Intelligent knowledge retrieval method and system for basic part manual

Abstract

The invention discloses an intelligent knowledge retrieval method and system oriented to a basic part manual, and belongs to the technical field of industrial knowledge retrieval. The method comprises the following steps: constructing a professional knowledge meta-model according to obtained professional knowledge of a basic file; wherein the professional knowledge meta-model comprises basic part knowledge classification, cross-standard equal-order mapping and a knowledge dependence graph; obtaining a basic part manual, and determining manual coordinate information of the basic part manual; establishing a mapping relation between chapter content of the basic part manual and basic part knowledge classification, and marking difference reasons of manual content of the same basic part knowledge classification; analyzing the chapter content to obtain analyzed content, and marking a mapping relation and manual coordinate information for the analyzed content to obtain knowledge atoms; analyzing to obtain user query information and executing retrieval to obtain an initial knowledge set; and generating an initial answer according to the initial knowledge set, and performing iterative optimization on the initial answer to obtain a target retrieval result.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of industrial knowledge retrieval, and particularly relates to an intelligent knowledge retrieval method and system for a basic component manual. BACKGROUND

[0002] In industrial production and engineering design, the selection and calculation of basic components such as bolts, bearings, and resistors highly depend on various professional manuals (such as national standard manuals, international standard manuals, and enterprise internal specifications). These manuals contain key knowledge such as product parameters, material properties, calculation methods, and testing processes. However, the existing technology has the following outstanding problems in manual knowledge retrieval and management: 1. Multiple manual management confusion: Enterprises often need to refer to multiple versions and systems of manuals (such as GB, ISO, and DIN). However, the existing technology treats manuals as independent documents, leading to scattered and repeated knowledge for the same basic component (such as "8.8 grade bolt strength" being expressed in multiple manuals), version conflicts (differences in parameter expression between old and new manuals), and standard confusion (cross-system terms cannot be associated). Engineers need to manually filter information, which is prone to selection errors.

[0003] 2. Serious knowledge fragmentation: Traditional retrieval technology uses fixed-length text segmentation, resulting in complete knowledge being split (such as bolt strength formulas and parameter tables belonging to different text blocks). When querying, a large amount of irrelevant content needs to be traversed, resulting in low retrieval efficiency and difficulty in obtaining complete practical knowledge (such as the source of parameters required for calculation and applicable conditions).

[0004] 3. Blind retrieval strategy: The existing intelligent retrieval technology lacks a professional framework for basic component manuals, cannot select an adaptive path according to the query type (such as selection or calculation), and relies mainly on full-text vector retrieval, which is prone to return irrelevant results. It also cannot actively complete missing information (such as missing working condition coefficients in calculations), requiring users to ask multiple times.

[0005] 4. Answer quality is difficult to guarantee: The existing technology relies mainly on single-agent retrieval, which cannot actively check answer defects (such as ignoring the applicable environmental constraints of basic components). It also lacks an optimization mechanism for basic component scenarios, making it difficult to meet engineering requirements in terms of answer completeness and accuracy (such as not considering the impact of impact load on bearing selection).

[0006] In summary, there is an urgent need for an intelligent retrieval solution that can unify knowledge frameworks, optimize retrieval strategies, and guarantee answer quality to meet the practical needs of engineering scenarios such as basic component selection and calculation. SUMMARY In view of this, the present invention provides an intelligent knowledge retrieval method and system for basic component manuals, which solves the problems that existing technologies cannot address in the storage of basic component manuals, such as large quantity, scattered distribution, chaotic management, slow retrieval speed, and offset and missing retrieval results.

[0007] To achieve the above objectives, this invention provides an intelligent knowledge retrieval method for basic component manuals, comprising the following steps: A professional knowledge meta-model is constructed based on the professional knowledge obtained from the basic documents; the professional knowledge meta-model includes: basic component knowledge classification, cross-standard order mapping, and knowledge dependency graph; Obtain the basic component manuals and determine the manual coordinate information; the manual coordinate information includes: time dimension, organization dimension, field dimension and authority information; Establish a mapping relationship between the chapter content of the basic component manual and the basic component knowledge classification, and indicate the reasons for the differences in the content of manuals within the same basic component knowledge classification; The chapter content is analyzed to obtain the analyzed content, and the mapped relationships and manual coordinate information are marked on the analyzed content to obtain knowledge atoms; The user query information is parsed and retrieved into a professional knowledge metamodel, and a retrieval is performed to obtain an initial knowledge set; Based on the game theory iterative optimization method, an initial answer is generated from the initial knowledge set and iteratively optimized to obtain the target retrieval result.

[0008] As an embodiment of the present invention, constructing a professional knowledge meta-model based on the acquired basic document professional knowledge includes: Determine the primary classifications, subcategories, core concepts, standard terminology, and parameter association rules in the acquired basic document professional knowledge; Based on natural language processing technology, we analyze basic component manuals and industry standard documents to extract high-frequency knowledge patterns and cross-standard equivalent terms to obtain cross-standard hierarchical mappings. Construct a knowledge dependency graph based on the dependencies of professional knowledge in the basic files and the acquisition methods of each parameter; When importing a new basic component manual, the coverage of the professional knowledge metamodel is checked, and the professional knowledge metamodel is updated based on the check results.

[0009] As an embodiment of the present invention, a knowledge dependency graph is constructed based on the dependencies of professional knowledge in the basic documents and the acquisition methods of each parameter, including: The calculation formulas, parameters, and parameter descriptions are extracted from the calculation method section of the basic component manual. Determine the dependencies between parameters in each formula; where dependencies include direct dependencies and indirect dependencies. A directed knowledge dependency graph is constructed with parameters as nodes and dependencies as edges, and the acquisition method of each parameter and the chapter from the basic component manual are labeled.

[0010] As an embodiment of the present invention, a mapping relationship is established between the chapter content of the basic component manual and the basic component knowledge classification, and the reasons for the differences in the manual content of the same basic component knowledge classification are marked, including: Determine the semantic relationship between the chapter titles of the basic component manual and the category names of the basic component knowledge classifications; Determine the degree of matching between the key information in the chapter content of the basic component manual and the characteristics of the basic component knowledge classification; The mapping relationship between the chapter content of the basic component manual and the basic component knowledge classification is obtained based on semantic relevance and matching degree. Note the reasons for the differences in the content of manuals that classify the same basic component knowledge.

[0011] As an embodiment of the present invention, parsing the chapter content yields parsed content, and annotating the parsed content with mapping relationships and manual coordinate information yields knowledge atoms, including: The text, images, tables, and formulas in the basic component manual are classified and parsed. Specifically, for table and image data, a dedicated large model with the ability to parse the format is called to complete the parsing of the manual content and obtain the parsed content. The parsed content is segmented according to semantic integrity; The parsed content is labeled with mapping relationships and manual coordinate information to obtain knowledge atoms.

[0012] As an embodiment of the present invention, user query information is parsed and obtained into a professional knowledge meta-model and a retrieval is performed to obtain an initial knowledge set, including: The search results are obtained by parsing the user's query information; the search results include: query intent and query scenario. Based on the query intent, key technology concepts are identified and a retrieval mapping relationship is established with basic component knowledge classification. Based on preset manual selection rules, the appropriate manual is determined according to the query scenario and manual coordinate information; the preset manual selection rules include: priority rules and knowledge conflict and supplementation rules. Based on the preset multi-strategy retrieval method, an initial knowledge set is obtained according to the adaptation manual and retrieval mapping relationship.

[0013] As an embodiment of the present invention, if a knowledge conflict is detected in the basic component manual, an appropriate manual is selected according to the knowledge conflict resolution rules. If the detected basic component manual has incomplete information, it will be supplemented from other basic component manuals under the same basic component knowledge category; The rules for resolving knowledge conflicts include: Version conflict: When different versions of the basic component manual under the same organization have different descriptions of the same parameter, the current valid version shall prevail, and the parameter information of the historical versions shall be marked for traceability. Standard system conflict: When different standard systems have different values ​​for the same concept parameter, the parameters of each standard are presented side by side, prompting the user to select the appropriate standard for the project and explaining the reasons for the differences.

[0014] Conflict between enterprise and general standards: When the enterprise's own basic component manual requirements are stricter than the general standards, both the enterprise's requirements and the general standard benchmark values ​​should be presented, and the scope of application of the enterprise's requirements should be indicated.

[0015] Conflicts and contradictions: When different basic component manuals present contradictory statements on the same issue and cannot be resolved by version conflict, standard system conflict, or conflict between enterprise and general standards, mark the conflict and provide the basis for each basic component manual, and offer suggestions in conjunction with consulting experts in the field based on the actual engineering situation.

[0016] As an embodiment of the present invention, based on a game-theoretic iterative optimization method, an initial answer is generated according to an initial knowledge set and iteratively optimized to obtain the target retrieval result, including: Construct adversarial dual agents; wherein the adversarial dual agents include: a retrieval agent and a verification agent; Based on the game-theoretic iterative optimization method, the retrieval agent and the verification agent generate an initial answer based on the initial knowledge set and iteratively optimize it to obtain the target retrieval result. Specifically, the retrieval agent generates the initial answer according to the retrieval strategy, and the verification agent raises questions about the initial answer according to the questioning strategy and provides manual evidence. The iteration terminates when there are no effective questions for several consecutive rounds or when a reasonable number of iterations is reached.

[0017] As an embodiment of the present invention, based on a game-theoretic iterative optimization method, an initial answer is generated from an initial knowledge set and iteratively optimized to obtain the target retrieval result, and the method further includes: The retrieval strategy of the retrieval agent and the questioning strategy of the verification agent are optimized based on the comprehensive reward mechanism to obtain the updated retrieval strategy and the updated questioning strategy. The retrieval agent and the verification agent are updated according to the updated retrieval strategy and the updated challenge strategy, respectively; The comprehensive reward mechanism includes: Search efficiency: Combining the number of search steps and the hit rate of effective knowledge units, the fewer the steps and the higher the hit rate, the higher the reward weight; Reasoning completeness: Evaluate the completeness of the reasoning logic in the answer and the degree to which parameter dependencies are satisfied; Answer accuracy: Assess the consistency between the answer and the content of the adaptation manual, as well as the traceability of the information source; Weighting adaptation scenarios: Adjust weights for different query types of basic components. Parameter queries focus on retrieval efficiency, selection calculations focus on reasoning completeness, and material recommendations focus on answer accuracy.

[0018] In addition, the present invention also provides an intelligent knowledge retrieval system for basic component manuals, including: The metamodel building module is used to build a professional knowledge metamodel based on the professional knowledge obtained from the basic documents. The manual management module is used to obtain basic component manuals and determine the manual coordinate information of the basic component manuals. The manual coordinate information includes: time dimension, organization dimension, domain dimension and authority information; establish the mapping relationship between the chapter content of the basic component manuals and the basic component knowledge classification, and mark the reasons for the differences in the manual content of the same basic component knowledge classification. The knowledge atom processing module is used to parse chapter content to obtain parsed content, and to annotate the parsed content with mapping relationships and manual coordinate information to obtain knowledge atoms. The query scheduling module is used to parse and obtain user query information into the professional knowledge metamodel and perform retrieval to obtain the initial knowledge set; The dual-agent optimization module is used to generate an initial answer based on the initial knowledge set and iteratively optimize it to obtain the target retrieval result, based on a game-theoretic iterative optimization method. The knowledge storage module is used to store professional knowledge meta-models, manual coordinate information, knowledge atoms, and retrieval strategies; The interaction module is used to obtain user query information, display the target search results and feedback.

[0019] The beneficial effects of this invention are as follows: 1. Unified knowledge across multiple manuals: By using a professional knowledge meta-model, the problem of knowledge chaos in multiple versions and systems of manuals is solved, achieving centralized management of basic component knowledge. Engineers no longer need to manually filter information, significantly improving knowledge acquisition efficiency; 2. Improved retrieval efficiency: Based on scenario-based manual selection and multi-strategy retrieval, irrelevant information traversal is avoided, and relevant knowledge of basic components is quickly located; combined with path caching and generalization, the retrieval efficiency of high-frequency queries is further improved, meeting the real-time needs of engineering scenarios; 3. Guaranteed answer quality: Through dual-agent game theory, answer defects (such as missing parameters or omitted working condition constraints) are actively checked and corrected to ensure the completeness and accuracy of the answers; Simultaneously, it marks the source of knowledge, enhances the credibility of answers, and reduces errors in the selection and calculation of basic components; 4. Wide applicability: It is compatible with various basic component manuals such as bolts, bearings, gears, and electronic components, and supports applications in multiple fields such as machinery, electronics, hydraulics, and pneumatics. It does not require system reconstruction for different components and has high practical value; 5. Strong engineering applicability: It fits the actual needs of basic component selection, calculation, and maintenance, and provides complete practical knowledge (such as parameter sources, applicable conditions, and calculation examples), directly serving engineering design and production maintenance, and reducing enterprise knowledge management costs; It solves the problems that existing technologies cannot solve in the process of storing basic component manuals, such as large quantity, scattered distribution, chaotic management, slow retrieval speed, and offset and missing search results.

[0020] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0021] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a flowchart illustrating an intelligent knowledge retrieval method for basic component manuals according to the present invention. Figure 2 This is a professional knowledge meta-model structure diagram of an intelligent knowledge retrieval method for basic component manuals according to the present invention; Figure 3 This is a schematic diagram of the coordinate annotation of a manual for an intelligent knowledge retrieval method for basic component manuals according to the present invention. Figure 4 This is a flowchart of a two-agent game theory process for an intelligent knowledge retrieval method for basic component manuals according to the present invention. Figure 5 This is an example diagram of the basic component retrieval process of an intelligent knowledge retrieval method for basic component manuals according to the present invention; Figure 6 This is a system architecture diagram of an intelligent knowledge retrieval system for basic component manuals according to the present invention. Detailed Implementation

[0022] like Figures 1-5 As shown, this invention provides an intelligent knowledge retrieval method for basic component manuals, comprising the following steps: A professional knowledge meta-model is constructed based on the professional knowledge obtained from the basic documents; the professional knowledge meta-model includes: basic component knowledge classification, cross-standard order mapping, and knowledge dependency graph; Obtain the basic component manuals and determine the manual coordinate information; the manual coordinate information includes: time dimension, organization dimension, field dimension and authority information; Establish a mapping relationship between the chapter content of the basic component manual and the basic component knowledge classification, and indicate the reasons for the differences in the content of manuals within the same basic component knowledge classification; The chapter content is analyzed to obtain the analyzed content, and the mapped relationships and manual coordinate information are marked on the analyzed content to obtain knowledge atoms; The user query information is parsed and retrieved into a professional knowledge metamodel, and a retrieval is performed to obtain an initial knowledge set; Based on the game theory iterative optimization method, an initial answer is generated from the initial knowledge set and iteratively optimized to obtain the target retrieval result.

[0023] The working principle of the above technical solution is as follows: In the application of basic component manuals, the following scenarios are particularly important for retrieving basic component data. This invention is applicable to the following basic component manual scenarios: Mechanical basic component manuals: selection and calculation manuals for standard parts such as bolts, nuts, bearings, gears, and couplings; Material basic manuals: performance and selection manuals for basic materials such as metallic materials and polymer materials; Electronic basic component manuals: selection and parameter manuals for general electronic components such as resistors, capacitors, and inductors; Hydraulic and pneumatic basic component manuals: selection and calculation manuals for basic hydraulic and pneumatic components such as hydraulic valves, cylinders, and air pipes; Process basic manuals: process operation manuals for basic components such as machining, heat treatment, and surface treatment; For different scenarios, the knowledge dimension classification rules and retrieval strategies are adjusted to adapt to the characteristics of various basic component manuals. Basic component domain professional knowledge metamodel A unified knowledge framework across manuals: Defining basic component knowledge categories (such as product classification, material properties, and calculation methods) from a professional perspective, independent of specific manual chapters, and establishing cross-standard equivalence mappings (such as the correlation between bolt strength grades in GB and ISO), thus resolving the problem of knowledge confusion across multiple manuals.

[0024] Knowledge dependency logic analysis: Construct parameter dependency graphs (such as the dependency between bearing life and rated load, operating condition coefficient), clarify the essential connections between knowledge, and provide a basis for accurate retrieval and information completion.

[0025] Intelligent scheduling and knowledge integration Scenario-based manual selection: Based on user scenarios (industry, region, enterprise) and manual coordinates (version, standard system), the system automatically selects the appropriate manual, avoiding manual screening; it resolves conflicts between multiple manual versions and standards, and presents applicable knowledge in a priority manner according to rules.

[0026] Cross-manual knowledge complementarity: When the information in a single manual is incomplete, supplement it from other manuals under the same basic component knowledge category to ensure that complete basic component knowledge is obtained (such as supplementing the material corrosion resistance description missing in the national standard manual from the ISO manual).

[0027] Dual-agent iterative optimization Proactive defect verification: By retrieving answers generated by the agent and verifying the agent's defects (such as missing parameters or omitted operating conditions), an iterative process of "generation-questioning-correction" is formed to ensure the completeness and accuracy of the answers.

[0028] Scenario-based strategy optimization: Optimize the reward mechanism based on basic component retrieval scenarios (such as selection and calculation) to improve retrieval efficiency and answer quality, accumulate general retrieval patterns (such as bearing selection must check life, load, and operating condition coefficient), and reuse optimized paths; The manual coordinate information is as follows: For imported basic component manuals (such as GB's Fastener Design Manual, Bearing-Mechanical Design Manual (single volume)), the following dimensions are marked: time dimension (version number, year of publication, validity period), organization dimension (standard type, issuing organization, standard number and corresponding international standard), field dimension (applicable industry, basic component type) and authority information (scope of application, priority, conflict resolution rules), which serve as the manual coordinate information. It is worth noting that when annotating coordinate information in the parts manual, the basic parts manual can also be annotated from a knowledge dimension. The process of identifying the knowledge dimension of the basic parts manual is as follows: Product information: includes static description information such as model, specifications, dimensions, and weight of the basic parts, such as "M12 bolt size table" and "6205 bearing basic parameters"; Material information: includes the material grade, performance parameters (such as tensile strength and hardness), and applicable environment of the basic parts, such as "bolt material mechanical property table" and "bearing steel applicable temperature range"; Calculation method: includes the selection formula, strength verification steps, and life calculation method of the basic parts, such as "bolt preload calculation formula" and "bearing life verification process"; Test method: includes the inspection items, test steps, and acceptance standards of the basic parts, such as "bolt torque test method" and "bearing vibration detection specification". The beneficial effects of the above technical solution are as follows: 1. Unified knowledge across multiple manuals: By using a professional knowledge meta-model, the problem of knowledge chaos across multiple versions and systems of manuals is solved, enabling centralized management of basic component knowledge. Engineers no longer need to manually filter information, significantly improving knowledge acquisition efficiency. 2. Improved retrieval efficiency: Based on scenario-based manual selection and multi-strategy retrieval, irrelevant information traversal is avoided, and relevant knowledge of basic components is quickly located. Combined with path caching and generalization, the retrieval efficiency of high-frequency queries is further improved, meeting the real-time needs of engineering scenarios. 3. Guaranteed answer quality: Through dual-agent game theory, answer defects (such as missing parameters or omitted operating condition constraints) are actively checked and corrected. 4. **Wide Applicability:** Adapts to various basic component manuals such as bolts, bearings, gears, and electronic components, supporting applications in multiple fields including machinery, electronics, hydraulics, and pneumatics. It eliminates the need for system reconstruction for different components, offering high practical value. 5. **Strong Engineering Practicality:** Meets the actual needs of basic component selection, calculation, and maintenance, providing complete practical knowledge (such as parameter sources, applicable conditions, and calculation examples), directly serving engineering design and production maintenance, reducing enterprise knowledge management costs. It solves the problems of large quantity, scattered distribution, and chaotic management in the storage of basic component manuals, as well as slow retrieval speed, offset search results, and missing information that existing technologies cannot address.

[0029] In one embodiment, constructing a professional knowledge meta-model based on the acquired basic document professional knowledge includes: Determine the primary classifications, subcategories, core concepts, standard terminology, and parameter association rules in the acquired basic document professional knowledge; Based on natural language processing technology, we analyze basic component manuals and industry standard documents to extract high-frequency knowledge patterns and cross-standard equivalent terms to obtain cross-standard hierarchical mappings. Construct a knowledge dependency graph based on the dependencies of professional knowledge in the basic files and the acquisition methods of each parameter; When importing a new basic component manual, the coverage of the professional knowledge metamodel is checked, and the professional knowledge metamodel is updated based on the check results.

[0030] The working principle and beneficial effects of the above technical solution are as follows: A professional knowledge meta-model is constructed based on the acquired basic document professional knowledge. Specifically, firstly, a domain knowledge classification ontology (i.e., basic component knowledge classification) is defined: independent of specific manuals, from the perspective of basic component engineering, a general knowledge classification system is constructed that includes product taxonomy, materials science, computational methodology, testing methodology, and applied engineering. Each meta-classification clarifies the essential relationships between core concepts, standard terms, and key parameters; for example, "bolt strength calculation" is associated with "material strength grade" and "stress cross-sectional area." Secondly, cross-standard equivalence mapping is established: equivalent technical concepts in different standard systems (such as GB, ISO, DIN) in the field of basic components are identified; for example, "8.8 grade bolt" in GB, "Property class 8.8" in ISO, and "8.8 grade Schraube" in DIN are mapped to the same basic component knowledge classification. Equivalent conversion relationships for physical parameter units (such as MPa and N / mm², kgf / mm²) are established. (conversion); Finally, construct a knowledge dependency graph: at the professional knowledge meta-model level, sort out the logical dependencies between basic component knowledge, such as "bearing life calculation" depends on parameters such as "rated dynamic load", "equivalent dynamic load" and "operating condition coefficient", and clarify the acquisition method of each parameter (look up table or calculate). In constructing the professional knowledge metamodel, the initial framework is defined under the guidance of domain experts. Specifically, experts in basic component engineering determine primary classifications and subcategories such as product taxonomy and materials science, clarifying core concepts, standard terminology, and parameter association rules. This is then expanded using professional text analysis techniques: natural language processing is used to analyze multiple versions and systems of basic component manuals and industry standard documents, extracting high-frequency knowledge patterns and cross-standard equivalent terms to refine the metamodel classification and mapping relationships. Simultaneously, iterative optimization is performed to adapt to new scenarios: when importing new manuals (such as newly added basic component type manuals), the metamodel coverage is checked. If coverage is insufficient, the uncovered knowledge classifications or mapping rules are expanded to ensure adaptation to various basic component manuals. Specifically, in the import... After importing the new basic component manual, the existing basic component manual will be updated simultaneously with the professional knowledge meta-model. Specifically: Version update detection: Regularly check the current version of the imported basic component manual and prompt users to update when a new version is released; Incremental import of differences: Only import the differences between the new and old versions (such as newly added parameter tables and revised formula descriptions), without needing to re-import the complete manual; Update mappings and atoms: Map the differences in the new version to the professional knowledge meta-model, update the manual coordinate information of the relevant knowledge atoms, and mark the substitution relationships of the old version atoms; Historical version tracking: Retain the old version manual data and support users to query historical version information (such as querying parameters in the old version manual of a certain basic component).

[0031] In one embodiment, a knowledge dependency graph is constructed based on the dependencies of expertise in the underlying files and the methods for obtaining each parameter, including: The calculation formulas, parameters, and parameter descriptions are extracted from the calculation method section of the basic component manual. Determine the dependencies between parameters in each formula; where dependencies include direct dependencies and indirect dependencies. A directed knowledge dependency graph is constructed with parameters as nodes and dependencies as edges, and the acquisition method of each parameter and the chapter from the basic component manual are labeled.

[0032] The working principle and beneficial effects of the above technical solution are as follows: In the process of constructing a knowledge dependency graph, the logical dependencies between basic component knowledge are sorted out. For example, "bearing life calculation" depends on parameters such as "rated dynamic load", "equivalent dynamic load", and "operating condition coefficient", and the acquisition method of each parameter is clarified (table lookup or calculation). Specifically, calculation-related knowledge is extracted: calculation formulas and corresponding parameter descriptions are extracted from the calculation method content of the basic component manual. Parameter dependencies are analyzed: the direct dependencies (such as "allowable stress" depending on "material tensile strength") and indirect dependencies (such as "material tensile strength" depending on "material grade") of each parameter in the formula are clarified. A knowledge dependency graph is constructed: with parameters as nodes and dependencies as edges, a directed dependency graph is constructed, and the acquisition method (such as table lookup or calculation) and source manual chapter of each parameter are marked, finally obtaining the knowledge dependency graph.

[0033] In one embodiment, a mapping relationship is established between the chapter content of the basic component manual and the basic component knowledge classification, and the reasons for the differences in the manual content of the same basic component knowledge classification are indicated, including: Determine the semantic relationship between the chapter titles of the basic component manual and the category names of the basic component knowledge classifications; Determine the degree of matching between the key information in the chapter content of the basic component manual and the characteristics of the basic component knowledge classification; The mapping relationship between the chapter content of the basic component manual and the basic component knowledge classification is obtained based on semantic relevance and matching degree. Reasons for the differences in the content of manuals that classify knowledge of the same basic components.

[0034] The working principle and beneficial effects of the above technical solution are as follows: The solution identifies the meta-knowledge classification corresponding to the chapter content of the manual. Even if different manuals have different chapter numbers or titles, they are still mapped to a unified basic component knowledge classification. For example, "Chapter 3 Bolt Strength Grade" in the GB Fastener Design Manual and "Section 5 Bolt Property Classes" in the ECSS Threaded Fastener Manual are both mapped to the basic component knowledge classification of "Bolt Strength Grade". Specifically: Semantic matching association: Compare the semantic association between the manual chapter titles and the meta-model classification names. For example, the chapter "Bearing Rated Load Table" is associated with the basic component knowledge classification of "Bearing Performance Parameters". Content feature identification: Extract key information from the manual chapters (such as parameter table structure, formula type, and operation verbs) and match it with the feature library of the professional knowledge meta-model classification. For example, a chapter containing "Selection Calculation Steps" is associated with the basic component knowledge classification of "Basic Component Selection Methods". Finally, manual verification and improvement: For chapters with low automatic mapping reliability, domain experts review and confirm them, and a verification case library is established to optimize the automatic mapping logic. Meanwhile, during the mapping process, record parameter difference information: for multiple manual contents mapped to the same basic component knowledge category, record the differences in parameter descriptions or values, and mark the reasons for the differences (such as version updates, different standard application scenarios).

[0035] In one embodiment, parsing chapter content yields parsed content, and annotating the parsed content with mapping relationships and manual coordinate information yields knowledge atoms, including: The text, images, tables, and formulas in the basic component manual are classified and parsed. Specifically, for table and image data, a dedicated large model with the ability to parse the format is called to complete the parsing of the manual content and obtain the parsed content. The parsed content is segmented according to semantic integrity; The parsed content is labeled with mapping relationships and manual coordinate information to obtain knowledge atoms.

[0036] The working principle and beneficial effects of the above technical solution are as follows: The text, images, tables, and formulas in the manual are categorized and parsed. Specifically, for table and image data, table parsing agents and image parsing agents are defined, and a dedicated large model with the ability to parse the format is called to complete the parsing of the manual content, obtaining the parsed content. Then, the parsed content is segmented according to semantic integrity, ensuring that each knowledge atom contains complete and practical information (such as a single calculation formula and parameter description, parameter table and applicable scope), avoiding the separation of formulas and parameters, and the disconnection between descriptions and tables. Simultaneously with the segmentation, dual attribution is marked: each segmented parsed content is marked with the professional knowledge meta-model attribution (corresponding to the basic component knowledge category) and manual coordinate information (source manual and version) to obtain the knowledge atom. For example, the knowledge atom "8.8 grade bolt room temperature tensile strength" is marked with its attribution to the basic component knowledge category "bolt strength grade" and its source "GB / T 3098.1". Finally, multi-source knowledge atoms are managed: multiple source knowledge atoms are allowed under the same basic component knowledge category, and the relationships between atoms (equivalence, supplementation, substitution) are marked, such as "8.8 grade bolt high temperature strength" in the enterprise design standard. Knowledge atoms serve as a supplement to the corresponding atoms in the national standard manual; In one embodiment, user query information is parsed and retrieved into a professional knowledge metamodel, and a retrieval is performed to obtain an initial knowledge set, including: The search results are obtained by parsing the user's query information; the search results include: query intent and query scenario. Based on the query intent, key technology concepts are identified and a retrieval mapping relationship is established with basic component knowledge classification. Based on preset manual selection rules, the appropriate manual is determined according to the query scenario and manual coordinate information; the preset manual selection rules include: priority rules and knowledge conflict and supplementation rules. Based on a pre-defined multi-strategy retrieval method, an initial knowledge set is obtained according to the adaptation manual and retrieval mapping relationship; Knowledge conflict and supplement rules include: If a knowledge conflict is detected in the basic component manual, the appropriate manual shall be selected according to the knowledge conflict resolution rules. If the detected basic component manual has incomplete information, it will be supplemented from other basic component manuals under the same basic component knowledge category; The rules for resolving knowledge conflicts include: Version conflict: When different versions of the basic component manual under the same organization have different descriptions of the same parameter, the current valid version shall prevail, and the parameter information of the historical versions shall be marked for traceability. Standard system conflict: When different standard systems have different values ​​for the same concept parameter, the parameters of each standard are presented side by side, prompting the user to select the appropriate standard for the project and explaining the reasons for the differences.

[0037] Conflict between enterprise and general standards: The enterprise's own basic component manual has stricter requirements than the general standards, while presenting the enterprise's requirements and the general standard benchmark values, and indicating the scope of application of the enterprise's requirements.

[0038] Conflicts and contradictions: When different basic component manuals present contradictory statements on the same issue and cannot be resolved by version conflict, standard system conflict, or conflict between enterprise and general standards, mark the conflict and provide the basis for each basic component manual, and offer suggestions in conjunction with consulting experts in the field based on the actual engineering situation.

[0039] The working principle and beneficial effects of the above technical solution are as follows: Analyze the acquired user query information to obtain query intent and query scenario; retrieve query intents for basic components (such as bolt selection, bearing life calculation), extract key technical concepts and map them to a professional knowledge meta-model, avoiding reliance on specific manual chapter structures; then select applicable manual examples: based on the user query scenario (such as industry, region, enterprise) and manual coordinate information, select suitable manuals. For example, domestic machinery companies prioritize national standard manuals when querying bolt parameters, while international projects prioritize ISO manuals; for multiple versions of manuals from the same organization, prioritize the currently valid version; when selecting suitable manuals, the selection is based on preset manual selection rules, including priority rules and knowledge conflict and supplementation rules; the priority rules are: First priority: the user's company's internal basic component manual (if it exists), adapting to the company's specific selection requirements; Second priority: standard manuals explicitly specified in the query (such as the user mentioning "according to GB standards"), performing the search according to the specified standard; Third priority: mainstream standard manuals in the user's region (such as domestic users prioritizing national standards, European users prioritizing EN standards). The first priority is to select the latest version of the manual within the same standard system, adapting it to local engineering practices. The second priority is to select the most recently released and valid version within the same standard system. The third priority is to select the manual with complete knowledge coverage when querying multiple types of knowledge, minimizing cross-manual splicing. When the selected manual has knowledge conflicts or omissions, it is supplemented using knowledge conflict and supplementation rules. When knowledge conflicts exist, they are handled according to knowledge conflict resolution rules. When knowledge is missing, it is supplemented using other basic component manuals. The supplementation rules are as follows: Information completion: When a manual does not explain a parameter of a basic component (e.g., the national standard manual does not mention the salt spray resistance of bearing materials), relevant information is extracted from other manuals (e.g., ISO manuals) under the same basic component knowledge category, and the source of the information is marked. Cross-validation: When multiple manuals have consistent descriptions of the same parameter, it is clearly stated that "multiple standards have consistent descriptions" to enhance the credibility of the answer. When there are differences in descriptions, they are handled according to knowledge conflict resolution rules. Knowledge chain integration: When calculations related to basic components require cross-category knowledge (e.g., bolt strength verification requires material parameters, stress formulas, and safety factors), corresponding knowledge is extracted from different manuals to form a complete calculation chain, and the source of information for each link is marked. Finally, based on the preset multi-strategy retrieval method, an initial knowledge set is obtained according to the adaptation manual and retrieval mapping relationship; specifically, according to the query type (such as parameter query, selection calculation, test process), a hierarchical retrieval path (chapter-level positioning, parameter-level completion) is selected, and applicable retrieval tools (chapter positioning, entity query, parameter completion, formula matching) are combined to retrieve the manual content to obtain the initial knowledge set.

[0040] In one embodiment, based on a game-theoretic iterative optimization method, an initial answer is generated from an initial knowledge set and iteratively optimized to obtain the target retrieval result, including: Construct adversarial dual agents; wherein the adversarial dual agents include: a retrieval agent and a verification agent; Based on the game-theoretic iterative optimization method, the retrieval agent and the verification agent generate an initial answer based on the initial knowledge set and iteratively optimize it to obtain the target retrieval result. Specifically, the retrieval agent generates the initial answer according to the retrieval strategy, and the verification agent raises questions about the initial answer according to the questioning strategy and provides manual support. The iteration terminates when there are no effective questions for several consecutive rounds or when a reasonable number of iterations is reached. Based on a game-theoretic iterative optimization method, an initial answer is generated from an initial knowledge set and iteratively optimized to obtain the target retrieval result. This also includes: The retrieval strategy of the retrieval agent and the questioning strategy of the verification agent are optimized based on the comprehensive reward mechanism to obtain the updated retrieval strategy and the updated questioning strategy. The retrieval agent and the verification agent are updated according to the updated retrieval strategy and the updated challenge strategy, respectively; The comprehensive reward mechanism includes: Search efficiency: Combining the number of search steps and the hit rate of effective knowledge units, the fewer the steps and the higher the hit rate, the higher the reward weight; Reasoning completeness: Evaluate the completeness of the reasoning logic in the answer and the degree to which parameter dependencies are satisfied; Answer accuracy: Assess the consistency between the answer and the content of the adaptation manual, as well as the traceability of the information source; Weighting adaptation scenarios: Adjust weights for different query types of basic components. Parameter queries focus on retrieval efficiency, selection calculations focus on reasoning completeness, and material recommendations focus on answer accuracy.

[0041] The working principle and beneficial effects of the above technical solution are as follows: In practice, an adversarial dual-agent system is constructed: the retrieval agent (R-Agent) plans the retrieval path and generates answers related to basic components (such as selection schemes and calculation results) based on the meta-model and manual information; the verification agent (V-Agent) checks the completeness and accuracy of the answers from the manual and identifies defects (such as missing parameters and omissions of working condition constraints); game iteration is executed: after the R-Agent generates the initial answer, the V-Agent raises questions and provides manual evidence; the R-Agent supplements the retrieval and corrects the answer based on the questions, iterating until the V-Agent has no effective questions for several consecutive rounds or reaches a reasonable number of iterations; the retrieval strategy is optimized: based on a comprehensive reward mechanism (retrieval efficiency, reasoning completeness, and answer accuracy), the retrieval path of the R-Agent and the questioning logic of the V-Agent are optimized, and the optimized paths of high-frequency queries are accumulated for subsequent retrieval; When the retrieval agent searches for knowledge bodies in the initial knowledge set, the retrieval strategies include the following two types: (I) When selecting materials for basic components, the strategy is as follows: 1. Extract constraints from the user query: working environment constraints (such as temperature, corrosive media), performance requirement constraints (such as strength, wear resistance), and economic constraints (such as cost range); 2. Preliminary screening of candidate materials: based on the material information index, screen material types that meet the basic constraints; 3. Verify material performance parameters: obtain detailed performance parameters of candidate materials through parameter dependency graphs and verify whether they meet all constraints; 4. Output suitable material solutions: sort by constraint satisfaction and economy, output a recommended list of materials, with material performance descriptions and sources. (ii) When searching for calculation methods for basic components, the following strategies are adopted: 1. Locate relevant calculation formulas: Search and query corresponding calculation formulas (such as bearing life formulas and bolt strength formulas) through the calculation method index; 2. Analyze the calculation logic path: For multi-step calculations (such as bearing selection under complex working conditions), trace back the required formulas and parameters from the final calculation target to form a complete calculation path; 3. Complete the required calculation parameters: Complete the missing parameters in the calculation process through the parameter dependency graph (such as obtaining them from the manual or through the calculation of the preceding steps); 4. Output complete calculation results: Present the calculation steps, the formulas used and their sources, and the basis for parameter values. If there are similar calculation examples in the manual, provide them for reference at the same time. The verification strategy for the intelligent agent is as follows: 1. Parameter completeness check: Check whether the calculation-related parameters in the answer are complete. If there are any unclear parameters (such as missing operating condition coefficients in bearing life calculation), point out the missing parameters and the way to obtain them (such as consulting a certain chapter of the manual); 2. Operating condition constraint check: Check whether the recommended solution for basic components considers the operating condition constraints in the query (such as whether heat-resistant materials are selected in high-temperature environments). If constraints are omitted, raise questions by citing the corresponding constraint requirements in the manual; 3. Logical consistency check: Check whether the answer is consistent with the statements in other chapters of the manual (such as whether the performance parameters of the same material match). If there are contradictions, point out the contradictory chapters and contents; 4. Standard validity check: Check whether the standard cited in the answer is the current valid version. If an old version of the standard is cited, indicate the latest version and related revisions. The iteration termination conditions are as follows: Consensus Termination: When the verification agent does not effectively question the retrieval agent's answer for two consecutive rounds, it is determined that a consensus has been reached, and the final answer is output; Number Limit Termination: When the number of iterations reaches the preset limit (a reasonable number is set according to the engineering scenario) and a consensus has not been reached, the current optimal answer is output and the content that needs manual confirmation is marked; Convergence Termination: When the difference between the answers corrected by the retrieval agent for two consecutive rounds is small, it is determined that the strategy has converged, the final answer is output and the remaining question points are marked; Deadlock Termination: When it is detected that the retrieval agent and the verification agent are stuck in a circular argument, the iteration is terminated and all candidate answers are output, and deadlock cases are recorded for subsequent optimization; The retrieval process also includes the optimization and reuse of retrieval paths: Path caching: For cases with excellent retrieval results (such as complete answers and high efficiency), cache query scenarios (such as "high temperature bolt selection"), retrieval paths (such as "material index → ​​parameter completion → calculation verification") and result information; Similar query reuse: When a new query is similar to a cached scenario, reuse the cached retrieval path or use it as the initial path for fine-tuning to reduce redundant planning; Path generalization: Regularly analyze cached paths, extract common retrieval patterns for basic components (such as "bearing selection requires checking rated load, life formula, and operating condition coefficient"), and solidify them as the default retrieval strategy.

[0042] like Figure 6 As shown, the present invention also provides an intelligent knowledge retrieval system for basic component manuals, comprising: The metamodel building module is used to build a professional knowledge metamodel based on the professional knowledge obtained from the basic documents. The manual management module is used to obtain basic component manuals and determine the manual coordinate information of the basic component manuals. The manual coordinate information includes: time dimension, organization dimension, domain dimension and authority information; establish the mapping relationship between the chapter content of the basic component manuals and the basic component knowledge classification, and mark the reasons for the differences in the manual content of the same basic component knowledge classification. The knowledge atom processing module is used to parse chapter content to obtain parsed content, and to annotate the parsed content with mapping relationships and manual coordinate information to obtain knowledge atoms. The query scheduling module is used to parse and obtain user query information into the professional knowledge metamodel and perform retrieval to obtain the initial knowledge set; The dual-agent optimization module is used to generate an initial answer based on the initial knowledge set and iteratively optimize it to obtain the target retrieval result, based on a game-theoretic iterative optimization method. The knowledge storage module is used to store professional knowledge meta-models, manual coordinate information, knowledge atoms, and retrieval strategies; The interaction module is used to obtain user query information, display the target search results and feedback.

[0043] The working principle and beneficial effects of the above technical solution: The knowledge dependency graph in the knowledge storage module is stored using a graph database and supports the following operations: Path query: query the dependency path between two parameters (such as the dependency relationship between "bearing life" and "material grade"); Subgraph extraction: extract dependency subgraphs related to a certain type of parameter (such as parameter dependencies related to bolt strength calculation); Impact analysis: query the impact of a change in a parameter on other parameters (such as the impact of a change in material strength on the allowable load of a bolt); Anomaly detection: identify loop logic in parameter dependencies (which may be an error in the manual) and prompt manual verification; The retrieval execution component in the query scheduling module includes the following tools: Chapter location tool: quickly locates the relevant chapter in the manual (e.g., locating the "Bolt Strength Calculation" chapter); Entity query tool: retrieves relevant information about basic component entities (e.g., the dimensions and material of "M12 Bolt"); Parameter completion tool: completes the parameters required for calculation based on the dependency graph (e.g., completes the operating condition coefficient in bearing life calculation); Formula matching tool: matches the corresponding calculation formula in the query (e.g., matches the relevant formula for bolt preload calculation); Unit conversion tool: automatically identifies the units of parameters in the query and the manual, and completes equivalent conversions (e.g., conversion between MPa and N / mm²).

[0044] The interactive module supports the following functions: Query input: receiving natural language queries from users regarding basic components (such as "how to select bolts for high-temperature environments"); Result display: presenting search results (such as selection schemes, calculation results) and information sources (manual name, chapter); Version tracking: allowing users to view the description of the same parameter in different manual versions; Standard comparison: allowing users to compare the parameter differences of the same basic component in different standard systems (such as GB and ISO).

[0045] This invention also provides the following methods and system application examples, as shown below: Example 1: Bolt selection cross-manual search Scenario: An engineer from a domestic machinery company is looking up "selection scheme for M12 bolts connecting 8mm steel plates at room temperature and bearing a working load of 20kN". The company has imported GB / T 3098.1-2010 (Mechanical properties of fasteners: bolts, screws and studs), ISO 898-1:2022 (International standard for fastener performance) and the company's bolt selection specifications (2023 version).

[0046] 1. Query parsing and meta-model mapping: In the query, “M12 bolt selection” is mapped to the basic component knowledge category “bolt selection method”, and “20kN working load” and “8mm steel plate” are mapped to the basic component knowledge categories “bolt load calculation” and “bolt connection working conditions”.

[0047] 2. Manual selection: Based on the user scenario (domestic machinery enterprises), prioritize internal enterprise specifications (first priority), followed by GB / T 3098.1-2023 (third priority, the mainstream domestic standard).

[0048] 3. Knowledge Integration and Retrieval: 4. Extract the "M12 bolt stress cross-sectional area" and "bolt strength grade requirements" from the company's specifications (the company's requirements are stricter than the national standard, requiring grade 8.8 or above).

[0049] 5. Supplement the GB manual with “Mechanical properties of grade 8.8 bolts at room temperature” and “Calculation method for preload”, and verify the consistency between the enterprise specifications and the national standard parameters (there is no conflict between the strength requirements of the two, and the enterprise has added economic constraints).

[0050] 6. Search according to the “Selection Calculation Steps”: Calculate the required bolt strength → Screen the material grade (8.8 grade) → Determine the preload → Recommend surface treatment (in combination with the “slightly corrosive environment” requirement in the enterprise specifications, galvanizing is recommended).

[0051] 1. Optimization of dual-agent game theory: 2. The initial answer retrieved by the agent: "Recommended 8.8 grade M12 bolts, preload calculated according to the formula, and galvanized surface treatment."

[0052] 3. Verification agent question: "The basis for bolt length selection (which needs to be determined in conjunction with steel plate thickness) and reference value for preload torque are not explained." The company specification "bolt length selection must cover the thickness of the connecting parts + nut height" and the GB manual "preload torque calculation method" are cited.

[0053] 4. Search agent correction: Add "Recommended bolt length 25-30mm (covering 8mm×2 steel plate + nut height), preload torque reference 72N・m (calculated according to GB manual formula)", verify that the agent has no further questions, and output the final answer.

[0054] Practical benefits: Engineers no longer need to manually consult multiple manuals; a single search provides a complete bolt selection solution (including material, length, surface treatment, and preload torque), avoiding omissions due to scattered knowledge and improving selection efficiency.

[0055] Example 2: Bearing life calculation retrieval Scenario: An engineer from a general equipment factory is looking up the "method for calculating the life of bearings for the output shaft of a reducer", referring to GB / T 272-2017 (National Standard Bearing Handbook) and ISO 15:2021 (International Standard Bearing Handbook).

[0056] 1. Meta-model mapping and knowledge organization: Querying "bearing life calculation" maps to the basic component knowledge category "bearing life calculation method", which relies on basic component knowledge categories such as "rated dynamic load", "equivalent dynamic load", and "operating condition coefficient".

[0057] 2. Multi-strategy retrieval: 3. Chapter-level positioning: The core formulas and parameter descriptions are located in Chapter 3 of the GB manual, "Bearing Life Calculation".

[0058] 4. Parameter-level completion: By using the dependency graph, it is clarified that "equivalent dynamic load needs to be calculated in conjunction with radial / axial load and load factor" and "operating condition factor needs to be selected with reference to the working scenario of the reducer". The load factor query method and recommended value of operating condition factor are automatically extracted from the manual.

[0059] 5. Cross-manual verification: Verify the consistency of lifespan formulas and parameter descriptions in GB and ISO manuals (the formulas are the same, but the parameter value ranges are slightly different). GB parameters should be used first for domestic scenarios.

[0060] 1. Answer generation and optimization: 2. The intelligent agent outputs the calculation steps: "1. Determine the bearing model and rated dynamic load; 2. Calculate the equivalent dynamic load (including load factor); 3. Determine the operating condition factor; 4. Substitute into the life formula for calculation", and indicate the manual reference for each step.

[0061] 3. Verify the agent: Confirm that there are no missing parameters, the applicable conditions of the formula are clear, there are no questions, and output the final calculation method and reference example (referencing the reducer bearing calculation case in the GB manual).

[0062] Practical benefits: Engineers can quickly obtain a complete bearing life calculation process, clarify the source and basis of each parameter, avoid calculation errors caused by missing parameters or misuse of formulas, and meet the bearing life verification needs in reducer design.

[0063] Example 3: Parameter Traceability in Multiple Version Manuals Scenario: An old equipment maintenance engineer is looking up the "tensile strength parameters of M10 bolts in the old version of GB manual". The company has imported GB / T 3098.1-2010 (current version) and GB / T 3098.1-2000 (old version).

[0064] 1. Query Analysis and Manual Selection: Query the "Old Version GB Manual" and clearly select GB / T 3098.1-2010, which maps to the basic component knowledge category "Bolt Strength Grade Parameter".

[0065] 2. Parameter retrieval and presentation: Extract the "M10 bolt 8.8 grade tensile strength parameter" from the old version of the manual, and present the corresponding parameter in the current version, mark the version differences (if any), and explain the applicable scenarios of the old version parameters (such as maintenance of old equipment).

[0066] Practical benefits: Engineers no longer need to manually search through old paper manuals, and can quickly obtain the historical parameters required for the maintenance of old equipment, ensuring the compatibility of parts during the repair of old equipment.

[0067] It is worth noting that the above-mentioned intelligent retrieval method and system for basic component manuals is deployed and used in practice according to the following process: 1. Hardware Deployment: This system can be deployed on enterprise servers or in the cloud, supports access from multiple terminals (such as engineers' office computers and workshop mobile terminals), and is adapted to the usage environment of industrial scenarios; the basic configuration can meet the basic component manual management needs of small and medium-sized enterprises, and the deployment cost is low.

[0068] 2. Software compatibility: The metamodel construction supports integration with mainstream natural language processing tools, knowledge storage adopts general graph databases and relational databases, the retrieval component is compatible with common search engines, does not depend on special software, and is compatible with the enterprise's existing IT architecture.

[0069] 3. Scenario Expansion: When adding new basic component types (such as adding a sealing manual), only the knowledge classification and mapping rules of the meta-model need to be expanded. There is no need to modify the core modules of the system, resulting in low expansion costs and adaptability to enterprise business expansion needs.

[0070] 4. Data Security: Supports encrypted storage and access control of internal enterprise manuals, ensuring the security of sensitive knowledge such as internal enterprise regulations, and meeting the data confidentiality requirements of industrial scenarios.

[0071] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. An intelligent knowledge retrieval method for basic component manuals, characterized in that, Includes the following steps: A professional knowledge meta-model is constructed based on the professional knowledge obtained from the basic documents; the professional knowledge meta-model includes: basic component knowledge classification, cross-standard order mapping, and knowledge dependency graph; Obtain the basic component manuals and determine the manual coordinate information; the manual coordinate information includes: time dimension, organization dimension, field dimension and authority information; Establish a mapping relationship between the chapter content of the basic component manual and the basic component knowledge classification, and indicate the reasons for the differences in the manual content of the same basic component knowledge classification; The chapter content is analyzed to obtain the analyzed content, and the mapped relationships and manual coordinate information are marked on the analyzed content to obtain knowledge atoms; The user query information is parsed and retrieved into a professional knowledge metamodel, and a retrieval is performed to obtain an initial knowledge set; Based on the game theory iterative optimization method, an initial answer is generated from the initial knowledge set and iteratively optimized to obtain the target retrieval result.

2. The intelligent knowledge retrieval method for basic component manuals according to claim 1, characterized in that, Construct a professional knowledge meta-model based on the acquired basic document expertise, including: Determine the primary classifications, subclasses, core concepts, standard terminology, and parameter association rules in the acquired basic document professional knowledge; Based on natural language processing technology, we analyze basic component manuals and industry standard documents to extract high-frequency knowledge patterns and cross-standard equivalent terms to obtain cross-standard hierarchical mappings. Construct a knowledge dependency graph based on the dependencies of professional knowledge in the basic files and the acquisition methods of each parameter; When importing a new basic component manual, the coverage of the professional knowledge metamodel is checked, and the professional knowledge metamodel is updated based on the check results.

3. The intelligent knowledge retrieval method for basic component manuals according to claim 2, characterized in that, A knowledge dependency graph is constructed based on the dependencies of professional knowledge in the underlying files and the methods for obtaining each parameter, including: The calculation formulas, parameters, and parameter descriptions are extracted from the calculation method section of the basic component manual. Determine the dependencies between parameters in each formula; where dependencies include direct dependencies and indirect dependencies. A directed knowledge dependency graph is constructed with parameters as nodes and dependencies as edges, and the acquisition method of each parameter and the chapter from the basic component manual are labeled.

4. The intelligent knowledge retrieval method for basic component manuals according to claim 1, characterized in that, Establish a mapping relationship between the chapter content of basic component manuals and the basic component knowledge categories, and indicate the reasons for the differences in the content of manuals within the same basic component knowledge category, including: Determine the semantic relationship between the chapter titles of the basic component manual and the category names of the basic component knowledge classifications; Determine the degree of matching between the key information in the chapter content of the basic component manual and the characteristics of the basic component knowledge classification; The mapping relationship between the chapter content of the basic component manual and the basic component knowledge classification is obtained based on semantic relevance and matching degree. Reasons for the differences in the content of manuals that classify knowledge of the same basic components.

5. The intelligent knowledge retrieval method for basic component manuals according to claim 1, characterized in that, The chapter content is parsed to obtain parsed content. The parsed content is then annotated with mapping relationships and manual coordinate information to obtain knowledge atoms, including: The text, images, tables, and formulas in the basic component manual are classified and parsed. Specifically, for table and image data, a dedicated large model with the ability to parse the format is called to complete the parsing of the manual content and obtain the parsed content. The parsed content is segmented according to semantic integrity; The parsed content is labeled with mapping relationships and manual coordinate information to obtain knowledge atoms.

6. The intelligent knowledge retrieval method for basic component manuals according to claim 1, characterized in that, The user query information is parsed and retrieved into a professional knowledge metamodel, and a retrieval is performed to obtain an initial knowledge set, including: The search results are obtained by parsing the user's query information; the search results include: query intent and query scenario. Based on the query intent, key technology concepts are identified and a retrieval mapping relationship is established with basic component knowledge classification. Based on preset manual selection rules, the appropriate manual is determined according to the query scenario and manual coordinate information; the preset manual selection rules include: priority rules and knowledge conflict and supplementation rules. Based on the preset multi-strategy retrieval method, an initial knowledge set is obtained according to the adaptation manual and retrieval mapping relationship.

7. The intelligent knowledge retrieval method for basic component manuals according to claim 6, characterized in that, Knowledge conflict and supplementation rules include: If a knowledge conflict is detected in the basic component manual, the appropriate manual shall be selected according to the knowledge conflict resolution rules. If the detected basic component manual has incomplete information, it will be supplemented from other basic component manuals under the same basic component knowledge category; The rules for resolving knowledge conflicts include: Version conflict: When different versions of the basic component manual under the same organization have different descriptions of the same parameter, the current valid version shall prevail, and the parameter information of the historical versions shall be marked for traceability. Standard system conflict: When different standard systems have different values ​​for the same concept parameter, the parameters of each standard are presented side by side, prompting the user to select the appropriate standard for the project and explaining the reasons for the differences. Conflict between enterprise and general standards: When the enterprise's own basic component manual requirements are stricter than the general standards, both the enterprise's requirements and the general standard benchmark values ​​should be presented, and the scope of application of the enterprise's requirements should be indicated. Conflicts and contradictions: When different basic component manuals present contradictory statements on the same issue and cannot be resolved by version conflict, standard system conflict, or conflict between enterprise and general standards, mark the conflict and provide the basis for each basic component manual, and offer suggestions in conjunction with consulting experts in the field based on the actual engineering situation.

8. The intelligent knowledge retrieval method for basic component manuals according to claim 1, characterized in that, Based on a game-theoretic iterative optimization method, an initial answer is generated from an initial knowledge set and iteratively optimized to obtain the target retrieval results, including: Construct adversarial dual agents; wherein the adversarial dual agents include: a retrieval agent and a verification agent; Based on the game-theoretic iterative optimization method, the retrieval agent and the verification agent generate an initial answer based on the initial knowledge set and iteratively optimize it to obtain the target retrieval result. Specifically, the retrieval agent generates the initial answer according to the retrieval strategy, and the verification agent raises questions about the initial answer according to the questioning strategy and provides manual evidence. The iteration terminates when there are no effective questions for several consecutive rounds or when a reasonable number of iterations is reached.

9. The intelligent knowledge retrieval method for basic component manuals according to claim 8, characterized in that, Based on a game-theoretic iterative optimization method, an initial answer is generated from an initial knowledge set and iteratively optimized to obtain the target retrieval result. This also includes: The retrieval strategy of the retrieval agent and the questioning strategy of the verification agent are optimized based on the comprehensive reward mechanism to obtain the updated retrieval strategy and the updated questioning strategy. The retrieval agent and the verification agent are updated according to the updated retrieval strategy and the updated challenge strategy, respectively; The comprehensive reward mechanism includes: Search efficiency: Combining the number of search steps and the hit rate of effective knowledge units, the fewer the steps and the higher the hit rate, the higher the reward weight; Reasoning completeness: Evaluate the completeness of the reasoning logic in the answer and the degree to which parameter dependencies are satisfied; Answer accuracy: Assess the consistency between the answer and the content of the adaptation manual, as well as the traceability of the information source; Weighting adaptation scenarios: Adjust weights for different query types of basic components. Parameter queries focus on retrieval efficiency, selection calculations focus on reasoning completeness, and material recommendations focus on answer accuracy.

10. An intelligent knowledge retrieval system for basic component manuals, characterized in that, include: The metamodel building module is used to build a professional knowledge metamodel based on the professional knowledge obtained from the basic documents. The manual management module is used to obtain basic component manuals and determine the manual coordinate information of the basic component manuals. The manual coordinate information includes: time dimension, organization dimension, domain dimension and authority information; establish the mapping relationship between the chapter content of the basic component manuals and the basic component knowledge classification, and mark the reasons for the differences in the manual content of the same basic component knowledge classification. The knowledge atom processing module is used to parse chapter content to obtain parsed content, and to annotate the parsed content with mapping relationships and manual coordinate information to obtain knowledge atoms. The query scheduling module is used to parse and obtain user query information into the professional knowledge metamodel and perform retrieval to obtain the initial knowledge set; The dual-agent optimization module is used to generate an initial answer based on the initial knowledge set and iteratively optimize it to obtain the target retrieval result, based on a game-theoretic iterative optimization method. The knowledge storage module is used to store professional knowledge meta-models, manual coordinate information, knowledge atoms, and retrieval strategies; The interaction module is used to obtain user query information, display the target search results and feedback.

Images