Intelligent document management method and system based on power plant data

Abstract

The invention discloses an intelligent document management method and system based on power plant data. The method comprises the steps that document information generated in the operation process of a power plant is acquired; the document information can comprise three types of information contents including texts, data and pictures; based on natural language processing and a neural network algorithm, performing multi-level vectorization processing on the document information; according to the vectorization result of each level, document association is carried out; performing classification simulation on the association result, and analyzing the fitness of classification; and screening the classification simulation result with the highest fitness, and managing the document. Manual intervention is reduced, the management cost is reduced, an efficient and intelligent data basis is provided for safe operation, operation and maintenance management and decision support of the power plant, and the method has good application and popularization prospects and actual economic value.

Description

Technical Field

[0001] This invention relates to the field of document management technology, specifically to an intelligent document management method and system based on power plant data. Background Technology

[0002] In the modern power industry, with the continuous advancement of automation and digitalization, power plants generate a large amount of structured and unstructured document information in their daily operation and management. These documents cover equipment operation records, alarm data, maintenance logs, operating procedures, inspection reports, and various images, tables, and monitoring data. The types of documents are diverse, originating from various sources, including text and tables manually filled out by staff, as well as electronic data and images generated in real time by automation systems and monitoring platforms. However, power plants still face many challenges in document management.

[0003] First, document collection and archiving methods are relatively traditional, with many steps relying on manual processing. This is not only inefficient but also prone to delays, omissions, or misclassifications, making it unsuitable for the rapid growth of power plant data. Second, with the diversification of document types and sources, information generated by different systems, platforms, and business processes lacks unified standards, making effective integration and centralized management difficult. Multimodal data such as text, tables, and images are often managed in a decentralized manner, resulting in significant information silos and hindering business collaboration and knowledge sharing.

[0004] Furthermore, most existing document management systems only possess basic classification and retrieval functions, lacking intelligent analysis of the deep semantics of document content and the ability to automatically correlate multimodal information. This makes it difficult to quickly extract key business information from large volumes of documents and fails to provide strong data support for practical applications such as equipment operation and maintenance, fault analysis, or safety decision-making. Given the increasingly complex equipment and stringent management requirements, how to efficiently collect, automatically integrate, intelligently analyze, and deeply utilize various document information from power plants has become a crucial issue in promoting the intelligent transformation of the power industry. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is that existing intelligent document management methods suffer from problems such as heterogeneous document types, weak automation and intelligent analysis capabilities, and insufficient information association.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent document management method based on power plant data, comprising:

[0008] Acquire document information generated during the operation of the power plant; the document information may include three types of information content: text, data, and images;

[0009] Based on natural language processing and neural network algorithms, the document information is processed into vectorization at multiple levels.

[0010] Based on the vectorization results at each level, documents are associated;

[0011] The association results are classified and simulated, and the fitness of the classification is analyzed; the classification simulation results with the highest fitness are selected for document management.

[0012] As a preferred embodiment of the intelligent document management method based on power plant data described in this invention, the document information includes: automatically collecting equipment operation data, alarm information, and system-generated reports by integrating power plant automation control systems, monitoring systems, or information platforms to form corresponding electronic documents, tables, and image information.

[0013] As a preferred embodiment of the intelligent document management method based on power plant data described in this invention, the vectorization process includes a multi-level division and a process of vectorizing the content of each level.

[0014] The multi-level division includes dividing the text content in the document into three levels: sentence, paragraph, and document; dividing the data content in the document into two levels: paragraph and document; and dividing the image content in the document into two levels: individual and document.

[0015] The text content is vector-encoded at three levels using a pre-trained natural language processing model.

[0016] After performing content analysis on the data and image content, the analysis results are vector-encoded.

[0017] As a preferred embodiment of the intelligent document management method based on power plant data described in this invention, the content analysis of the data content includes marking the data source, corresponding physical quantity, and timestamp on each data; and classifying data with the same data source, the same physical quantity, and consecutive timestamps as the same segment of data content according to the marked content.

[0018] The data content within the same segment is input into a pre-trained encoder to obtain the encoding result of the data content within the segment; the vectorization result of the data content at the document level is obtained through principal component analysis.

[0019] Content analysis of the image content includes inputting a single image into a pre-trained neural network to identify the image content and the device status within the image content, and generating an encoding result of the image content at the individual level based on the identification result; and obtaining a vectorized result of the image content at the document level through principal component analysis.

[0020] As a preferred embodiment of the intelligent document management method based on power plant data described in this invention, the step of associating documents according to the vectorization results at each level includes:

[0021] Step 1: Within the same document, analyze the correlation between any two vectors at any level in the document; after performing dimensionality reduction on the two vectors at the document level using PCA, calculate the Euclidean distance between the dimensionality reduction results to obtain the intermediate parameter for the correlation analysis between the two vectors; add one to the intermediate parameter and take the reciprocal as the correlation between the two vectors.

[0022] Step 2: Analyze the correlation between any two vectors at any level in different documents; after performing dimensionality reduction on the two vectors at the document level using PCA, perform dimensionality reduction again on the dimensionality reduction results in different documents, calculate the Euclidean distance between the dimensionality reduction results, and obtain the intermediate parameter for the correlation analysis between the two vectors. Add one to the intermediate parameter and take the reciprocal as the correlation between the two vectors.

[0023] Step 3: Based on the relationship between every two vectors, treat each vector as a node and the relationship as an edge to generate a relationship network;

[0024] Step 4: Using the Bayesian algorithm, allocate attention weights to the nodes in the relational network: After clustering the vectors in the documents, use the Bayesian algorithm to analyze whether each cluster center is the basis for classifying the corresponding document, calculate the prior probability, and obtain the prior probability of each cluster family; after normalizing the prior probabilities of all cluster centers, divide the normalized prior probability results corresponding to each family, analyze the contribution of each vector to the principal components within the family, and allocate the normalized prior probability results proportionally according to the size of each vector's contribution, obtaining the probability value corresponding to each vector as the attention weight of each vector.

[0025] As a preferred embodiment of the intelligent document management method based on power plant data described in this invention, the classification simulation includes clustering all reference points in each document using any node as a reference point, and the clustering result serves as a classification management scheme.

[0026] In a preferred embodiment of the intelligent document management method based on power plant data described in this invention, the fitness is measured using the attention weights of vectors and the correlations between vectors in the relational network after attention weight allocation.

[0027] Let two vectors a and b belong to two documents A and B respectively, and serve as reference points for documents A and B respectively. A and B are different documents.

[0028] Multiply the association between vector a and any vector i in document B by the attention weight of vector i to obtain the fitness value between vector a and vector i; after traversing all vectors in document B, sum the fitness values ​​to obtain the fitness of vector a and document B.

[0029] After calculating the fitness of vector b with document A, sum the fitness of vector a with document B and the fitness of vector b with document A to obtain the fitness of document A and document B in the same cluster.

[0030] After calculating the fitness of each pair of documents separately, the fitness within a family is averaged to obtain the fitness within the family; the fitness within each family is summed and averaged to obtain the fitness of each classification scheme.

[0031] An intelligent document management system based on power plant data, employing the method described in this invention, wherein:

[0032] The data acquisition unit acquires document information generated during the operation of the power plant; the document information may include three types of information content: text, data, and images.

[0033] The processing unit performs multi-level vectorization processing on the document information based on natural language processing and neural network algorithms;

[0034] The association unit associates documents based on the vectorization results at each level;

[0035] The classification unit performs classification simulations on the associated results and analyzes the fitness of the classification; it then selects the classification simulation results with the highest fitness and manages the documents.

[0036] A computer device includes: a memory and a processor; the memory stores a computer program, wherein: when the processor executes the computer program, it implements the steps of the method described in any one of the present invention.

[0037] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the present invention.

[0038] The beneficial effects of this invention are as follows: The intelligent document management method based on power plant data provided by this invention can automatically collect, standardize, and multi-level vectorize various types of document information from power plants, significantly improving the intelligence and automation level of document management. Through multimodal fusion and principal component analysis, it can effectively integrate heterogeneous content such as text, data, and images, achieving deep correlation and intelligent classification of document content, greatly improving retrieval efficiency and knowledge discovery capabilities. The introduction of Bayesian algorithms and attention mechanisms can dynamically allocate document node weights, improving the accuracy and interpretability of clustering, archiving, and intelligent recommendation. The overall solution reduces manual intervention and management costs, providing an efficient and intelligent data foundation for power plant safe operation, maintenance management, and decision support, and has good application prospects and practical economic value. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 The first embodiment of the present invention provides an overall flowchart of an intelligent document management method based on power plant data. Detailed Implementation

[0041] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0042] Example 1, referring to Figure 1 As an embodiment of the present invention, an intelligent document management method based on power plant data is provided, comprising:

[0043] S1: Obtain document information generated during the operation of the power plant; the document information may include three types of information content: text, data, and images.

[0044] Furthermore, the document information includes the automatic collection of equipment operation data, alarm information, and system-generated reports through the integration of power plant automation control systems, monitoring systems, or information platforms, forming corresponding electronic documents, tables, and image information.

[0045] By integrating automated control systems, monitoring systems, and information platforms, key document content such as equipment operation data, alarm information, and system reports can be acquired in real time, covering various formats including text, tables, and images. This ensures data integrity and timeliness, reduces the workload of manual data entry and archiving, and avoids information omissions and misalignments, laying a solid data foundation for subsequent standardized processing, intelligent analysis, and in-depth utilization. It significantly improves the automation and standardization of document management, meeting the needs of efficient document collection and accurate management in large-scale, complex power plant scenarios.

[0046] S2: Based on natural language processing and neural network algorithms, the document information is processed into vectorization at multiple levels.

[0047] The vectorization process includes multi-level partitioning and vectorizing the content of each level.

[0048] Furthermore, the multi-level segmentation includes dividing the text content in the document into three levels: sentence, paragraph, and document; dividing the data content in the document into two levels: paragraph and document; and dividing the image content in the document into two levels: individual and document. Dividing the text content into sentence, paragraph, and document levels allows for the capture of fine-grained information (such as operating steps and anomaly descriptions), meso-level information structure (such as process flow and event sequence), and the overall theme and intent of the document. This facilitates more accurate semantic understanding, content retrieval, and contextual reasoning. Dividing the data content into paragraph and document levels allows for the grasp of the data's temporal sequence, phased patterns, and overall characteristics. Data "segments" are typically grouped together based on data source, physical quantity, and collection time, automatically grouping data with consistent sources, identical physical quantities, and continuous time. This segmentation helps reveal the phased changes in equipment operation, operating condition characteristics, and their correlation with events, facilitating subsequent local analysis and comprehensive modeling of data trends and anomaly fragments. Image content is divided into two levels: individual and document. This allows for focusing on the information contained in a single image (such as equipment status or fault location) as well as integrating multiple images to reflect the overall business scenario or complex event process. This approach facilitates both local recognition and global perception, as well as collaborative analysis of multimodal content. In summary, this multi-level division enhances the expressiveness and flexibility of content modeling, enabling intelligent analysis, feature fusion, and business decision-making for different scenarios and levels.

[0049] This approach utilizes pre-trained natural language processing models to perform vector encoding on text content at three levels. Specifically, at the sentence level, each sentence is input into a pre-trained model such as BERT or RoBERTa to extract sentence-level semantic vectors, enabling fine-grained content representation. At the paragraph level, paragraph-level semantic vectors are obtained by fusing sentence vectors within the same paragraph or by utilizing paragraph-level context modeling, better reflecting the paragraph's theme and structure. At the document level, all paragraph-level vectors are further aggregated, and document-level semantic modeling yields the overall semantic expression of the document. This hierarchical vector encoding method simultaneously considers both the detailed information and overall semantics of the text, providing a high-quality feature foundation for subsequent document association, intelligent retrieval, and content aggregation.

[0050] After performing content analysis on the data and image content, the analysis results are vector-encoded. Specifically, the content analysis of the data content includes marking the data source, corresponding physical quantity, and timestamp on each data point; according to the marked content, data from the same data source, with the same physical quantity, and consecutive timestamps are considered as the same data segment.

[0051] The data content within the same segment is input into a pre-trained encoder to obtain the encoded result of the data content within the segment. Principal component analysis (PCA) is used to obtain the vectorized result of the data content at the document level. The encoder can generate vectors from the following dimensions: data source, corresponding physical quantity, trend, eigenvalues ​​(preset mean, standard deviation, variance, etc.), peak frequency, etc. Besides using a neural network-based model, the encoder can also employ traditional or unsupervised feature extraction methods such as autoencoders and PCA. In specific implementations, different types of encoders can be flexibly selected to compress and vectorize the data within the segment based on the data structure and actual needs, thereby obtaining a representative low-dimensional representation. This design not only improves the applicability of the method but also enhances the flexibility and robustness of the system in various business scenarios.

[0052] Content analysis of the images involves inputting a single image into a pre-trained neural network to identify the image content and device status within it, generating an encoding result at the individual level based on the identification results, and obtaining a vectorized result at the document level through principal component analysis. By intelligently recognizing single images, not only can device information be extracted from the images, but also key business information such as device operating status and fault conditions can be automatically identified. Encoding image content using a neural network and further integrating the features of multiple images through principal component analysis achieves a vectorized representation of image information at the document level. This method efficiently transforms large amounts of image information into computable and searchable structured features, providing a solid data foundation for comprehensive document analysis, intelligent association, and automatic classification, thereby improving the automation and intelligence level of document management.

[0053] In this embodiment, the neural network is preferably a deep learning model such as a Convolutional Neural Network (CNN) or a Visual Transformer (ViT), which can automatically learn multi-level features of the image to effectively identify and represent image content such as device type and status. Optionally, the neural network can also employ a deep model with transfer learning, a multimodal fusion network, or a composite model combining object detection and image segmentation algorithms. Depending on the actual application scenario and the complexity of the image content, other deep model structures such as autoencoders and generative adversarial networks (GANs) can also be selected to further improve the representation ability and recognition accuracy of the image content.

[0054] S3: Based on the vectorization results at each level, associate the documents.

[0055] Step 1: Within the same document, analyze the correlation between any two vectors at any level. After performing dimensionality reduction on both vectors at the document level using PCA, calculate the Euclidean distance between the dimensionality reduction results to obtain an intermediate parameter for the correlation analysis between the two vectors. Add one to this intermediate parameter and take its reciprocal as the correlation between the two vectors (the document-level vectors obtained in step S2 through principal component analysis can be directly used in the PCA dimensionality reduction process, simplifying the computation). Dimensionality reduction using Principal Component Analysis (PCA) not only removes redundant information and highlights key features but also maps high-dimensional features of different types and levels to a comparable low-dimensional space. Based on this, using Euclidean distance as a quantitative indicator of the difference between vectors, and then converting the distance into correlation by adding one and taking its reciprocal, ensures that the measurement result is always positive, and that vectors that are closer and more related have a higher correlation. This design provides a solid theoretical foundation and efficient technical means for structured modeling of document content, information fusion, and subsequent intelligent aggregation, retrieval, and recommendation functions. It also enhances the scalability and practicality of the system when processing multimodal and multi-level document content.

[0056] Step Two: Within different documents, analyze the correlation between any two vectors at any level within the documents. After performing dimensionality reduction on the two vectors at the document level using Principal Component Analysis (PCA), the dimensionality reduction results are then further reduced across different documents. The Euclidean distance between the dimensionality reduction results is calculated to obtain an intermediate parameter for the correlation analysis between the two vectors. This intermediate parameter is incremented by one, and the reciprocal is taken as the correlation between the two vectors. (These two dimensionality reductions are equivalent to first mapping the lower-level vectors to low-dimensional "document-level" vectors; then, the principal components of the "document-level" vectors are mapped again across different documents, thus making them comparable at both the document and overall levels.) Through two rounds of Principal Component Analysis (PCA) dimensionality reduction, the high-dimensional vectors at the lower level are first uniformly mapped to the low-dimensional space of their respective document levels, eliminating the influence between different content levels and modalities, and standardizing the vector representation within the same document. Subsequently, these document-level vectors are further dimensionality reduced across different documents, further normalizing and aligning the representations of each document, ensuring consistency and comparability of their feature distribution at the overall level. Building upon this foundation, Euclidean distance is employed for quantitative measurement, and the reciprocal is added to convert it into a correlation index. This not only enhances the intuitiveness and practicality of the measurement but also ensures that all comparisons are based on a unified space and standard. This design effectively strengthens the capabilities for structured comparison of information across different documents, intelligent clustering, and deep semantic analysis, providing solid technical support for cross-document content fusion, knowledge discovery, and intelligent recommendation.

[0057] Step 3: Based on the relationship between every two vectors, treat each vector as a node and the relationship as an edge to generate a relationship network.

[0058] Step 4: Using the Bayesian algorithm, attention weights are assigned to nodes in the relational network. After clustering the vectors in the documents, the Bayesian algorithm is used to analyze whether each cluster center is the basis for classifying the corresponding document, calculating the prior probability to obtain the prior probability of each cluster family. After normalizing the prior probabilities of all cluster centers, the normalized prior probability results for each family are distributed, analyzing the contribution of each vector to the principal components within the family. Based on the magnitude of each vector's contribution, the normalized prior probability results are distributed proportionally, obtaining the probability value corresponding to each vector as the attention weight for each vector. This method can more scientifically reflect the influence and importance of different document content and features in the relational network, strengthen the focus on key information in the structure, and improve the accuracy and interpretability of overall content analysis, classification, and intelligent recommendation. Simultaneously, it provides a quantitative theoretical basis and technical foundation for subsequent document fitness assessment, classification simulation, and management decision-making.

[0059] The contribution calculation process includes: principal component analysis (PCA) to extract principal components; calculating the projection of each vector onto the principal component direction; normalizing the projection results to obtain the contribution.

[0060] S4: Perform classification simulation on the association results and analyze the fitness of the classification; select the classification simulation results with the highest fitness and manage the documents.

[0061] The classification simulation involves clustering all reference points within each document, using any node as a reference point. The clustering result serves as a classification management scheme. By performing full-range clustering operations with any node as a reference point, multiple different document classification management schemes can be simulated and generated. This allows for a comprehensive exploration of the organizational structure and inherent relationship patterns of document content from different reference perspectives, avoiding limitations or information omissions caused by a single classification standard. Each clustering result obtained with different nodes as reference points can be considered as an alternative classification management method. By clustering all reference points and forming multiple classification management schemes, the optimal or most suitable document classification structure for business needs can be selected in the subsequent fitness evaluation stage.

[0062] In the relational network after attention weight allocation, the fitness is measured using the attention weights of vectors and the relationships between vectors:

[0063] Let two vectors a and b belong to two documents A and B respectively, and serve as reference points for documents A and B respectively. A and B are different documents.

[0064] Multiply the association between vector a and any vector i in document B by the attention weight of vector i to obtain the fitness value between vector a and vector i; after traversing all vectors in document B, sum the fitness values ​​to obtain the fitness of vector a and document B.

[0065] After calculating the fitness of vector b with document A, sum the fitness of vector a with document B and the fitness of vector b with document A to obtain the fitness of document A and document B in the same cluster.

[0066] After calculating the fitness of each pair of documents separately, the fitness within a family is averaged (by summing the fitness of any two documents within a family and dividing by the number of times any two documents are enumerated) to obtain the fitness within the family; the fitness within each family is summed and averaged to obtain the fitness of each classification management scheme.

[0067] To address this, quantitative and interpretable indicators are introduced into the evaluation process of classification management schemes to systematically measure the strength of content associations and overall clustering effects among different documents. By combining the attention weight of each vector with its relevance in the relationship network, the actual degree of association between document content and the influence of core information can be comprehensively reflected. Specifically, weighted summation of the relevance of reference points to all nodes of other documents not only highlights the dominant role of key content but also eliminates the interference of noise or marginal information on the evaluation results. Using bidirectional fitness metrics (i.e., document A to B and B to A) and averaging all document pairs avoids the randomness and bias caused by a single perspective, improving the objectivity and comprehensiveness of the scheme evaluation. Finally, by uniformly quantifying and comparing the fitness of each classification management scheme, the optimal classification structure can be efficiently and automatically selected. This significantly improves the scientific rigor, accuracy, and adaptive capabilities of intelligent document management, contributing to the most rational document archiving, retrieval, and knowledge discovery.

[0068] Example 2 provides an intelligent document management system based on power plant data, which includes:

[0069] The data acquisition unit acquires document information generated during the operation of the power plant; the document information may include three types of information content: text, data, and images.

[0070] The processing unit performs multi-level vectorization processing on the document information based on natural language processing and neural network algorithms.

[0071] The association unit associates documents based on the vectorization results at each level.

[0072] The classification unit performs classification simulations on the associated results and analyzes the fitness of the classification; it then selects the classification simulation results with the highest fitness and manages the documents.

[0073] If the above functions 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, the technical solution of this 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 steps of the methods described in the various embodiments of this 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.

[0074] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0075] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0076] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0077] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart document management method based on power plant data, characterized in that, include: Obtain document information generated during the operation of the power plant; The document information may include three types of information content: text, data, and images. Based on natural language processing and neural network algorithms, the document information is processed into vectorization at multiple levels. Based on the vectorization results at each level, documents are associated; The association results are classified and simulated, and the fitness of the classification is analyzed; the classification simulation results with the highest fitness are selected for document management.

2. The intelligent document management method based on power plant data as described in claim 1, characterized in that: The document information includes electronic documents, tables, and images generated by automatically collecting equipment operation data, alarm information, and system reports through the integration of power plant automation control systems, monitoring systems, or information platforms.

3. The intelligent document management method based on power plant data as described in claim 2, characterized in that: The vectorization process includes multi-level partitioning and vectorizing the content of each level; The multi-level division includes dividing the text content in the document into three levels: sentence, paragraph, and document; dividing the data content in the document into two levels: paragraph and document; and dividing the image content in the document into two levels: individual and document. The text content is vector-encoded at three levels using a pre-trained natural language processing model. After performing content analysis on the data and image content, the analysis results are vector-encoded.

4. The intelligent document management method based on power plant data as described in claim 3, characterized in that: Content analysis of the data includes marking the data source, corresponding physical quantity, and timestamp on each data point; According to the marking content, data from the same data source, the same physical quantity, and consecutive timestamps are regarded as the same data segment; The data content within the same segment is input into a pre-trained encoder to obtain the encoding result of the data content within the segment; the vectorization result of the data content at the document level is obtained through principal component analysis. Content analysis of the image content includes inputting a single image into a pre-trained neural network to identify the image content and the device status within the image content, and generating an encoding result of the image content at the individual level based on the identification result; and obtaining a vectorized result of the image content at the document level through principal component analysis.

5. The intelligent document management method based on power plant data as described in claim 4, characterized in that: The process of associating documents based on the vectorization results at each level includes: Step 1: Within the same document, analyze the correlation between any two vectors at any level in the document; after performing dimensionality reduction on the two vectors at the document level using PCA, calculate the Euclidean distance between the dimensionality reduction results to obtain the intermediate parameter for the correlation analysis between the two vectors; add one to the intermediate parameter and take the reciprocal as the correlation between the two vectors. Step 2: Analyze the correlation between any two vectors at any level in different documents; after performing dimensionality reduction on the two vectors at the document level using PCA, perform dimensionality reduction again on the dimensionality reduction results in different documents, calculate the Euclidean distance between the dimensionality reduction results, and obtain the intermediate parameter for the correlation analysis between the two vectors. Add one to the intermediate parameter and take the reciprocal as the correlation between the two vectors. Step 3: Based on the relationship between every two vectors, treat each vector as a node and the relationship as an edge to generate a relationship network; Step 4: Using the Bayesian algorithm, allocate attention weights to the nodes in the relational network: After clustering the vectors in the documents, use the Bayesian algorithm to analyze whether each cluster center is the basis for classifying the corresponding document, calculate the prior probability, and obtain the prior probability of each cluster family; after normalizing the prior probabilities of all cluster centers, divide the normalized prior probability results corresponding to each family, analyze the contribution of each vector to the principal components within the family, and allocate the normalized prior probability results proportionally according to the size of each vector's contribution, obtaining the probability value corresponding to each vector as the attention weight of each vector.

6. The intelligent document management method based on power plant data as described in claim 5, characterized in that: The classification simulation includes clustering all reference points in each document using any node as a reference point, and using the clustering results as a classification management scheme.

7. The intelligent document management method based on power plant data as described in claim 6, characterized in that: In the relational network after attention weight allocation, the fitness is measured using the attention weights of vectors and the relationships between vectors: Let two vectors a and b belong to two documents A and B respectively, and serve as reference points for documents A and B respectively. A and B are different documents. Multiply the association between vector a and any vector i in document B by the attention weight of vector i to obtain the fitness value between vector a and vector i. After traversing all vectors in document B, sum the fitness values ​​to obtain the fitness of vector a with respect to document B. After calculating the fitness of vector b with document A, sum the fitness of vector a with document B and the fitness of vector b with document A to obtain the fitness of document A and document B in the same cluster. After calculating the fitness of each pair of documents separately, the fitness within a family is averaged to obtain the fitness within the family; the fitness within each family is summed and averaged to obtain the fitness of each classification scheme.

8. An intelligent document management system based on power plant data, employing the method described in any one of claims 1-7, characterized in that: The data acquisition unit acquires document information generated during the operation of the power plant; the document information may include three types of information content: text, data, and images. The processing unit performs multi-level vectorization processing on the document information based on natural language processing and neural network algorithms; The association unit associates documents based on the vectorization results at each level; The classification unit performs classification simulations on the associated results and analyzes the fitness of the classification; it then selects the classification simulation results with the highest fitness and manages the documents.

9. A computer device, comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images