for performing named entity recognition of power operation and maintenance text: inputting an operation and maintenance work order, automatically extracting the corresponding information, batch processing multiple operation and maintenance work orders, and batch exporting the recognition results.

[0030] In a second aspect, embodiments of the present invention provide an electronic device, comprising:

[0031] At least one processor, at least one memory, a communication interface, and a bus; wherein,

[0032] The processor, memory, and communication interface communicate with each other through the bus;

[0033] The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the information-enhanced named entity recognition system for power operation and maintenance text provided by any of the various implementations of the first aspect.

[0034] Thirdly, embodiments of the present invention provide a non-transitory computer-readable storage medium that stores computer instructions that enable a computer to implement the information-enhanced named entity recognition system for power operation and maintenance text provided by any of the various implementations of the first aspect.

[0035] The information-enhanced named entity recognition system and device for power operation and maintenance text provided in this invention integrates a dataset construction module, a word embedding module, a location-aware global attention mechanism module, a fine-grained enhancement module, a conditional random field module, and a visualization interaction and storage module. This enables the extraction of key information from power operation and maintenance text, improves the efficiency and intelligence level of power equipment operation and maintenance management, and helps to promote the standardization of equipment management, reduce operation and maintenance costs, and optimize decision support. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the structure of an information-enhanced named entity recognition system for power operation and maintenance text provided in an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the entity category division structure provided in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the fine-grained information enhancement module structure provided in an embodiment of the present invention;

[0041] Figure 5This is a schematic diagram of the power equipment operation and maintenance management system provided in an embodiment of the present invention;

[0042] Figure 6 This is a schematic diagram showing the functional page display effect of the employee registration and management module provided in an embodiment of the present invention;

[0043] Figure 7 This is a schematic diagram showing the functional page display effect of the maintenance work order information extraction module provided in this embodiment of the invention;

[0044] Figure 8 This is a schematic diagram showing the functional page display effect of the structured information management module for operation and maintenance work orders provided in this embodiment of the invention;

[0045] Figure 9 This is a schematic diagram illustrating the dynamic effects of model training provided in an embodiment of the present invention.

[0046] Figure 10(a) is a schematic diagram of the confusion matrix thermal effect of the benchmark model provided in the embodiment of the present invention;

[0047] Figure 10(b) is a schematic diagram of the confusion matrix thermal effect of the proposed model provided in the embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined with each other to form feasible technical solutions. Such combinations are not constrained by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0049] This invention provides an information-enhanced named entity recognition system for power operation and maintenance texts. (See also...) Figure 1The system includes: a dataset construction module for correcting characters in power equipment maintenance work orders, determining entity categories, and labeling entities; a word embedding module for converting words from discrete symbolic forms into continuous vector representations; a location-aware global attention mechanism module for performing location encoding and attention score calculation, and obtaining correlation information between different samples; a fine-grained enhancement module for capturing local information and more accurately identifying entities in maintenance work orders; a conditional random field module for constructing a probabilistic framework and capturing dependencies between labels; and a visualization interaction and storage module for providing visual interaction for named entity recognition in power maintenance texts.

[0050] Based on the above system embodiments, as an optional embodiment, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes a dataset construction module comprising: a preprocessing module for deduplicating the acquired data, removing content containing useless characters, missing information, and erroneous statements; an entity category determination module for determining the equipment name, voltage level, line, substation, damaged part, specific operation and maintenance content, and time point; and an entity annotation module for annotating the text data using the BMES annotation system, where "B" represents the beginning of an entity, "M" represents the middle part of an entity, "E" represents the end of an entity, and "S" represents a single word entity.

[0051] Specifically, all data in the preprocessing module originates from power equipment operation and maintenance work orders recorded in the State Grid Production Management System (PMS). The acquired data undergoes preprocessing, first by deduplication, and then by removing records containing useless characters, numerous missing information, and erroneous statements. The main purpose of the NER task in the entity category determination module is to extract key information about equipment operation and maintenance, facilitating further management and analysis of equipment by power companies, including equipment lifespan prediction, defect correlation analysis, and smart grid construction. Based on these considerations, seven entity categories with the highest analytical value were identified: equipment name, voltage level, line, substation, damaged location, specific operation and maintenance content, and time, such as... Figure 3 As shown.

[0052] Based on the above system embodiments, as an optional embodiment, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention, wherein the word is converted from discrete symbolic form to continuous vector representation, includes: using BERT-wwm-ext to embed characters into the input sequence, which enhances the word embedding of the Chinese corpus, and consists of 12 Transformer layers, each with 768 hidden units and 12 attention heads.

[0053] Specifically, character embeddings are essential in Named Entity Recognition (NER) tasks because they transform words from discrete symbolic forms into continuous vector representations. This process effectively reduces the dimensionality of the word space, creating a low-dimensional continuous mapping. By incorporating character-level information, the model gains a richer understanding of the semantics of words in the surrounding text, improving its ability to interpret context and accurately identify entities.

[0054] Based on the above system embodiments, as an optional embodiment, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes the following steps: performing position encoding and attention score calculation, and obtaining association information between different samples.

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062] Where Attn is the position-aware global attention mechanism; softmax is the softmax function; The positional awareness score is the value matrix between positions t and j. This is global information; The query vector for the token at position t; Let j be the key vector of the token at position j; is the relative position embedding vector between two tokens; u is the first learnable bias vector, which adjusts the attention based on the inherent properties of the tokens; v is the second learnable bias vector, which adjusts the attention based on the direction and distance between the two tokens, further enhancing the model's ability to capture spatial and sequential relationships; T is the matrix transpose symbol; t is the current position; j is the position to be focused on; i is the frequency index of the position encoding, with a value range of [0, d]. k / 2]; Let be the dimension of the key vector; L-DNorm is a learnable dual normalization strategy; Q is the query matrix; K is the key matrix; , These are two globally shared learnable memory units, where, This helps calculate similarity to generate attention weights. By combining with attention weights, the input features are updated, thereby affecting the final output; X is the input sequence; , , It is a parameter matrix used to map the input X to different spaces.

[0063] Specifically, the Transformer is the most mainstream deep learning model, thanks to its core attention mechanism, which makes it far superior to other models in acquiring long-range dependencies and achieving efficient parallel processing. In the Transformer model, the input sequence is processed simultaneously, lacking inherent sequence dependencies. While this improves computational efficiency, the model struggles to grasp the order of the input sequence. Therefore, the Transformer model introduces positional embeddings to explicitly define the positional information of each token, enabling the model to use this information to distinguish and understand the sequential structure of the data.

[0064] However, in the original transformer encoder, positional encoding is obtained through a linear combination of sine and cosine functions. This positional encoding method enhances the model's distance awareness, but the model struggles to distinguish the direction from which different tokens originate. Furthermore, the original self-attention mechanism only focuses on the relationships between different positional features within a single sample, ignoring potential associations between different samples. To address these issues, a position-aware global attention mechanism is designed. This mechanism employs a novel positional encoding method and attention score calculation method, and introduces two globally learnable units to acquire association information between different samples.

[0065] Based on the above system embodiments, as an optional embodiment, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes a learnable dual normalization strategy, comprising:

[0066] (7)

[0067] (8)

[0068] (9)

[0069] in, Let be the intermediate matrix element obtained through matrix multiplication, representing the initial associated value in the i-th row and j-th column; These are the normalized values ​​of the column; The learnable parameter for adjusting the strength of column normalization; exp is the exponential function; This represents the original association value between the nth row and the jth column; n is the row and column dimension index. This is the final output value after double normalization; This represents the value in the i-th row and n-th column of the intermediate matrix after column normalization. Learnable parameters for adjusting row normalization strength.

[0070] Specifically, an overbalanced attention distribution may affect the model's feature recognition ability. Therefore, two learnable parameters, α and β, were set, and a learnable dual normalization strategy was constructed as shown in equations (7) to (9) (L-DNorm), so that the model can dynamically adjust the intensity of normalization during training.

[0071] Based on the above system embodiments, as an optional embodiment, the information enhancement named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes a fine-grained enhancement module comprising:

[0072] (10)

[0073] (11)

[0074] (12)

[0075] in, This is the result of fusing the nonlinear features of the second convolutional layer with the features of the first convolutional layer as the main focus, capturing the local details in the input that are focused by different convolutional kernels; Norm is a layer normalization operation; This is the first one-dimensional convolutional layer; This is the second one-dimensional convolutional layer; Hadamard product symbol; GELU is the activation function; This is the result of fusing the nonlinear features of the first convolutional layer with the features of the second convolutional layer as the main driver, thus mining potential fine-grained semantic differences from another perspective; Out is the final output of the fine-grained enhancement module; This is the third one-dimensional convolutional layer.

[0076] Specifically, while position-aware global attention mechanisms excel at handling global contextual information, they may lack precision in capturing details. This is particularly true in named entity recognition tasks, where the model needs to accurately identify specific entities with a small percentage of characters within long sentences. Therefore, a fine-grained information enhancement module was designed to run parallel to the position-aware global attention mechanism to improve the model's ability to capture local information, enabling it to more accurately identify entities in the text. The specific structure of this module is as follows: Figure 4As shown, it mainly consists of two parallel convolutional networks. CNN is good at extracting features from local windows and can capture phrase-level word features. These local features are particularly helpful for named entity recognition. The fine-grained information enhancement module uses two convolutional layers with different kernel sizes to process in parallel. The small kernel is used to capture local fine-grained features, and the large kernel is used to capture relatively longer-range contextual dependencies. The two convolutional layers adopt an interactive design. The output of each layer modulates the feature extraction of the other layer, and the feature interaction is achieved through element-wise multiplication. The calculation method of this module is shown in equations (10) to (12). Conv1 and Conv2 represent two one-dimensional convolutional layers with kernel sizes of 1 and 5, respectively. Norm indicates that performing layer normalization can improve training stability, and GELU is the activation function used. Conv3 maps the features after the previous processing back to the original feature dimension. Finally, a residual mechanism is added to alleviate the gradient vanishing problem.

[0077] Based on the above system embodiments, as an optional embodiment, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes, as an example, constructing a probabilistic framework and capturing the dependencies between tags:

[0078] (13)

[0079] Where P(Y|X) is the conditional probability; A quantification symbol for compatibility between adjacent tags; Assign a label to position i The appropriateness evaluation symbol; Y represents a complete label sequence; The set of all possible labeled sequences corresponding to the observed sequence X; It is the quantization symbol for the compatibility between the label at position i and the label at position i+1 in all possible label sequences in the denominator; The evaluation symbol for assigning a specific label at position i in all possible label sequences in the denominator.

[0080] Specifically, Conditional Random Fields (CRFs) represent a probabilistic framework designed to address sequence labeling challenges in machine learning. This architecture excels at capturing dependencies between labels, making it particularly advantageous for tasks where traditional independent classification methods have proven insufficient. Consider an input sequence X = [x1, ..., xn] paired with its corresponding label sequence Y = [y1, ..., yn]. The CRF framework establishes a probability mapping between these sequences using equation (13). In equation (13), the denominator serves as a normalization term, summing over all possible label sequences (Y and X) to ensure appropriate probability distribution characteristics. The optimization objective focuses on maximizing the conditional probability P(Y|X) during the training phase.

[0081] Based on the above system embodiments, as an optional embodiment, the information-enhanced named entity recognition system for power operation and maintenance text provided in this embodiment of the invention includes the following visual interaction for performing named entity recognition of power operation and maintenance text: inputting an operation and maintenance work order, automatically extracting the corresponding information, batch processing multiple operation and maintenance work orders, and batch exporting the recognition results.

[0082] Specifically, this module also features storage capabilities, automatically saving user-input sentences and named entity information identified by the model. This stored data can be used for subsequent data analysis and mining to further optimize the model and improve recognition accuracy. It also provides power maintenance personnel with historical data query services. For example, maintenance personnel can quickly find named entity recognition results for similar past faults through keyword searches, providing a reference for current fault handling and improving the efficiency of fault diagnosis and processing. Simultaneously, the stored data can also serve as part of a knowledge base, working collaboratively with other modules to provide rich data support for downstream applications such as building power maintenance knowledge graphs.

[0083] The information-enhanced named entity recognition system and device for power operation and maintenance text provided in this invention integrates a dataset construction module, a word embedding module, a location-aware global attention mechanism module, a fine-grained enhancement module, a conditional random field module, and a visualization interaction and storage module. This enables the extraction of key information from power operation and maintenance text, improves the efficiency and intelligence level of power equipment operation and maintenance management, and helps to promote the standardization of equipment management, reduce operation and maintenance costs, and optimize decision support.

[0084] Specifically, the information-enhanced named entity recognition system and device for power operation and maintenance text provided in this embodiment of the invention constructs a large-scale standardized text dataset for the power operation and maintenance field, filling the gap in high-quality training data in this field. By labeling seven core entities, including equipment name, voltage level, and associated line, it provides an important data foundation for the intelligent management of power equipment operation and maintenance. The construction of this dataset provides reliable training resources for subsequent intelligent applications such as operation and maintenance data analysis, fault diagnosis, and lifespan prediction. The standardized processing of the dataset provides reusable benchmark data for other related research and applications in the industry, promoting the development of the field's technology. The word embedding based on BERT-wwm-ext has powerful contextual understanding capabilities, accurately capturing the semantic representation of words in different contexts. The rich semantic features provided by the pre-trained model provide a solid foundation for subsequent deep feature extraction, effectively improving the overall performance of the model. The dynamic characteristics of word embedding enable the model to adaptively adjust word representations according to context, enhancing the flexibility of feature extraction. The improved positional encoding method enhances the model's ability to perceive sequence positional information, improving the capture effect of long-distance dependencies. The design incorporates globally shared learnable memory units, enabling the model to acquire information from the entire dataset, overcoming the limitations of traditional methods that focus only on a small number of samples. Through the decomposition of attention scores, effective fusion of content and location information is achieved, improving the accuracy of feature extraction. This effectively avoids the problem of anomalous feature values ​​of a single token dominating the entire attention distribution, improving the model's robustness. Dynamically adjustable normalization parameters allow the model to adaptively balance the attention distribution of different samples, optimizing feature extraction performance. The application of a learnable dual normalization strategy enhances the model's training stability, ensuring high-quality feature learning. The parallel multi-scale convolutional network design enables accurate extraction of local detail features, improving entity recognition accuracy. An interactive feature fusion mechanism enhances the expressive power of features at different scales, optimizing the model's feature representation. The application of residual connections effectively alleviates the gradient vanishing problem in deep networks, ensuring the stability of model training. By establishing dependencies between label sequences, the accuracy of entity boundary recognition is significantly improved. The application of a probabilistic framework enables the model to learn label transfer rules, improving the rationality of sequence labeling. The final label prediction process has been optimized, reducing unreasonable label sequences in the prediction results. Batch processing and result export of single and multiple maintenance work orders are supported, improving efficiency in practical applications. The automatic storage function enables continuous accumulation of identification results, providing support for subsequent data analysis and model optimization. The rapid retrieval function of historical data provides decision-making references for maintenance personnel, improving fault handling efficiency. The stored structured data can be directly used to build a power maintenance knowledge graph, promoting the intelligent construction of the power system.

[0085] The specific application areas or related products of this invention.

[0086] The information-enhanced named entity recognition system for power operation and maintenance text proposed in this invention can be further used to build a power equipment operation and maintenance management system. This system can be ported to mobile electronic devices such as smartphones and tablets. The system structure is as follows: Figure 5 As shown.

[0087] The power equipment operation and maintenance management system consists of an employee registration and management module, an operation and maintenance work order information extraction module, and an operation and maintenance work order structured information management module.

[0088] The main function of the employee registration and management module is to add and manage the information of operation and maintenance employees. Employee information includes basic information such as employee number, name, gender, and phone number. It supports fuzzy search based on a single employee information and allows modification of employee information and permissions. Figure 6 This displays the functional pages for this module.

[0089] The main functions of the maintenance work order information extraction module are to import and export original equipment maintenance work orders, perform named entity recognition on the work orders, and convert unstructured maintenance information into a structured form. This module also features batch management and batch structuring capabilities. Figure 7 This displays the functional pages for this module.

[0090] The maintenance work order structured information management module is responsible for storing and querying the structured information after named entity recognition. The structured information is displayed according to the defined entity categories. Figure 8 This displays the functional pages for this module.

[0091] Evidence related to the technical effects obtained by the embodiments of the present invention.

[0092] The named entity recognition model proposed in this invention was tested on the power operation and maintenance dataset constructed in this invention to verify its recognition performance. In the experiments, three evaluation metrics—precision (P), recall (R), and F1 score—were used to verify the model's performance. The formula for calculating precision (P) is as follows:

[0093]

[0094] Where TP represents the number of samples correctly predicted as positive, and FP represents the number of samples incorrectly predicted as positive. Precision (P) measures how many of the samples predicted as positive are actually positive. Recall (R) is calculated as follows:

[0095]

[0096] Where FN represents the number of samples that the model incorrectly predicted as negative. Recall (R) measures how many of the true positive samples were correctly predicted as positive by the model. The formula for calculating the F1 score is as follows:

[0097]

[0098] The F1 score is the harmonic mean of precision and recall, designed to balance these two metrics. It represents a trade-off between precision and recall, providing a comprehensive measure that considers both.

[0099] 1. Ablation test

[0100] To investigate the effectiveness of different modules in the named entity recognition model proposed in this invention, an ablation experiment was conducted on the power operation and maintenance dataset constructed in this invention. The experimental results are shown in Table 1.

[0101] Table 1 Ablation Experiment Results

[0102] Module P(%) R(%) F1(%) benchmark model 76.8 73.9 75.3 ① +Location-aware global attention mechanism module (does not use global memory unit) 80.5 78.3 79.4 ② +Location-aware global attention mechanism module 85.7 84.9 85.3 ③ + Fine-grained enhancement module 85.4 83.1 84.2 ④ + Position-aware global attention mechanism module + Fine-grained enhancement module 88.3 88.9 88.6 ⑤ +BERT-wwm-ext + Position-aware global attention mechanism module + Fine-grained enhancement module 92.8 91.9 92.3

[0103] The baseline model in Table 1 is Transformer+CRF, with an F1 score of 75.3%. The experimental results in the table show that adding a position-aware global attention mechanism module to the baseline model consistently improves all metrics. The use of a global memory unit further enhances performance, achieving an F1 score of 85.3%, a 5.9% improvement compared to the model without a global memory unit. This demonstrates the crucial role of dataset-scale global information modeling in named entity recognition tasks.

[0104] When the fine-grained enhancement module was added to the baseline model, the F1 score was 84.2%. Furthermore, the combined use of the position-aware global attention mechanism module and the fine-grained enhancement module yielded even better results, with an F1 score of 88.6%. The balanced improvement in precision (88.3%) and recall (88.9%) indicates that the two modules work together to enhance entity boundary detection and classification.

[0105] The complete model proposed in this invention achieves optimal performance with an F1 score of 92.3%, so the addition of the pre-trained model BERT-wwm-ext is also crucial. Figure 9 The training dynamics of the model over 50 iterations are shown with different modules used.

[0106] As shown in the figure, the use of each module improves the model's convergence speed, stability, and final performance to varying degrees. In summary, the superior prediction accuracy of the model proposed in this invention can be attributed to its ability to accurately capture long-range contextual dependencies and entity boundaries. In practical applications, this model can efficiently and accurately identify technical terms in power industry maintenance work orders.

[0107] 2. Confusion matrix heatmap

[0108] To intuitively evaluate the performance of the proposed model, a detailed comparison of its recognition results with a benchmark model across different entity categories was conducted. Figures 10(a) and 10(b) show the confusion matrix heatmaps of the two models, where the vertical axis represents the ground truth entity categories, the horizontal axis represents the entity categories predicted by the model, and the color from light to dark indicates a change in the prediction proportion from low to high. The value in each cell represents the corresponding prediction percentage. The diagonal cells represent the proportion of model prediction accuracy for each entity category, while the other cells represent the proportion of samples predicted as one category but actually belonging to other entity categories.

[0109] The comparison of the confusion matrix heatmaps shows that the proposed model significantly outperforms the baseline model in terms of recognition accuracy (values ​​on the main diagonal) across all entity categories. Furthermore, the off-diagonal elements reveal a significant reduction in the false recognition rate of the improved model.

[0110] Specifically, the entity "Operations and Maintenance Content" exhibits significant variations in text length and content diversity, making it prone to confusion with other entity categories and challenging to identify. The proposed simulation effectively addresses this issue, greatly enhancing the accuracy of "Operations and Maintenance Content" entity recognition. Furthermore, two consecutively occurring entity categories or those with similar text formats are easily confused, such as "affiliated substation" and "affiliated line," "equipment name" and "damaged part," and "voltage level" and "time." The proposed model significantly mitigates this problem. These performance enhancements can be attributed to the model's accurate feature extraction of complex long text semantics and entity boundaries.

[0111] Comparative Analysis of Predicted Examples

[0112] To visually verify the entity recognition capability of the proposed model in power equipment operation and maintenance work order text, especially its performance when handling domain-specific terms and complex entity structures, Table 2 presents three representative model prediction examples for comparative analysis. By comparing with the benchmark model, the actual effect of the proposed model is evaluated.

[0113] Table 2

[0114] Example 1 The communication of the monitoring and control device of the 110kV No.1 main transformer is not working. Correct label: <![CDATA[[flow] SUB [110 kV] VL [Main Transformer No. 1] EN [Monitoring and Control Device] DP [Communication Failure Check] MS。 > Baseline model: <![CDATA[[flow] SUB [110 kV] VL [Main Transformer 1 Monitoring and Control Device] EN [Communication [Check for Disconnection]] MS .]]> The proposed model: <![CDATA[[Drip] SUB [110 kV] VL [Main transformer No. 1] EN [Monitoring and control device] DP [Check for communication failure] MS .]]> Example 2 Handling the defect of damaged C-phase closing stop contact of the 110kV Chang 04 disconnector on the Yunchang Second Circuit. Correct label: <![CDATA[[Yunchang Second Circuit Line] LINE [110 kV] VL [Chang] SUB [04 Disconnector] EN Phase C [Closing Stop Contact] DP [Damage Defect Handling] MS 。]]> Baseline model: <![CDATA[[Yunchang Second Circuit Line] SUB [110 kV] VL [04 Disconnector of Chang] EN Closing of Phase C DP Stop [Treatment of Contact Damage Defect] MS .]]> The proposed model: <![CDATA[[Yunchang Second Circuit Line] LINE [110 kV] VL [Chang] SUB [04 Disconnector] EN Phase C [Closing Stop Contact] DP [Handling of Damage Defects] MS 。]]> Example 3 Defect elimination task: Eliminate the damage to the energy storage motor of the operating mechanism of the 110kV bus coupler Hua 23 circuit breaker at the QX23008 Huashan substation. Correct label: <![CDATA[Defect elimination task: Eliminate QX23008 [Huashan Substation] SUB [110kV] VL Bus coupler [Hua] SUB [23 Circuit breaker] EN [Operating mechanism energy storage motor] DP [Damage handling] MS 。]]> Baseline model: <![CDATA[Defect elimination task: Eliminate QX

[23008] TIME [Huashan Substation] SUB [110 kV] VL [Mother connection Huahua] SUB [23 Circuit breaker] EN Operating mechanism [Energy storage motor] DP [Damage handling] MS .]]> The proposed model: <![CDATA[Defect elimination task: Eliminate QX23008 [Huashan Substation] SUB [110kV] VL Bus coupler [Hua] SUB [23 Circuit breaker] EN [Operating mechanism energy storage motor] DP [Damage handling] MS .]]>

[0115] In Table 2, the content in [] represents entities, and the superscripts represent different entity categories. SUB represents the substation, VL represents the voltage level, EN represents the equipment name, DP represents the damaged part, MS represents the specific maintenance content, LINE represents the line, and TIME represents the time. These texts contain different sentence lengths and writing styles. In Example 1, the baseline model incorrectly identified the damaged part as part of the equipment name, and also failed to accurately identify the specific maintenance content. In Example 2, the baseline model failed to correctly identify the easily confused line and substation, as well as the easily distinguishable damaged part and specific maintenance content. In Example 3, the baseline model incorrectly identified irrelevant information in the same numerical format as time, and other entity boundaries were not accurately identified. However, the model proposed in this invention effectively avoids the above errors and has a better ability to distinguish entity categories that are easily confused semantically and formally. This indicates that the proposed model can accurately capture the location information, contextual information, and boundary information of entities, and understand the dependencies between entities.

[0116] The system in this embodiment of the invention is implemented using electronic devices; therefore, it is necessary to introduce the relevant electronic devices. For this purpose, the embodiments of the present invention provide an electronic device, such as… Figure 2 As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor can invoke logical instructions stored in the at least one memory to implement various systems provided in the system embodiments.

[0117] Furthermore, when the logical instructions in at least one of the aforementioned memories can be 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, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the system described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0118] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0119] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to implement the methods or systems described in or in part of the various embodiments.

[0120] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Based on this understanding, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0121] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Any expressions such as "predetermined threshold," "preset threshold," etc., without specifying a particular value, can be determined by those skilled in the art through simple experimentation or appropriate adjustments.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An information-enhanced named entity recognition system for power operation and maintenance text, characterized in that, include: The dataset construction module is used to correct characters in power equipment maintenance work orders, determine entity categories, and annotate entities; the word embedding module is used to convert words from discrete symbolic forms into continuous vector representations; the location-aware global attention mechanism module is used to perform location encoding and attention score calculation, and obtain correlation information between different samples; the fine-grained enhancement module is used to capture local information and more accurately identify entities in maintenance work orders; and the conditional random field module is used to construct a probabilistic framework and capture dependencies between labels. The visualization interaction and storage module is used for the visualization interaction of named entity recognition in power operation and maintenance text. The fine-grained enhancement module includes: ； ； ； Where X is the input sequence, This is the result of fusing the nonlinear features of the second convolutional layer with the features of the first convolutional layer as the main focus, capturing the local details in the input that are focused by different convolutional kernels; Norm is a layer normalization operation; This is the first one-dimensional convolutional layer; This is the second one-dimensional convolutional layer; Hadamard product symbol; GELU is the activation function; This is the result of fusing the nonlinear features of the first convolutional layer with the features of the second convolutional layer as the main driver, thus mining potential fine-grained semantic differences from another perspective; Out is the final output of the fine-grained enhancement module; This is the third one-dimensional convolutional layer; The process of performing position encoding and attention score calculation, and obtaining correlation information between different samples, includes: ； ； ； ； ； ； ； Where Q, K, and V are the query matrix, key matrix, and value matrix, respectively; X is the input sequence; , , It is a parameter matrix used to map the input X to different spaces; The relative position embedding vector between the two tokens; t is the current position; j is the position to be focused on; T is the matrix transpose symbol; i is the frequency index of the position encoding, with a value range of [0, d]. k / 2]; The dimension of the key vector; A positional awareness score is given between positions t and j. The query vector for the token at position t; Let be the key vector of the token at position j; u is the first learnable bias vector, which adjusts the attention based on the inherent properties of the token; v is the second learnable bias vector, which adjusts the attention based on the direction and distance between the two tokens, further enhancing the model's ability to capture spatial and sequential relationships. This provides global information; L-DNorm is a learnable dual normalization strategy. , These are two globally shared learnable memory units, where, This helps calculate similarity to generate attention weights. By combining with attention weights, the input features are updated, thereby affecting the final output; Attn is a position-aware global attention mechanism; softmax is the softmax function.

2. The information-enhanced named entity recognition system for power operation and maintenance text as described in claim 1, characterized in that, The dataset construction module includes: a preprocessing module for deduplicating the acquired data, removing content containing useless characters, missing information, and erroneous statements; an entity category determination module for determining the equipment name, voltage level, line, substation, damaged part, specific maintenance content, and time point; and an entity annotation module for annotating the text data using the BMES annotation system, where "B" represents the beginning of an entity, "M" represents the middle part of an entity, "E" represents the end of an entity, and "S" represents a single word entity.

3. The information-enhanced named entity recognition system for power operation and maintenance text as described in claim 2, characterized in that, The process of converting words from discrete symbolic forms to continuous vector representations includes: using BERT-wwm-ext to embed characters into the input sequence, which enhances the word embedding of the Chinese corpus. It consists of 12 Transformer layers, each with 768 hidden units and 12 attention heads.

4. The information-enhanced named entity recognition system for power operation and maintenance text as described in claim 1, characterized in that, The learnable dual normalization strategy includes: ； ； ； in, Let be the intermediate matrix element obtained through matrix multiplication, representing the initial associated value in the i-th row and j-th column; These are the normalized values ​​of the column; The learnable parameter for adjusting the strength of column normalization; exp is the exponential function; This represents the original association value between the nth row and the jth column; n is the row and column dimension index. This is the final output value after double normalization; This represents the value in the i-th row and n-th column of the intermediate matrix after column normalization. Learnable parameters for adjusting row normalization strength.

5. The information-enhanced named entity recognition system for power operation and maintenance text according to claim 4, characterized in that, The construction of the probabilistic framework and the capture of dependencies between tags include: ； Where P(Y|X) is the conditional probability; A quantification symbol for compatibility between adjacent tags; Assign a label to position i The appropriateness evaluation symbol; Y represents a complete label sequence; The set of all possible labeled sequences corresponding to the observed sequence X; It is the quantization symbol for the compatibility between the label at position i and the label at position i+1 in all possible label sequences in the denominator; The evaluation symbol for assigning a specific label at position i in all possible label sequences in the denominator.

6. The information-enhanced named entity recognition system for power operation and maintenance text according to claim 5, characterized in that, The visual interaction for power operation and maintenance text named entity recognition includes: inputting an operation and maintenance work order, automatically extracting the corresponding information, batch processing multiple operation and maintenance work orders, and batch exporting the recognition results.

7. An electronic device, characterized in that, include: At least one processor, at least one memory, a communication interface, and a bus; wherein, The processor, memory, and communication interface communicate with each other through the bus; The memory stores program instructions that can be executed by the processor, which invokes the program instructions to implement the system as described in any one of claims 1 to 6.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that enable the computer to implement the system as claimed in any one of claims 1 to 6.

Images