An intelligent verification method, device, equipment and product for power equipment procurement parameters

Abstract

The application relates to the technical field of power equipment, and discloses a power equipment procurement parameter intelligent checking method, device, equipment and product, the method comprising the following steps: obtaining a plurality of standard specification unstructured documents and procurement files of a target power equipment, and constructing a dynamic knowledge base and determining the equipment category label of the target power equipment; based on the procurement files and the equipment category label, the core context fragments of the core parameter chapters in the procurement files are obtained through field adaptive BERT model and named entity recognition method processing; based on the dynamic knowledge base, the equipment category label and the core context fragments, the parameter checking method based on the dynamic prompt engineering and the large model logical reasoning is used to determine whether the parameters of the procurement files are abnormal, the abnormal results are bound with the procurement files through data association, and a procurement parameter structured early warning report table of the target power equipment is generated, so that the checking processing efficiency, the parameter recognition accuracy and the complex parameter logical conflict recognition capability are considered.

Description

Technical Field

[0001] This invention relates to the field of power equipment technology, specifically to a method, device, equipment, and product for intelligent verification of power equipment procurement parameters. Background Technology

[0002] With the continuous expansion of power grid construction, the procurement volume of power equipment has increased dramatically. The compliance review of technical parameters in procurement documents has become a crucial link in ensuring the safe operation of the power grid. Currently, traditional equipment parameter review mainly relies on manual labor. However, due to the wide variety of power equipment (such as transformers, GIS, switchgear, etc.), the complex technical standards and specifications involved, and the large amount of data, procurement documents are usually in unstructured PDF or Word document format, leading to the following significant drawbacks of manual review: Inefficient and error-prone: Manually checking parameters against thick standard specifications is time-consuming and labor-intensive, and has a high rate of missed or incorrect detections.

[0003] Lagging updates to technical specifications and standards for power equipment: National and industry standards related to technical specifications for power equipment are updated frequently, making it difficult for reviewers to keep abreast of the latest standard changes, which can easily lead to problems of conducting inspections based on outdated standards.

[0004] Unstructured data processing is difficult: Traditional information technology methods are insufficient to accurately extract core data such as "technical parameter tables" from tender documents with complex formats and varying layouts. Furthermore, the way parameters are described in different documents (such as unit conversion and alias usage) varies. Conventional keyword matching or regular expression methods have weak generalization ability and are difficult to cope with complex contexts.

[0005] Lack of logical correlation verification: Existing technologies focus more on whether a single parameter exceeds the limit, and lack the ability to intelligently reason about the implicit logical relationships between parameters (such as the correspondence between "rated capacity, connection group and short-circuit impedance").

[0006] Therefore, there is an urgent need for a solution that can automatically build a standard knowledge base, intelligently parse procurement documents, and perform in-depth logical comparisons. Summary of the Invention

[0007] This invention provides a method, device, equipment, and product for intelligent verification of procurement parameters for power equipment, in order to solve the problem that existing methods for reviewing and verifying procurement documents for power equipment rely on manual review and cannot automatically build a standard knowledge base or intelligently parse procurement documents.

[0008] In a first aspect, the present invention provides an intelligent verification method for power equipment procurement parameters, the method comprising: The process involves: acquiring multiple unstructured standard and specification documents and procurement documents for the target power equipment; constructing a dynamic knowledge base based on these documents; processing the procurement documents using a pre-defined classification and fusion decision-making method to obtain equipment category tags for the target power equipment; processing the procurement documents and equipment category tags using a domain-adaptive BERT model and named entity recognition method to obtain core context fragments for the core parameter sections of the procurement documents; using the dynamic knowledge base, equipment category tags, and core context fragments, and employing a parameter verification method based on dynamic prompting engineering and large-scale model logical reasoning to determine whether the parameters in the procurement documents are abnormal; and when abnormal parameters are determined, associating and binding the abnormal results with the procurement documents, and generating a structured early warning report for the procurement parameters of the target power equipment.

[0009] The intelligent verification method for power equipment procurement parameters provided by this invention, by acquiring standard specification documents and procurement documents, can completely aggregate the authoritative benchmark data sources and business documents to be verified, unifying the input entry points for the two types of documents and eliminating the pre-processing barriers caused by the dispersion and inconsistent formats of multiple source documents. Furthermore, by constructing a dynamic knowledge base using standard documents, it can transform scattered, unstructured industry standards into unified, searchable, and inferable authoritative parameter benchmarks, solving the problems of lagging standard document updates, low efficiency of manual standard review, and inconsistent standard bases. Furthermore, by obtaining equipment category tags through a preset classification fusion decision method, it achieves accurate differentiation of equipment types by fusing the metadata rules and textual semantic features of procurement documents, thereby automatically generating differentiated verification indexes, avoiding the verification logic confusion caused by adapting unified verification rules to multiple types of equipment, and realizing customized verification scheduling for different equipment. Furthermore, by combining a domain-adaptive BERT model and named entity recognition methods to extract core context fragments, it can filter invalid noise content in procurement documents, thereby accurately locating the target equipment parameter paragraphs, suppressing text illusions, context overflow, and cross-equipment parameter confusion problems during large model inference, and significantly reducing subsequent inference interference. Furthermore, by combining a dynamic knowledge base, equipment tags, and core context-driven large-model logical reasoning verification, this invention overcomes the limitations of traditional single-parameter numerical comparison. It simultaneously achieves parameter semantic alignment, single-point threshold verification, and multi-parameter relational logical deduction, enabling dual identification of both surface-level parameter deviations and deep-level logical contradictions, thus improving the depth and comprehensiveness of verification. Finally, abnormal data is correlated with procurement document metadata to generate structured early warning reports, connecting verification results with business information and forming a closed-loop business process of verification, early warning, and rectification, reducing the cost of problem location and rectification for procurement personnel. Therefore, by implementing this invention, the fully manual item-by-item verification model is replaced, solving multiple industry pain points such as low efficiency of manual review, false positives and false negatives, untimely standard updates, reliance on single-value comparisons, and lack of traceability support for anomalies. It balances verification processing efficiency, parameter identification accuracy, and the ability to identify complex parameter logical conflicts, and is adaptable to complex procurement tender document scenarios with multiple equipment chapters.

[0010] In one alternative implementation, a dynamic knowledge base is constructed based on multiple standard specification unstructured documents, including: Using optical character recognition (OCR) technology, each standard specification unstructured document is parsed to obtain multiple original standard specification structured documents. Redundancy is removed from each original standard specification structured document to obtain multiple first standard specification structured documents. Using a preset standardized mapping function, each first standard specification structured document is normalized to obtain multiple second standard specification structured documents. The multiple second standard specification structured documents are reviewed and verified to obtain multiple target standard specification structured documents. The parameters in each target standard specification structured document are decomposed, and a dynamic knowledge base is constructed.

[0011] The intelligent verification method for power equipment procurement parameters provided by this invention uses optical character recognition (OCR) technology to parse unstructured standard and specification documents, overcoming the problem that image-based standard and specification documents such as PDFs cannot be read by machines. This allows for the batch conversion of unstructured standard text and parameter tables into computable structured raw data, opening up a machine-readable channel for standard documents. Furthermore, through redundancy removal, duplicate, incomplete, and invalid standard parameter entries are eliminated, simplifying the baseline data volume and reducing the computational overhead of subsequent vector storage, parameter matching, and logical calculations, thus avoiding interference from redundant data with parameter comparison accuracy. Furthermore, by using a preset standardized mapping function for normalization, parameter expressions of different units and magnitudes are unified, eliminating comparison deviations caused by inconsistencies in units and magnitudes between procurement documents and industry standards, ensuring the uniqueness of the parameter value comparison benchmark. Furthermore, through review and verification, typos, conversion errors, and omissions in standard clauses caused by OCR errors can be corrected, ensuring the authority and accuracy of parameter thresholds and logical rules within the dynamic knowledge base, eliminating verification misjudgments caused by distorted baseline data from the source. Furthermore, by constructing a dynamic knowledge base through parameter splitting and separate storage, it is possible to separate and schedule rule retrieval and semantic matching, while also possessing the dual capabilities of precise numerical querying and fuzzy alias parameter recognition. Therefore, by implementing this invention, the problems of fragmented standards, chaotic units, inability to be reused by machines, and inability to match alias parameters in traditional methods are completely solved.

[0012] In one optional implementation, based on the procurement documents and equipment category tags, and processed by a domain-adaptive BERT model and named entity recognition method, the core context fragment of the core parameter section in the procurement documents is obtained, including: Based on equipment category tags, the domain-adaptive BERT model and named entity recognition method are used to calculate the comprehensive matching degree value between the chapter title text and the preset core keyword text in the procurement document. When the comprehensive matching degree value is greater than the preset matching threshold, the chapter corresponding to the chapter title text is determined to be the core parameter chapter of the target power equipment, and the core context fragment of the core parameter chapter is extracted.

[0013] The intelligent verification method for power equipment procurement parameters provided by this invention calculates the comprehensive matching degree of chapters and keywords through multi-model joint calculation. It integrates power-specific semantic features and equipment entity constraints, taking into account both text semantic similarity and equipment attribution constraints, thus avoiding cross-equipment chapter mis-extraction caused by relying solely on keyword matching. Furthermore, if the comprehensive matching degree value is greater than a preset matching threshold, the chapter corresponding to the chapter title text is determined to be the core parameter chapter of the target power equipment. The core context fragments of the core parameter chapter are extracted, automatically filtering out irrelevant business chapters and other equipment parameter paragraphs, retaining only the target equipment parameter text fragments. This helps to significantly reduce the length of the input text for large models, reduce inference computational power consumption, and avoid verification misjudgments caused by cross-equipment parameter confusion from the source. Therefore, by implementing this invention, intelligent and accurate filtering of target equipment parameter chapters is achieved. While retaining complete parameter information, invalid noise text is removed, solving the defects of general text matching algorithms that easily confuse different equipment chapters and cause illusions from long text inputs of large models. It provides clean and low-interference input corpus for parameter extraction of large models, simultaneously improving the speed and accuracy of parameter extraction.

[0014] In one optional implementation, based on equipment category labels, a domain-adaptive BERT model and named entity recognition method are used to calculate the comprehensive matching score between the chapter title text and the preset core keyword text in the procurement document, including: Using a domain-adaptive BERT model, the chapter title text and the preset core keyword text in the procurement document are encoded to obtain enhanced semantic vectors for the chapter titles and text. Based on these enhanced semantic vectors, the multi-scale basic semantic similarity value between the chapter title text and the preset core keyword text is calculated. Named entity recognition is used to extract equipment entity words from the chapter title text. Based on the equipment entity words and equipment category labels, an entity penalty factor is determined. Finally, based on the multi-scale basic semantic similarity value and the entity penalty factor, the comprehensive matching degree value between the chapter title text and the preset core keyword text is determined.

[0015] The intelligent verification method for power equipment procurement parameters provided by this invention encodes and generates two types of enhanced semantic vectors using a domain-adaptive BERT model. It integrates shallow lexical features and deep global semantics, accurately capturing subtle semantic differences in power industry terminology and parameter aliases, significantly improving the accuracy of short text matching in the industry. Furthermore, by calculating multi-scale basic semantic similarity values, it quantifies the semantic closeness between chapter titles and parameter keywords, objectively measuring text content relevance and replacing fuzzy keyword hitting rules, achieving flexible semantic matching in scenarios without fixed keywords. Furthermore, it uses named entity recognition to extract equipment entity words from chapters, achieving accurate identification of equipment category entities within titles. Furthermore, by combining equipment entity words with equipment tag matching to determine a penalty factor, it can apply a matching degree attenuation penalty to non-target equipment chapters, thereby proactively lowering the matching score of cross-equipment chapters and preventing other equipment chapters from being mistakenly identified as target parameter chapters. Furthermore, a comprehensive matching degree is obtained by combining multi-scale basic semantic similarity values ​​and entity penalty factors, while taking into account both text semantic relevance and device attribution constraints. This balances the flexibility of semantic matching with the rigid constraints of device classification, avoiding missed or incorrect parameter chapters caused by single-dimensional judgment.

[0016] In one optional implementation, based on a dynamic knowledge base, equipment category tags, and core context fragments, a parameter verification method based on dynamic prompting engineering and large-scale model logical reasoning is used to determine whether the parameters of the procurement documents are abnormal, including: Based on a dynamic knowledge base, equipment category tags, and core context fragments, a structured parameter set is obtained through large language model processing. Based on the structured parameter set, a comprehensive anomaly index value is calculated. When the comprehensive anomaly index value is greater than a preset judgment threshold, the parameters of the procurement document are determined to be abnormal. When the comprehensive anomaly index value is less than or equal to the preset judgment threshold, the parameters of the procurement document are determined to be normal.

[0017] The intelligent verification method for power equipment procurement parameters provided by this invention transforms free-format tender documents into standardized key-value pairs using a large language model, eliminating the need for manual parameter table compilation. Furthermore, it calculates a comprehensive anomaly index value based on a structured parameter set, converting discrete parameter issues into comparable numerical indicators and achieving quantitative grading of anomaly levels. Moreover, it compares the comprehensive anomaly index value with a preset judgment threshold to determine whether the parameters in the procurement documents are abnormal. By setting a unified quantitative judgment standard, it avoids subjective judgment differences during manual review, achieving standardized and automated anomaly judgment without the need for manual line-by-line subjective screening of parameter issues. Therefore, by implementing this invention, objective, unified, and multi-dimensional automatic parameter compliance judgment is achieved, significantly reducing errors from manual subjective review and improving the throughput of batch document verification.

[0018] In one optional implementation, the comprehensive anomaly index value is calculated based on a structured set of parameters, including: Obtain the original semantic encoding vector of each structured parameter in the structured parameter set; based on the principle of maximum mapping confidence, map and match each original semantic encoding vector with the semantic vector of each parameter in the dynamic knowledge base to obtain the mapping parameter set; obtain the subset of parameters in the mapping parameter set that have implicit physical or standard constraints; use the logical mapping function in the dynamic knowledge base to calculate the logical conflict degree value of the parameter combination in the parameter subset of the dynamic knowledge base; calculate the comprehensive anomaly index value based on the logical conflict degree value.

[0019] The intelligent verification method for power equipment procurement parameters provided by this invention extracts the original semantic vectors of parameters output by a large model, preserving the original semantic features of the parameters and providing a basic feature carrier for cross-knowledge base vector space alignment. This method can adapt to scenarios where the encoding spaces of the large model and the standard vector library are inconsistent. Furthermore, mapping matching is performed based on the principle of the maximum mapping confidence, eliminating semantic offset between the large model and the standard knowledge base. Simultaneously, entity masks are used to shield parameters irrelevant to different devices, enabling accurate matching of parameter aliases and similar parameters with different expressions, solving the problem of comparison failures caused by inconsistent parameter naming. Furthermore, by filtering a subset of parameters with associated constraints, parameter combinations with physical and standard linkage relationships can be automatically identified, avoiding invalid logical calculations for unrelated parameters and reducing computational waste. Furthermore, by using logical mapping functions in the dynamic knowledge base to calculate logical conflict degree values, the degree to which parameter combinations deviate from industry-mandated linkage standards is quantified, thereby capturing hidden errors where a single numerical value is compliant but the parameter combination does not conform to the specifications, compensating for the blind spots of single-point threshold verification. Furthermore, by combining logical conflict degree values ​​to calculate a comprehensive anomaly index value, ordinary numerical deviations and high-risk parameter combination contradictions can be distinguished, achieving hierarchical quantification of anomaly severity.

[0020] Secondly, the present invention provides an intelligent verification device for power equipment procurement parameters, the device comprising: The acquisition module is used to acquire multiple standard specification unstructured documents and procurement documents for the target power equipment; the construction module is used to build a dynamic knowledge base based on multiple standard specification unstructured documents. The first processing module, based on the procurement documents, processes them using a pre-defined classification and fusion decision-making method to obtain equipment category tags for the target power equipment. The second processing module, based on the procurement documents and equipment category tags, processes them using a domain-adaptive BERT model and named entity recognition method to obtain core context fragments for the core parameter sections of the procurement documents. The judgment module, based on a dynamic knowledge base, equipment category tags, and core context fragments, uses a parameter verification method based on dynamic prompting engineering and large-scale model logical reasoning to determine whether the parameters in the procurement documents are abnormal. The generation module, when the parameters in the procurement documents are determined to be abnormal, associates and binds the abnormal results with the procurement documents and generates a structured early warning report of the procurement parameters for the target power equipment.

[0021] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the intelligent verification method for power equipment procurement parameters described in the first aspect or any corresponding embodiment.

[0022] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the intelligent verification method for power equipment procurement parameters described in the first aspect or any corresponding embodiment thereof.

[0023] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the intelligent verification method for power equipment procurement parameters described in the first aspect or any corresponding embodiment. Attached Figure Description

[0024] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific 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 from these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the intelligent verification method for power equipment procurement parameters according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the intelligent verification method for power equipment procurement parameters based on dynamic knowledge base and large model reasoning according to an embodiment of the present invention. Figure 4This is a structural block diagram of an intelligent verification device for power equipment procurement parameters according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] 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.

[0027] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0028] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0029] As an optional application scenario of this invention, considering the specific application environment architecture or specific hardware architecture upon which the intelligent verification method for power equipment procurement parameters depends, the specific application environment architecture or specific hardware architecture is described here. For example... Figure 1 As shown, the architecture system may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0030] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0031] According to an embodiment of the present invention, an embodiment of an intelligent verification method for power equipment procurement parameters is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0032] This embodiment provides an intelligent verification method for power equipment procurement parameters, which can be used in the aforementioned mobile terminals, such as mobile phones and tablets. Figure 2 This is a flowchart of an intelligent verification method for power equipment procurement parameters according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Obtain multiple standard specification unstructured documents and procurement documents for the target power equipment.

[0033] In one optional embodiment, the standard unstructured document is the technical specification document corresponding to the current national and industry standards of the power grid industry, and the carrier is mostly in PDF format. Furthermore, the standard unstructured document does not contain a unified structured data table, and contains a large number of text clauses, parameter tables, and technical terms. It has characteristics such as equipment parameter thresholds, forced logical constraints between parameters (such as the peak withstand current of the switchgear being 2.5 times the short-time withstand current), and mixed unit labels.

[0034] Furthermore, the procurement document refers to the tender document issued by the power company for the external bidding of power equipment such as transformers, switchgear, and GIS, and is in PDF / Word format. Moreover, this procurement document may include commercial terms, qualification requirements, sections on mixed technologies for multiple equipment, and equipment technical parameter tables, among other information.

[0035] In one optional embodiment, multiple unstructured standard and specification documents can be obtained in batches through the power standard resource library and standardization archive system, and obsolete and expired standard documents can be automatically filtered out, leaving only the latest valid specifications.

[0036] Furthermore, it can connect to the enterprise's internal procurement management system via API interface, and obtain tender documents (i.e. procurement documents) to be reviewed in batches, while simultaneously capturing the supporting metadata for each procurement document.

[0037] Furthermore, basic verification can be performed on the collected files. For example, damaged, blank, or encrypted documents that cannot be parsed can be removed.

[0038] Step S202: Construct a dynamic knowledge base based on multiple standard specifications of unstructured documents.

[0039] In one optional embodiment, the dynamic knowledge base is composed of two types of storage carriers: one is storing parameter thresholds, parameter threshold ranges, and logical mapping functions between parameters. The first is a relational database, and the second is to store semantic vectors of various standard parameters. A vector database.

[0040] In one optional embodiment, by performing a complete transformation of multiple standard unstructured documents into a benchmark data base that is standardized and machine-reasonable, a knowledge base, namely a dynamic knowledge base, can be constructed that has dynamic update capabilities and dual-database collaborative retrieval capabilities.

[0041] Step S203: Based on the procurement documents, the equipment category label of the target power equipment is obtained through a preset classification and fusion decision-making method.

[0042] In one optional embodiment, the preset classification fusion decision method represents a multi-source fusion device classification algorithm that takes into account both file metadata rules and text semantic features.

[0043] Furthermore, the equipment category label represents a unique identifier index for the power equipment corresponding to the procurement documents, such as switchgear, transformers, GIS, etc.

[0044] In one alternative embodiment, two types of features are separated from the procurement documents: metadata features. It can include structured business information such as file name, project procurement category, and business system annotations; content semantic features Extract unstructured text content such as the full text of the tender document and equipment description paragraphs.

[0045] Furthermore, based on preset industry keywords and procurement category rules, the prior probability of the rules is obtained by calculating metadata matching. Simultaneously, the trained text classification model can be used to parse the semantics of the entire text and obtain the posterior probability from machine learning. .

[0046] Furthermore, the equipment category label is determined using a classification fusion decision formula. The following relation (1) is shown:

[0047] In the formula: Represents the complete set of features of the input file. At that time, the procurement document belonged to the first... The overall probability of attribution to power equipment; This represents the classification fusion weight coefficient, used to balance the weight ratio of metadata rule classification and machine learning semantic classification, and its value is between 0 and 1. Indicates based on file metadata features The calculated prior probabilities of the rules are used to determine the file's ownership based solely on structured business rules such as filename and procurement system annotations. The probability of such devices; Represents semantic features based on document content The calculated posterior probability from machine learning is used to parse the full text of the tender document using a text semantic model to determine the file's ownership. The probability of such devices; Indicates the first Equipment category identifiers corresponding to various types of electrical equipment.

[0048] Step S204: Based on the procurement documents and equipment category labels, the core context fragments of the core parameter section in the procurement documents are obtained through domain-adaptive BERT model and named entity recognition method.

[0049] In one optional embodiment, the domain-adaptive BERT model represents a transfer learning paradigm that, based on a general pre-trained BERT, undergoes "continued pre-training" using unlabeled corpora from a specific domain to master the terminology, syntax, and semantic distribution of that domain, and then fine-tunes it for downstream tasks. In this embodiment, the domain-adaptive BERT model is a model fine-tuned using a masked language model (MLM) task on a standard corpus of the power industry.

[0050] Masked Language Model (MLM) represents a self-supervised pre-training task. Its core is to randomly mask some words in the input text, forcing the model to use bidirectional context to predict the masked content, thereby learning deep semantic representations.

[0051] Furthermore, Named Entity Recognition (NER) is a technique in natural language processing for identifying specific meaningful entities such as names of people, places, and organizations in text. Its essence is a sequence labeling task that transforms text into structured information.

[0052] Furthermore, the core parameters section refers to the independent section in the procurement documents that contains the technical parameter table and parameter constraint clauses of the target power equipment. Only this section contains the valid parameter information to be verified, while the other sections, such as business, qualifications, and other equipment, are all invalid noise content.

[0053] Furthermore, the core context fragment refers to the clean text paragraph extracted from the core parameters section after stripping away images, blank spaces, and irrelevant business text; it retains only descriptions related to all the technical parameters of the target device.

[0054] In one alternative embodiment, the procurement documents have problems such as a large amount of redundant business text, mixed chapters of multiple devices, and mixed use of title synonyms. General keyword matching cannot distinguish between different device paragraphs. Directly inputting the complete tender document into the large model will cause text illusion, context overflow and mismatch of cross-device parameters.

[0055] In this embodiment, target device parameter content is filtered by device category tags, and invalid text such as cross-device chapters and business qualifications is removed by fusing a domain-adaptive BERT model and entity recognition. This ultimately extracts a pure context, i.e., a core context fragment, containing only the target device parameters.

[0056] Step S205: Based on the dynamic knowledge base, equipment category tags, and core context fragments, a parameter verification method based on dynamic prompting engineering and large model logical reasoning is used to determine whether the parameters in the procurement documents are abnormal.

[0057] In one optional embodiment, the parameter verification method based on dynamic prompt engineering and large model logical reasoning represents a technical paradigm that guides the large model to perform step-by-step logical deduction and self-verification during the generation process by adaptively adjusting the prompt structure at runtime, thereby performing real-time consistency, integrity, and compliance checks on input parameters or output results.

[0058] In one alternative embodiment, existing traditional verification methods can only perform simple numerical comparisons and cannot identify parameter aliases, semantic offset issues across coding spaces, and lack the reasoning ability of multi-parameter linkage constraints. At the same time, directly feeding raw text without standard constraints to large models can easily produce parameter extraction illusions.

[0059] In this embodiment, a dynamic knowledge base, equipment category tags, and core context fragments are used as inputs. A large model is used to perform a complete reasoning process from free text to standardized parameters and then to compliance quantification. This replaces the manual comparison of standards and tender parameters item by item. At the same time, it covers two types of abnormal scenarios: single-point parameter exceeding limits and logical contradictions in parameter combinations. In the end, it can output standardized parameter anomaly judgment results.

[0060] Step S206: When it is determined that the parameters in the procurement document are abnormal, the abnormal result is linked and bound to the procurement document, and a structured early warning report of the procurement parameters of the target power equipment is generated.

[0061] In one optional embodiment, the structured early warning report for procurement parameters may include anomaly indices, descriptions of logical conflicts, and tracing information.

[0062] In one optional embodiment, abnormal indicators, conflict logic, and parameter deviation information are linked and bound to metadata such as project and department provided in the procurement documents. This enables the automatic generation and archiving of standardized early warning reports, which solves the pain points of traditional manual review of anomalies, lack of unified records, lack of traceability of problems, and unclear responsibility for rectification. At the same time, it can provide procurement personnel with intuitive and actionable basis for parameter rectification.

[0063] The intelligent verification method for power equipment procurement parameters provided in this embodiment replaces the manual item-by-item verification mode, solving multiple industry pain points such as low efficiency of manual review, false detection and missed detection, untimely standard updates, only single value comparison, and lack of traceability support for anomalies. It takes into account the verification processing efficiency, parameter recognition accuracy and the ability to identify complex parameter logical conflicts, and can adapt to complex procurement tender documents with multiple equipment mixed chapters.

[0064] In some optional implementations, step S202 above includes: Step S2021: Using optical character recognition technology, each standard specification unstructured document is parsed and processed to obtain multiple original standard specification structured documents.

[0065] In one alternative embodiment, optical character recognition (OCR) technology refers to computer vision technology that detects characters in paper documents using electronic devices and converts the text into editable text using image processing and pattern recognition techniques.

[0066] In one optional embodiment, OCR technology is used to overcome the problem that tables and text cannot be directly extracted from standard PDF images and scanned documents. This allows unstructured standard files without a fixed data format to be converted into original structured data tables that can be read, split, and calculated by a program.

[0067] For example, optical character recognition technology is used to parse each standard specification unstructured document page by page and accurately locate parameter tables and clause text areas within the document.

[0068] Furthermore, the system automatically extracts text, values, and unit labels from the table rows and columns, and converts the mixed text and image content into raw structured data in key-value pairs and two-dimensional table format. Finally, all parsing results are stored according to document number, forming multiple original standard structured documents.

[0069] Step S2022: Redundancy is removed from each original standard specification structured document to obtain multiple first standard specification structured documents.

[0070] In one alternative embodiment, by simplifying and denoising each original standard specification structured document and removing duplicate, incomplete, and invalid parameter entries, the computational power consumption of subsequent normalization, storage, and matching can be reduced, and redundant dirty data can be avoided from interfering with the accuracy of the knowledge base benchmark.

[0071] For example, iterate through all parameter entries in each original structured document, compare parameter names, device types, and threshold values, and then delete completely duplicate entries.

[0072] Furthermore, the system filters out incomplete and invalid parameter rows that contain missing values, blank units, or incorrect equipment category labeling. At the same time, it removes parameter records corresponding to obsolete standards and outdated clauses, retaining only the parameters of the current valid standards, and generates a set of first standard specification structured documents after cleaning and removing redundancy, i.e., multiple first standard specification structured documents.

[0073] Step S2023: Using a preset standardized mapping function, normalize each first standard specification structured document to obtain multiple second standard specification structured documents.

[0074] In an optional embodiment, the preset standardized mapping function is a normalized calculation model that unifies the units and magnitudes of power equipment parameters, as shown in the following relation (2):

[0075] In the formula: This represents the standardized parameter value; Represents the original numerical value; Indicates the unit conversion factor; This indicates the magnitude adjustment index.

[0076] Furthermore, after normalizing each first standard specification structured document using the above relation (2), multiple corresponding second standard specification structured documents are obtained.

[0077] Step S2024 involves reviewing and verifying multiple second standard specification structured documents to obtain multiple target standard specification structured documents.

[0078] In one optional embodiment, the deviation of the normalized data is corrected by a dual mechanism of automatic algorithm verification and manual review, which can eliminate problems such as OCR recognition errors, unit conversion errors, and incorrect recording of standard clauses, thereby ensuring the authority and accuracy of the knowledge base benchmark data.

[0079] For example, the algorithm is first automatically verified. Specifically, by comparing whether the parameter values ​​conform to common sense in the power industry and whether the logical constraints are consistent, parameter entries with abnormal values ​​or logical contradictions are marked.

[0080] Furthermore, the marked abnormal entries and all documents are pushed to the manual review interface, where professional power standards personnel verify and correct identification and conversion errors. Additionally, the timeliness of standard clauses, parameter thresholds, and parameter alias correspondences are manually confirmed, and missing parameter logical constraints are supplemented.

[0081] Furthermore, the documents that have passed the review are uniformly marked as qualified, and finally the corresponding target standard specification structured documents are output.

[0082] Step S2025: Decompose the parameters in each target standard specification structured document and build a dynamic knowledge base.

[0083] In one optional embodiment, by splitting the standardized and qualified parameter data and adopting a dual-database separate storage architecture, a dynamically updatable verification benchmark base is constructed, which can support two core verification capabilities: precise numerical rule retrieval and parameter alias semantic matching.

[0084] For example, the data within each target standard specification structured document is split to obtain a function containing parameter thresholds, parameter threshold ranges, and logical relationships between parameters. Structured rules, as well as the textual semantics corresponding to the names of standard parameters and aliases.

[0085] Furthermore, the parameter thresholds, threshold ranges, and logical constraint functions are stored in a relational database (such as MySQL). Simultaneously, a power-tuned BERT model is used to generate semantic vectors from the text encoding of all standard parameters. And store it in a vector database (such as Milvus).

[0086] Furthermore, an associated index of device categories, parameters, vectors, and logical rules is established to support incremental updates after the new standard is introduced, and ultimately form a complete dynamic knowledge base.

[0087] In some optional implementations, step S204 above includes: Step S2041: Based on the equipment category label, use the domain-adaptive BERT model and named entity recognition method to calculate the comprehensive matching degree value between the chapter title text and the preset core keyword text in the procurement document.

[0088] Specifically, step S2041 includes: Step a1: Using the domain-adaptive BERT model, the chapter title text and the preset core keyword text in the procurement document are encoded to obtain the chapter title enhanced semantic vector and the text enhanced semantic vector.

[0089] In an optional embodiment, a domain-adaptive BERT model (PowerBERT) can be used to encode the document chapter title text Tsec and the preset core keyword text Tkey.

[0090] Specifically, to capture the local features and global semantics of electrical engineering terminology, the outputs of different hidden layers of PowerBERT are extracted and fused at multiple scales to generate enhanced semantic vectors. The following relation (3) is shown:

[0091] In the formula: This represents the global classification vector of the last layer of the model; The hidden state matrix of the intermediate layers (such as layers 6 and 8) contains rich local lexical features. , Represents a trainable linear projection matrix; The average pooling calculation method introduces intermediate layer feature pooling concatenation. Because the bottom layer of BERT often contains more fine-grained lexical features, it is very effective for matching short texts such as parameter table titles, thus enhancing the feature expression capability. This represents the activation function.

[0092] Furthermore, the above can be used to obtain the enhanced semantic vector of chapter titles. and text-enhanced semantic vectors .

[0093] Step a2: Based on the enhanced semantic vector of the chapter title and the enhanced semantic vector of the text, calculate the multi-scale basic semantic similarity value between the chapter title text and the preset core keyword text.

[0094] In an alternative embodiment, the obtained chapter title is combined to enhance the semantic vector. and text-enhanced semantic vectors Calculate multi-scale basic semantic similarity values The following relation (4) is shown:

[0095] in, The value ranges from 0 to 1, with higher values ​​indicating closer semantic relevance.

[0096] Step a3: Using named entity recognition, extract the chapter title text to obtain device entity words.

[0097] In one optional embodiment, the chapter title text is input into the NER entity recognition model in the power field, and then matched with the built-in device alias dictionary of the model to extract all device-related entity words in the title.

[0098] Furthermore, the storage and retrieval results are used as device entity words. Furthermore, no-device terms are marked as empty sets. .

[0099] Step a4: Determine the entity penalty factor based on the device entity term and the device category label.

[0100] In an alternative embodiment, this can be achieved by using device entity words. With the target equipment category Perform alignment checks and calculate entity penalty factors. The following relation (5) is shown:

[0101] In the formula: Indicates the target device category The official name and alias dictionary for transformers includes "main transformer" and "station service transformer"; This indicates a strong penalty coefficient (value greater than 0.5). This indicates that the title does not recognize the device entity.

[0102] Step a5: Determine the comprehensive matching degree between the chapter title text and the preset core keyword text based on the multi-scale basic semantic similarity value and the entity penalty factor.

[0103] In an alternative embodiment, this can be achieved by combining multi-scale basic semantic similarity values. and entity penalty factor The overall matching degree value is calculated as shown in the following formula (6):

[0104] In the formula: This represents the overall matching score.

[0105] Step S2042: When the comprehensive matching degree value is greater than the preset matching threshold, the chapter corresponding to the chapter title text is determined to be the core parameter chapter of the target power equipment, and the core context fragment of the core parameter chapter is extracted.

[0106] In an optional embodiment, the overall matching score is... and preset matching threshold Perform a comparison, if If so, then the chapter corresponding to the chapter title text is determined to be the core parameter chapter of the current target power equipment.

[0107] Furthermore, for the core parameter sections that pass the judgment, page images, blank lines, and text irrelevant to business qualifications are removed, leaving only the complete device parameter description text, thereby generating the noise-reduced core context fragments.

[0108] Furthermore, if If so, then simply discard that chapter.

[0109] In some optional implementations, step S205 above includes: Step S2051: Based on the dynamic knowledge base, device category tags, and core context fragments, a structured parameter set is obtained through large language model processing.

[0110] In one alternative embodiment, a Large Language Model (LLM) refers to a deep learning model trained on a large amount of text data, which enables the model to generate natural language text or understand the meaning of language text.

[0111] The parameters in the original procurement documents are scattered in texts and tables without a unified format. The parameter names and aliases are diverse and contain irrelevant business content. Large models without standard rules are prone to extracting incorrect parameters and creating illusions.

[0112] In one optional embodiment, the "device type-extraction rule" mapping matrix is ​​retrieved using the device category tag as an index to obtain the exclusive extraction list and logical constraints for that type of device. Simultaneously, the standard parameter thresholds and logical mapping functions corresponding to that category are extracted from a dynamic knowledge base. It is then dynamically assembled with core context fragments into a prompt word template and input into a large language model.

[0113] Furthermore, large language models can output a set of structured parameters based on the input prompt word templates. .

[0114] Specifically, a complete prompt word template containing standard constraints, clean parameter text, and output specifications is fed into the large model for inference calculation. The standard rules within the prompt word template are used to constrain the output behavior of the large model, suppress the generation of irrelevant text and parameter illusions, and avoid extracting interference parameters from other devices in the tender document.

[0115] Furthermore, the large model reads tables and parameter description text from core context fragments; identifies equipment parameter names, parameter aliases, corresponding values, and units within the text; distinguishes target equipment parameters from residual noise parameters of other equipment; and filters out textual content with no parameter value, such as qualifications, business information, and project schedule.

[0116] Furthermore, the large model outputs all extracted target device parameters according to the format specified by the prompt words, and organizes them into a structured parameter set. .

[0117] Step S2052: Calculate the comprehensive anomaly index value based on the structured parameter set.

[0118] Specifically, step S2052 includes: Step b1: Obtain the original semantic encoding vector for each structured parameter in the structured parameter set.

[0119] Step b2 involves mapping each original semantic encoding vector to each parameter semantic vector in the dynamic knowledge base using the principle of maximizing mapping confidence, thereby obtaining a mapping parameter set.

[0120] Step b3: Obtain the subset of parameters in the mapping parameter set that contain implicit physical or standard constraints.

[0121] Step b4: Calculate the logical conflict degree of parameter combinations in the parameter subset of the dynamic knowledge base using the logical mapping function in the dynamic knowledge base.

[0122] Step b5: Calculate the comprehensive anomaly index value based on the logical conflict degree value.

[0123] In an alternative embodiment, each parameter in the structured parameter set is traversed. The parameter names are text-encoded using the built-in encoder of the large language model to generate the original semantic encoded vector. .

[0124] Furthermore, the original semantic encoding vector With the semantic vector of each parameter in the dynamic knowledge base Perform mapping. Further, the confidence level of the mapping alignment. The calculation formula is shown in the following relation (7):

[0125] In the formula: This represents a learnable cross-spatial alignment projection matrix used to eliminate semantic offsets between different encoders; Indicates the vector dimension; Represents the entity-aware mask term, if the standard parameters Not belonging to the current equipment category ,but This indicates that the exponent term is set to zero, directly shielding cross-device parameter interference; otherwise... ; This indicates the total number of candidate parameters.

[0126] Furthermore, the principle of maximum confidence level can be applied. This completes the precise mapping and matching of extracted parameters to standard parameters, and obtains the matched mapping parameter set.

[0127] Furthermore, there exists a subset of parameters in the mapped parameter set that contain implicit physical / standard constraints. (For example, transformer capacity-impedance combinations) can utilize the logic mapping functions in the dynamic knowledge base. Calculate the logical conflict degree of the actual parameter combinations within the parameter subset. The following relation (8) is shown:

[0128] In the formula: This indicates the actual logical relationship value followed by the extracted parameters; This represents the nominal logical relation value in the standard knowledge base.

[0129] Furthermore, the single-parameter threshold deviation and logical conflict degree can be combined. Calculate the comprehensive anomaly index The following relation (9) is shown:

[0130] In the formula: Indicates parameter weights; This indicates a single-point over-limit indicator function (1 for over-limit, 0 otherwise); Indicates the allowable deviation threshold; Indicates the logical conflict penalty factor; This represents the first item stored in the dynamic knowledge base under the corresponding device type. The nominal benchmark value of the parameter in the national / industry standard.

[0131] Step S2053: When the comprehensive anomaly index value is greater than the preset judgment threshold, the parameters of the procurement document are determined to be abnormal.

[0132] In an optional embodiment, the comprehensive anomaly index value is... and preset judgment threshold Perform a comparison, if If the parameter combination is found to be abnormal, the parameters in the procurement document will be deemed abnormal, and an error record and logical conflict deduction path will be output.

[0133] Step S2054: When the comprehensive anomaly index value is less than or equal to the preset judgment threshold, the parameters of the procurement document are determined to be normal.

[0134] In an alternative embodiment, if If the parameter combination is normal, then the parameters in the procurement document are considered normal.

[0135] In one instance, such as Figure 3 As shown, an intelligent verification method for power equipment procurement parameters based on dynamic knowledge base and large model reasoning is provided, including the following steps: Step 1: Structured extraction and knowledge base solidification of standard and specification documents (building a comparison benchmark).

[0136] This step is fundamental to intelligent verification. Its core objective is to transform unstructured standard specification documents into structured knowledge base data, providing authoritative and accurate standard references for subsequent parameter comparisons. It mainly includes the following key steps in sequence: OCR technology is used to parse standard specification PDF documents, extract tabular data and convert it into a structured format; a data cleaning engine is built to remove redundancy and unify units in the extracted data, providing an authoritative benchmark data source for subsequent parameter comparison and logical reasoning. In the unit unification process, the parameter values ​​are normalized using the following standardized mapping function, as shown in the above relation (2).

[0137] Furthermore, after manual review and algorithm verification, the data is split into two storage methods: structured mapping rules (including parameter threshold ranges and logical relationship functions between parameters). Stored in a relational database; semantic vector of standard parameters. Stored in a vector database; these two types of knowledge bases serve as the underlying data support for prompt word injection and logical conflict calculation in step four.

[0138] Step 2: Batch acquisition and multi-level classification of procurement documents (determine verification dimensions).

[0139] This step identifies the equipment attributes in the procurement documents, providing a navigation index for subsequent loading of differentiated parsing and inference rules. It also involves batch retrieving file metadata and documents through an interface connecting to the procurement management system. The main key steps include: Equipment category labels are determined using a classification fusion decision formula. As shown in the above relation (1).

[0140] Furthermore, the output of this step is the target device category. This output is directly used as an index key to dynamically match the corresponding extraction rules and prompt word templates from the mapping matrix in step four.

[0141] Step 3: Semantic parsing and core context extraction of procurement documents (filtering inference noise).

[0142] This step involves in-depth analysis of the procurement documents, accurately extracting core content containing equipment technical parameters from complex document structures. Because procurement documents contain significant commercial and qualification noise, directly inputting them into a large model can lead to "illusions" and context overflow. Furthermore, chapter titles in power equipment procurement documents often suffer from issues such as "long modifiers," "synonym substitution," and "cross-equipment chapter confusion" (e.g., a single document containing both "transformer technical parameters" and "switchgear technical parameters"). Conventional single-granularity semantic matching is highly prone to misjudgment or omission. This step is based on the equipment categories determined in step two. The core context containing the technical parameters of this type of equipment is extracted in a targeted manner; a multi-scale semantic matching algorithm based on domain adaptation and entity perception is used to accurately extract the core context, specifically including: (1) Domain-adaptive multi-scale feature extraction.

[0143] A domain-adaptive BERT model (PowerBERT), fine-tuned using a masked language model (MLM) task on a standard corpus for the power industry, is employed to encode document chapter title text (Tsec) and preset core keyword text (Tkey). To capture the local features and global semantics of power industry terminology, the outputs of different hidden layers of PowerBERT are extracted and fused at multiple scales to generate enhanced semantic vectors. As shown in relation (3) above. The chapter titles and keywords are encoded respectively to obtain... and .

[0144] (2) Basic semantic similarity calculation: Calculate the multi-scale basic semantic similarity between chapter titles and keywords. As shown in the above relation (4).

[0145] (3) Entity perception constraint and punishment mechanism.

[0146] To avoid mismatches across equipment sections (e.g., when searching for "transformer" parameters, mistakenly matching the "switchgear technical parameter table" in the same file), an entity-aware penalty term is introduced. Named Entity Recognition (NER) is used to extract equipment entity words from section titles. And compare it with the target device category output in step two. Perform alignment checks and calculate entity penalty factors. As shown in the above relation (5).

[0147] (4) Comprehensive matching degree determination: The final comprehensive matching degree is calculated by combining multi-scale semantic similarity and entity penalty factor. As shown in the above relation (6).

[0148] Furthermore, when When a preset matching threshold is reached, the chapter is determined to be a core parameter chapter for the current device. The system automatically removes irrelevant fields and images, extracting the core context fragments. The output of this step is the denoised "core context fragment," which serves as the input corpus for direct processing by the large model in step four. This significantly reduces inference noise and avoids logical misjudgments caused by cross-device parameter confusion.

[0149] Step 4: Parameter verification based on dynamic prompts and large model logical reasoning.

[0150] This step deeply integrates the baseline data from Step 1, the category index from Step 2, and the context fragments from Step 3. Through a large model, it bridges the gap between unstructured text and structured logical conflict determination. Specifically, this includes: (1) Dynamic loading of rules and assembly of prompt words: based on the device category output in step two The system retrieves the "Device Type - Extraction Rule" mapping matrix to obtain the exclusive extraction list and logical constraints for this type of device. Simultaneously, it extracts the standard parameter thresholds and logical mapping functions corresponding to this category from the relational database in step one. The core context fragments output from step three are dynamically assembled into the prompt word template and input into the large language model. (2) Alignment between cross-spatial projection and entity perception: The large model outputs a set of structured parameters based on the prompt words. Because the vector space generated by the large model has a different distribution than the encoding space of the standard knowledge base in step one, and procurement documents often have cross-device chapter parameter confusion issues, this example uses an alignment confidence algorithm based on cross-spatial projection and entity perception to accurately map the extracted parameters to the standard parameters.

[0151] Specifically, this refers to the original semantic encoding vector of the i-th parameter name (such as "impedance voltage") extracted from the procurement documents by the large language model. Compare this vector with the standard vector in the vector library from step one. Perform mapping, confidence level of mapping alignment The calculation formula is shown in the above relation (7).

[0152] Furthermore, based on the principle of maximum confidence... This completes the accurate mapping and matching of extracted parameters to standard parameters.

[0153] (3) Quantitative deduction of logical conflicts: for the parameter subset of the existence of implicit physical / standard constraints after mapping. (e.g., transformer capacity-impedance combination), extract the preset standard logic function from step one. Calculate the logical conflict degree of the actual parameter combination. As shown in the above relation (8).

[0154] (4) Comprehensive anomaly determination: Calculate the comprehensive anomaly index by combining the single parameter threshold deviation and the degree of logical conflict. As shown in the above relation (9).

[0155] Furthermore, when When the (preset judgment threshold) is reached, the parameter combination is judged to be abnormal, and the abnormal record and logical conflict deduction path are output.

[0156] Step 5: Source tracing of abnormal results and generation of early warning reports.

[0157] For the abnormal results identified in step four, extract the file metadata (project number, responsible department, etc.) obtained in step two, perform data association and binding, and generate a structured early warning report containing anomaly index, logical conflict description and source tracing information to complete the verification closed loop.

[0158] The intelligent verification method for power equipment procurement parameters based on dynamic knowledge base and large model reasoning provided in this example has the following beneficial effects: 1. Dual improvement in review efficiency and accuracy: By building a high-quality standard knowledge base through "OCR + manual review" and combining the semantic understanding capabilities of large models, the process has been transformed from manual visual review to machine intelligent review, increasing the processing speed by dozens of times and effectively avoiding missed detections caused by human fatigue.

[0159] 2. High versatility and adaptability: Adopting dynamic prompting engineering technology, the system can automatically switch extraction rules according to the device type, without rewriting code logic for each device. It has strong scalability and can quickly adapt to the review needs of newly added device types.

[0160] 3. Possesses deep logic verification capabilities: Breaking through the limitations of traditional technologies that can only perform numerical comparisons (greater than / less than), it utilizes the reasoning capabilities of large models to identify complex errors such as "parameter combinations do not conform to standard logic mapping" (e.g., capacity and impedance mismatch), thereby enhancing the depth of data analysis and comparison.

[0161] 4. High degree of data standardization: The measurement units are standardized during the cleaning stage, which solves the comparison problem caused by inconsistent unit writing in the procurement documents and ensures the consistency of the verification benchmark.

[0162] Furthermore, taking the parameter verification of 10kV switchgear procurement documents (including scenarios with mixed chapters for multiple devices) as an example, the above-mentioned intelligent verification method for power equipment procurement parameters based on dynamic knowledge base and large model reasoning is described in detail.

[0163] Specifically, this embodiment takes the parameter verification of a power company's procurement document for "10kV switchgear and supporting transformers" as an example. This procurement document is a typical mixed document containing multiple devices. Furthermore, the intelligent verification process is as follows: 1. Standard knowledge base construction phase.

[0164] The system first reads PDF standard documents such as "3.6kV~40.5kV AC Metal-Enclosed Switchgear and Controlgear" (GB / T 3906-2020). The OCR recognition engine extracts the switchgear parameter table, and the rated current column in the original data contains records with the unit "0.0315kA".

[0165] Furthermore, the data cleaning engine initiates unit standardization processing, employing the standardized mapping function shown in the above relation (2). For "0.0315kA", its unit is identified as kA, and the target standard unit is... Then the conversion factor Magnitude Adjustment Index The standardized parameter values ​​were calculated. .

[0166] At the same time, the system structures the logical constraint rules specified in the standard, such as extracting the switchgear's "rated peak withstand current ( The rated short-time withstand current should be ( The mandatory requirement of "2.5 times" is transformed into a standard logic mapping function. After manual review and confirmation, the structured mapping rules and logical functions are stored in a MySQL relational database, and the semantic vectors of parameter names (including aliases) are stored in a Milvus vector database.

[0167] 2. Rule dynamic loading phase.

[0168] The system connects to the procurement platform via API to batch retrieve documents named "2023 10kV Switchgear and Supporting Transformer Procurement Tender Document.pdf". A multi-level classification model is constructed to extract the metadata features of the documents. With content semantic features Because the filename contains "switch cabinet", the rule-based prior probability... Based on text semantic analysis, posterior probability Set weighting factors. Calculate the fusion probability: Due to the high probability of merging, the system tagged the file with the "switch cabinet" equipment category. This serves as the index key for subsequent differential verification.

[0169] 3. Semantic parsing and core context extraction stage.

[0170] Analysis of the PDF file revealed that it contains "Chapter 1: Technical Parameter Table for 10kV Switchgear" and "Chapter 2: Technical Parameter Table for Distribution Transformers".

[0171] The system employs a domain-adaptive PowerBERT model to extract multi-scale features from chapter titles, resulting in a fused semantic vector. and with preset core keyword vectors ("Switchgear technical parameters") Calculate basic semantic similarity Assuming the calculation yields the results of the two chapters... Both are 0.85.

[0172] To accurately filter out interference from the transformer chapter, entity-aware constraints are introduced. NER identifies the entity in the first chapter title. The target category is "switch cabinet". Punishment factor Overall matching degree The entity identified in the Chapter 2 title is "Transformer," which does not belong to... A stronger penalty δ=1.0 is imposed. , .

[0173] Set matching threshold Chapter 1 only The transformer section was successfully extracted, and the core context fragment of the switchgear was extracted.

[0174] 4. Large-scale model analysis and comparison stage.

[0175] The system identifies the device category. For "switch cabinet", dynamically load the switch cabinet-specific Prompt template and extract the content from step one. The logical rules and core context fragments from step three are injected into the large model. The large model outputs a structured parameter set, including: "Dynamic stability current 63kA" "Thermal stability current 25kA" "No-load loss 1500W" (The large model mistakenly extracted transformer parameters due to residual noise interference).

[0176] (1) Alignment between cross-spatial projection and entity perception: The system initiates the alignment algorithm. For incorrectly extracted parameters... ("No-load current"), calculate its alignment confidence level. .when When matching switchgear parameters in the standard library, mask item ;when When checking the transformer parameter "no-load loss" in the standard library, a mask penalty is triggered because the current verification category is switchgear. Therefore, the exponent term is zero. The confidence level of the parameters aligned to the switchgear was extremely low, and they were automatically discarded by the system as irrelevant parameters. For After projection matrix After mapping, the confidence level is aligned to the standard parameter "rated peak withstand current". ; Alignment to "rated short-time withstand current" confidence level Successfully achieved semantic alignment of aliases.

[0177] (2) Quantification of logical conflict deduction.

[0178] The system identifies the aligned parameter subset. A multiple constraint exists. The actual value is calculated by calling a logical relation function. To obtain the nominal value, call a standard logic function: Calculate the degree of logical conflict. The conflict rate is 0.8%, indicating that the logic of the peak value and the short-time withstand current is consistent and there is no conflict.

[0179] (3) Comprehensive anomaly determination.

[0180] The system further compared the single-point threshold. According to GB / T 3906 standard, the rated short-time withstand current standard value of this specification of switchgear should be 31.5kA, while the value extracted from the document is 25kA, indicating that a single point exceeds the limit.

[0181] Set parameter weights tolerance Indicator function Logical conflict penalty factor Calculate the comprehensive anomaly index: The anomaly index exceeded the parameter weight, and the system ultimately determined that the parameters of the procurement document were abnormal.

[0182] 5. Result Output Stage (Step Five): Based on the above judgment results, the system automatically associates the file metadata (Project No.: 2023-SW-01, Responsible Department: Project Management Department), generates and pushes a structured early warning report. The report details not only point out the single-point over-limit problem of "rated short-time withstand current 25kA is lower than the standard value of 31.5kA", but also explain through logical deduction that "the relationship between the peak value and the short-time withstand current of 2.52 times is consistent, and the root cause of the problem is that the current level selection is too low, resulting in the overall parameters not meeting the standard." Based on this, the procurement personnel directly requested the supplier to correct the current level configuration, effectively avoiding the short-circuit withstand risk after the equipment was connected to the network.

[0183] This embodiment also provides an intelligent verification device for power equipment procurement parameters. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0184] This embodiment provides an intelligent verification device for power equipment procurement parameters, such as... Figure 4 As shown, the device includes: Module 401 is used to acquire multiple standard specification unstructured documents and procurement documents for the target power equipment.

[0185] Module 402 is used to build a dynamic knowledge base based on the multiple standard specifications of unstructured documents.

[0186] The first processing module 403 is used to obtain the equipment category label of the target power equipment by processing the procurement documents through a preset classification fusion decision method.

[0187] The second processing module 404 is used to obtain the core context fragment of the core parameter section in the procurement document by processing the procurement document and the equipment category label through the domain adaptive BERT model and the named entity recognition method.

[0188] The determination module 405 is used to determine whether the parameters of the procurement document are abnormal based on the dynamic knowledge base, the equipment category tags, and the core context fragment, using a parameter verification method based on dynamic prompting engineering and large model logical reasoning.

[0189] The generation module 406 is used to, when it is determined that the parameters of the procurement document are abnormal, associate and bind the abnormal result with the procurement document, and generate a structured early warning report of the procurement parameters of the target power equipment.

[0190] The intelligent verification device for power equipment procurement parameters provided in this embodiment of the invention can execute the intelligent verification method for power equipment procurement parameters provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules are the same as in the corresponding embodiments described above, and will not be repeated here.

[0191] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0192] The following is a detailed reference. Figure 5The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from memory 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device. The processor 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0193] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0194] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a memory 508, or installed from a ROM 502. When the computer program is executed by the processor 501, it performs the functions defined in the intelligent verification method for power equipment procurement parameters according to embodiments of the present invention.

[0195] Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0196] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the intelligent verification method for power equipment procurement parameters shown in the above embodiments is implemented.

[0197] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0198] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for intelligent verification of procurement parameters for power equipment, characterized in that, The method includes: Obtain multiple standard specification unstructured documents and procurement documents for the target power equipment; A dynamic knowledge base is constructed based on the aforementioned multiple standard specifications and unstructured documents; Based on the procurement documents, the equipment category label of the target power equipment is obtained after processing by a preset classification and fusion decision method; Based on the procurement documents and the equipment category labels, the core context fragments of the core parameter section in the procurement documents are obtained through processing with the domain-adaptive BERT model and the named entity recognition method. Based on the dynamic knowledge base, the equipment category tags, and the core context fragments, a parameter verification method based on dynamic prompting engineering and large model logical reasoning is used to determine whether the parameters of the procurement document are abnormal. When the parameters of the procurement document are determined to be abnormal, the abnormal result is linked and bound to the procurement document, and a structured early warning report of the procurement parameters of the target power equipment is generated.

2. The method according to claim 1, characterized in that, Based on the aforementioned multiple standard specification unstructured documents, a dynamic knowledge base is constructed, including: Using optical character recognition technology, each standard specification unstructured document is parsed and processed to obtain multiple original standard specification structured documents; Redundancy is removed from each original standard specification structured document to obtain multiple first standard specification structured documents; By using a pre-defined standardized mapping function, each first standard specification structured document is normalized to obtain multiple second standard specification structured documents; The multiple second standard specification structured documents are reviewed and verified to obtain multiple target standard specification structured documents; The parameters in each target standard specification structured document are broken down, and the dynamic knowledge base is constructed.

3. The method according to claim 1, characterized in that, Based on the procurement documents and the equipment category labels, and processed using a domain-adaptive BERT model and named entity recognition method, the core context fragments of the core parameter section in the procurement documents are obtained, including: Based on the equipment category label, using the domain-adaptive BERT model and the named entity recognition method, the comprehensive matching degree value of the chapter title text and the preset core keyword text in the procurement document is calculated; When the overall matching degree value is greater than the preset matching threshold, the chapter corresponding to the chapter title text is determined to be the core parameter chapter of the target power equipment, and the core context fragment of the core parameter chapter is extracted.

4. The method according to claim 3, characterized in that, Based on the equipment category tags, using the domain-adaptive BERT model and the named entity recognition method, the comprehensive matching degree value between the chapter title text and the preset core keyword text in the procurement document is calculated, including: Using the domain-adaptive BERT model, the chapter title text and the preset core keyword text in the procurement document are encoded to obtain the chapter title enhanced semantic vector and the text enhanced semantic vector. Based on the enhanced semantic vector of the chapter title and the enhanced semantic vector of the text, calculate the multi-scale basic semantic similarity value between the chapter title text and the preset core keyword text; Using the named entity recognition method, the chapter title text is extracted to obtain device entity words; The entity penalty factor is determined based on the device entity words and the device category tags; Based on the multi-scale basic semantic similarity value and the entity penalty factor, the comprehensive matching degree value between the chapter title text and the preset core keyword text is determined.

5. The method according to claim 1, characterized in that, Based on the dynamic knowledge base, the equipment category tags, and the core context fragments, a parameter verification method based on dynamic prompting engineering and large-scale model logical reasoning is used to determine whether the parameters of the procurement documents are abnormal, including: Based on the dynamic knowledge base, the device category tags, and the core context fragments, a structured parameter set is obtained through large language model processing. Based on the structured parameter set, the comprehensive anomaly index value is calculated; When the comprehensive anomaly index value is greater than the preset judgment threshold, the parameters of the procurement document are determined to be abnormal. When the comprehensive anomaly index value is less than or equal to the preset judgment threshold, the parameters of the procurement document are determined to be normal.

6. The method according to claim 5, characterized in that, Based on the aforementioned structured parameter set, the comprehensive anomaly index value is calculated, including: Obtain the original semantic encoding vector for each structured parameter in the set of structured parameters; Based on the principle of maximum mapping confidence, each original semantic encoding vector is mapped and matched with each parameter semantic vector in the dynamic knowledge base to obtain a mapping parameter set. Obtain a subset of parameters in the mapping parameter set that contain implicit physical or standard constraints; Using the logical mapping function in the dynamic knowledge base, calculate the logical conflict degree value of the parameter combination in the parameter subset of the dynamic knowledge base; The comprehensive anomaly index value is calculated based on the logical conflict degree value.

7. An intelligent verification device for power equipment procurement parameters, characterized in that, The device includes: The acquisition module is used to acquire multiple standard specification unstructured documents and procurement documents for the target power equipment; The building module is used to construct a dynamic knowledge base based on the multiple standard specifications of unstructured documents; The first processing module is used to obtain the equipment category label of the target power equipment based on the procurement documents and through a preset classification fusion decision method. The second processing module is used to obtain the core context fragment of the core parameter section in the procurement document by processing the procurement document and the equipment category label through the domain adaptive BERT model and the named entity recognition method. The determination module is used to determine whether the parameters of the procurement document are abnormal based on the dynamic knowledge base, the equipment category tags, and the core context fragments, using a parameter verification method based on dynamic prompting engineering and large model logical reasoning. The generation module is used to determine that the parameters of the procurement document are abnormal, associate and bind the abnormal result with the procurement document, and generate a structured early warning report of the procurement parameters of the target power equipment.

8. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the intelligent verification method for power equipment procurement parameters as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the intelligent verification method for power equipment procurement parameters as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes computer instructions, which are used to cause a computer to execute the intelligent verification method for power equipment procurement parameters as described in any one of claims 1 to 6.

Images