Heterogeneous data management method based on rule engine and dynamic classification coding

Abstract

The invention discloses a heterogeneous data management method based on a rule engine and dynamic classification coding, and belongs to the technical field of information management and engineering informatization. According to the method, on the basis of an IEC 61355 classification table framework, multi-dimensional labels (such as project stages, unit numbers, systems, equipment and document types) of actual engineering documents and natural language processing semantic features are fused, and a dynamic classification system capable of being evolved regularly is constructed; a rule engine and a deep learning model are combined, a unique identifier conforming to an IEC structure is automatically generated, and classification rules are dynamically optimized through reinforcement learning; unified management of identifiers and intelligent recommendation based on user behaviors are realized among heterogeneous systems such as MRS, PLM, ERP, SCADA and the like. According to the method, the automation, accuracy and cross-system cooperation capability of document classification coding are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the fields of information management, enterprise content governance and engineering informatization technology, and specifically relates to a heterogeneous data governance method based on rule engine and dynamic classification coding. Background Technology

[0002] Static classification table-driven coding system: The current IEC 61355 provides a statically defined classification table, and the codes for each document type and its subclasses are fixed when the standard is published. Various engineering documents (such as design drawings, manuals, calculation sheets, test reports, etc.) are assigned corresponding classification codes in projects through manual selection or semi-automatic methods. In enterprise information systems, the common practice is to import the IEC 61355 classification table into a DMS or MRS system, relying on a predefined classification table, requiring manual matching of document types and coding labels.

[0003] Data fusion solutions driven by rule engines: Data routing, cleaning, or event fusion are achieved through rule bases, but they lack coding generation capabilities and do not integrate industrial classification standards.

[0004] The disadvantages of existing technology are: 1. Highly reliant on manual classification and coding processes: Currently, most solutions still rely on manual judgment of file attributes (title, author, date, etc.) and manual selection of codes for document classification. In large projects, the manual classification of hundreds of thousands of documents is costly and prone to errors. When documents are transferred across stages or systems, manual reconfirmation and adjustment of codes are required, increasing workload and error risk.

[0005] 2. Insufficient multi-dimensional information processing capabilities: The IEC 61355 static classification table only defines the document type dimension, but it lacks a direct correlation mechanism for key information such as project stage, unit number, system, and equipment. Currently, most companies typically add prefixes / suffixes to include custom project information, but this method lacks unified rules and is prone to conflicts during cross-system data exchange.

[0006] 3. Lack of content-driven intelligent classification: Existing systems primarily classify documents based on metadata (filename, author, date, etc.), lacking semantic analysis of document content and unable to automatically classify documents intelligently based on technical terms, component names, system numbers, etc. This makes the accuracy of document classification dependent on the user's professional knowledge and input quality, resulting in low automation. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide a heterogeneous data governance method based on rule engine and dynamic classification coding. This invention significantly improves the automation, accuracy and cross-system collaboration capabilities of document classification coding.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A heterogeneous data governance method based on rule engine and dynamic classification coding includes the following steps: S1. Based on the framework of the International Electrotechnical Commission IEC 61355 classification table, a dynamic classification system that can be evolved according to rules is constructed by analyzing the multi-dimensional tags and semantic features of natural language processing in actual engineering documents. S2. By combining a rule engine with a deep learning model, unique identifiers that conform to the IEC code structure are automatically generated, and classification rules are dynamically adjusted based on historical data analysis. S3. Unify the management of the unique identifier among heterogeneous systems such as Material Requirements System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA), and implement intelligent recommendations to optimize document management.

[0009] Preferably, in step S1, the construction of the dynamic classification system includes the following steps: S1-1. Extract multi-dimensional tags from actual engineering documents, including equipment manuals, design drawings, test reports, and other engineering documents. The multi-dimensional tags include project stage, unit number, system, equipment, and document type. S1-2. Perform entity recognition, keyword extraction, and theme modeling on the title, body text, header, footer, or label of the actual engineering document; S1-3. Use a bidirectional encoder to represent the transformation model to encode the semantics of the actual engineering document and generate a 768-dimensional semantic embedding vector. S1-4. The 768-dimensional semantic embedding vector and the multi-dimensional label are weighted and fused to construct a dynamic classification feature space.

[0010] Preferably, in step S2, the generation of the unique identifier includes the following steps: S2-1. Based on the dynamic classification feature space, predict the generation probability distribution of each coding segment using a bidirectional long short-term memory network model; S2-2. The decision path of the rule engine is optimized by using a reinforcement learning algorithm based on the Q-learning framework, and the rule priority is dynamically adjusted according to the historical classification accuracy. S2-3. Generate an encoding structure containing project code, stage code, unit code, system code, device code, file type code, sub-file type code, sequence code, and check bit; S2-4. Verify the integrity of the encoding structure using a cyclic redundancy check algorithm to ensure compliance with the IEC 61355 standard.

[0011] Preferably, in step S3, the heterogeneous system governance includes the following steps: S3-1. Establish a unified identifier mapping table between Material Requirements Management System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA). S3-2. Real-time analysis of user operation behavior data on actual engineering documents, wherein the user operation behavior data includes access frequency, modification history and associated documents; S3-3. Generate a personalized recommendation strategy based on the user operation behavior data to provide users with the best document classification suggestions; S3-4. Dynamically adjust the governance strategy according to the system load, and prioritize the processing of critical system documents when the load is high.

[0012] Preferably, in steps S1-3, the bidirectional encoder represents the input of the transformation model to encode the semantics of the actual engineering document as a text sequence, and the output is a 768-dimensional semantic embedding vector.

[0013] Preferably, in steps S1-4, the weighted fusion adopts an adaptive weight allocation mechanism, and the weight coefficients are dynamically calculated based on the correlation strength between the document type and the project stage. The calculation formula is as follows: ; in, For the first The weight coefficient of each label, The semantic distance between the tag and the document. The attenuation coefficient is... This represents the total number of tags.

[0014] Preferably, in step S2-1, the bidirectional long short-term memory network model takes the dynamic classification feature space as its input and outputs the conditional probability distribution of each coding segment.

[0015] Preferably, in step S2-2, the state space of the reinforcement learning algorithm based on the Q-learning framework includes historical classification accuracy, document type distribution, and system load; the action space includes rule priority adjustment strategies; and the reward function is defined as: ; in, For the first The reward value of each step To improve classification accuracy, For rule coverage, For system latency, , , These are the weighting coefficients.

[0016] Preferably, in step S3-2, the analysis of user behavior data related to actual engineering documents includes constructing a user-document interaction matrix and identifying high-frequency association patterns using singular value decomposition (SVD) technology. The matrix decomposition formula is as follows: ; in, For user-document interaction matrix, For the user latent factor matrix, It is a singular value matrix. This is the document's latent factor matrix.

[0017] Preferably, in step S3-3, the personalized recommendation strategy is generated based on association rule mining, and the recommendation results include a set of actual engineering documents with high probability of association. The formula for calculating the support of association rules is: ; in, and This is a collection of actual engineering documents. To contain simultaneously and The actual number of engineering documents For inclusion The actual number of engineering documents.

[0018] A heterogeneous data governance system based on a rule engine and dynamic classification coding, characterized in that it includes: Includes the following steps: The dynamic classification system construction module is used to construct a rule-based and evolving dynamic classification system based on the IEC 61355 classification table framework by analyzing the multi-dimensional tags and natural language processing semantic features of actual engineering documents. The unique identifier generation module, connected to the dynamic classification system construction module, is used to automatically generate unique identifiers that conform to the IEC code structure by combining a rule engine with a deep learning model, and to dynamically adjust the classification rules based on historical data analysis. The heterogeneous system governance module, connected to the unique identifier generation module, is used to uniformly govern the unique identifier among heterogeneous systems such as Material Requirements System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA), and to achieve intelligent recommendations to optimize document management. The rule engine optimization module, connected to the unique identifier generation module, is used to perform the reinforcement learning algorithm optimization based on the Q-learning framework.

[0019] Preferably, the dynamic classification system construction module includes a semantic feature extraction unit and a feature fusion unit, wherein the semantic feature extraction unit performs the deep vectorization processing of the bidirectional encoder representation transformation model.

[0020] Preferably, the unique identifier generation module includes an encoding prediction unit and a rule optimization unit, wherein the encoding prediction unit performs the prediction of the bidirectional long short-term memory network model.

[0021] Preferably, the heterogeneous system governance module includes a behavior analysis unit and a recommendation generation unit, wherein the behavior analysis unit performs the user's operation behavior data processing on the actual engineering documents.

[0022] The present invention can achieve the following beneficial effects: 1. Significantly improved automation of classification and coding: By leveraging rule engines and NLP content understanding, the system can automatically classify and locate documents / data and generate IEC structured codes, significantly reducing manual intervention. 2. Achieve semantic-level precise matching through NLP to improve classification accuracy and intelligence. NLP extracts entities / keywords / topics from titles, body text, headers, footers, and image tags, and integrates them with rule weights to make decisions, prioritizing the most specific and compliant classification path. Attached Figure Description

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0024] Preferred solutions include Figure 1 As shown, a heterogeneous data governance method based on rule engines and dynamic classification coding includes the following steps: S1. Based on the framework of the International Electrotechnical Commission IEC 61355 classification table, a dynamic classification system that can be evolved according to rules is constructed by analyzing the multi-dimensional tags and semantic features of natural language processing in actual engineering documents. S2. By combining a rule engine with a deep learning model, unique identifiers that conform to the IEC code structure are automatically generated, and classification rules are dynamically adjusted based on historical data analysis. S3. Unify the management of the unique identifier among heterogeneous systems such as Material Requirements System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA), and implement intelligent recommendations to optimize document management.

[0025] The overall architecture of this method aims to transform the static classification table of IEC 61355 into a dynamic classification system that can evolve according to rules. It integrates "multi-dimensional labels + NLP semantic features" to automatically generate unique identifiers (UIDs) that conform to the IEC code structure and to achieve unified governance across heterogeneous systems such as MRS / PLM / ERP / SCADA. The entire process starts from document input, goes through feature extraction, rule reasoning, encoding generation, and cross-system synchronization, forming a closed-loop data governance system.

[0026] Furthermore, in step S1, the construction of the dynamic classification system includes the following steps: S1-1. Extract multi-dimensional tags from actual engineering documents, including equipment manuals, design drawings, test reports, and other engineering documents. The multi-dimensional tags include project stage, unit number, system, equipment, and document type. S1-2. Perform entity recognition, keyword extraction, and theme modeling on the title, body text, header, footer, or label of the actual engineering document; S1-3. Use a bidirectional encoder to represent the transformation model to encode the semantics of the actual engineering document and generate a 768-dimensional semantic embedding vector. S1-4. The 768-dimensional semantic embedding vector and the multi-dimensional label are weighted and fused to construct a dynamic classification feature space.

[0027] In this process, the system first performs multi-dimensional label extraction and standardization: Filenames / paths are parsed using regular expressions and a predefined dictionary. For example, "W74 wind turbine foundation pile installation diagram.pdf" can be parsed to have Unit=W74, System=pile foundation, and SubType=installation diagram. Simultaneously, metadata fields from source systems such as MRS / PLM are mapped, and synonyms and value ranges are aligned. Further cross-validation with project master data, such as system / equipment ledgers, is performed to correct spelling errors and aliases. Each label is accompanied by a confidence level (conf_label) for subsequent fusion and weighting. Then, NLP semantic feature extraction is performed: technical entities and keyword phrases are identified, such as "foundation pile," "torque," "epoxy grouting," and "220kV GIS." A fine-tuned Transformer model is used to classify document types, with input including title, abstract, header / footer key fields, and first and last paragraphs. System-equipment relationships are extracted based on rules and denotation techniques. For example, if both "grounding transformer" and "GIS" appear in a document, it is inferred that it belongs to the GIS equipment under the electrical system. Finally, a constrained breadth-first search (BFS) is performed on the LSM classification tree to generate candidate classification paths and filter out invalid nodes that do not meet business rules, such as stage mismatch or system mismatch.

[0028] Further, in step S2, the generation of the unique identifier includes the following steps: S2-1. Based on the dynamic classification feature space, predict the generation probability distribution of each coding segment using a bidirectional long short-term memory network model; S2-2. The decision path of the rule engine is optimized by using a reinforcement learning algorithm based on the Q-learning framework, and the rule priority is dynamically adjusted according to the historical classification accuracy. S2-3. Generate an encoding structure containing project code, stage code, unit code, system code, device code, file type code, sub-file type code, sequence code, and check bit; S2-4. Verify the integrity of the encoding structure using a cyclic redundancy check algorithm to ensure compliance with the IEC 61355 standard. The unique identifier (UID) coding structure must possess foresight, stability, flexibility, and applicability, ensuring a high degree of alignment with the company's business rules. Specific coding segments are defined as follows: Project Code (Proj), Phase Code (divided into five phases according to ISO 21500: initiation, planning, execution, monitoring, and closure), Unit Code (Unit, which is the turbine number in the wind power field), System Code, Equipment Code (Equip, all determined based on the project's equipment and facilities list), Document Type Code (Dcc), and Sub-Document Type Code (SubType, strictly adhering to the IEC 61355 standard). During the generation process, the system first maps the final determined classification path to standardized labels. <system> / <equip> / <dcc> / <subtype>The sequence is divided into segments; then, sequence numbers are assigned according to the dimensions (Proj, Phase, System, Equip, DCC, SubType), and an optimistic locking mechanism is used to ensure auto-incrementing uniqueness; then, all preceding encoded segments are concatenated and the Mod11 or Luhn check bit is calculated; finally, a uniqueness check is performed. If a duplicate (uid_code, hash_md5) combination is found, a new version (version+1) is automatically created instead of overwriting the original record.

[0029] Furthermore, in step S3, the heterogeneous system governance includes the following steps: S3-1. Establish a unified identifier mapping table between Material Requirements Management System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA). S3-2. Real-time analysis of user operation behavior data on actual engineering documents, wherein the user operation behavior data includes access frequency, modification history and associated documents; S3-3. Generate a personalized recommendation strategy based on the user operation behavior data to provide users with the best document classification suggestions; S3-4. Dynamically adjust the governance strategy according to the system load, and prioritize the processing of critical system documents when the load is high.

[0030] To support the above governance process, the system needs to build a core data model. This mainly includes: The DOC (Document Master Table) contains fields such as doc_uid, title, hash_md5, source_system, project_code, stage, unit_no, system_code, equipment_code, doc_kind_dcc (IEC document type code), doc_subtype, LSM_path, uid_code, version, status, created_at, and audit_trail (JSON). The RULE (Rule Table) defines rule_id, scope (applicable project / stage / system), priority, condition (DSL / JSON), and action (code segment generation). The system includes: a `<constraint>` table for data entry and processing, containing data for each data type (e.g., `<constraint>`, `<from>`, `<to>`, `<version>`, `<author>`); a `<txonomy>` (LSM tree) table storing `node_id`, `parent_id`, `name`, `node_type` (Project / System / Equipment / DocKind / Subtype), `iec_code`, `aliases`, `constraints`, and `status`); and a `<feature_store>` table storing intermediate results such as `doc_uid`, `tokens`, `entities`, `keyphrases`, `embedding`, and `struct_features` (tables / header levels / footers). Through the collaborative work of these tables, the system can establish a reliable identifier mapping between heterogeneous systems and support subsequent intelligent recommendation and load scheduling.

[0031] Furthermore, in steps S1-3, the bidirectional encoder represents the input of the transformation model to encode the semantics of the actual engineering document as a text sequence, and the output is a 768-dimensional semantic embedding vector.

[0032] This 768-dimensional semantic embedding vector, as a deep semantic representation of the document, is stored in the FEATURE_STORE table for subsequent similarity calculation, cluster analysis, and classification decisions. It serves as a key bridge connecting the original document content with the structured tags.

[0033] Furthermore, in steps S1-4, the weighted fusion adopts an adaptive weight allocation mechanism, and the weight coefficients are dynamically calculated based on the correlation strength between the document type and the project stage. The calculation formula is as follows: ; in, For the first The weight coefficient of each label, The semantic distance between the tag and the document. The attenuation coefficient is... This represents the total number of tags.

[0034] This adaptive weighting mechanism ensures that the importance of relevant tags can be dynamically adjusted at different project stages. For example, in the "execution" stage, the weight of "equipment code" and "installation diagram" types will be significantly higher than in the "planning" stage, thereby guiding the classification system to better fit the current business context.

[0035] Furthermore, in step S2-1, the bidirectional long short-term memory network model takes the dynamic classification feature space as its input and outputs the conditional probability distribution of each coding segment.

[0036] The role of the BiLSTM model here is to transform the fused feature vector (containing structured labels and unstructured semantics) into a conditional probability distribution for each encoded segment (such as Dcc and SubType), providing a probabilistic basis for subsequent deterministic encoding and effectively handling ambiguous or conflicting input information.

[0037] Further, in step S2-2, the state space of the reinforcement learning algorithm based on the Q-learning framework includes historical classification accuracy, document type distribution, and system load; the action space includes rule priority adjustment strategy; and the reward function is defined as: ; in, For the first The reward value of each step To improve classification accuracy, For rule coverage, For system latency, , , These are the weighting coefficients.

[0038] This reinforcement learning mechanism enables the rule engine to self-optimize. The priority field in the RULE table is dynamically updated based on the output of Q-learning, giving higher priority to rules with high accuracy and coverage, while degrading inefficient or outdated rules, thereby continuously improving the overall classification performance.

[0039] Furthermore, in step S3-2, the analysis of user behavior data related to actual engineering documents includes constructing a user-document interaction matrix and identifying high-frequency association patterns using singular value decomposition (SVD) technology. The matrix decomposition formula is as follows: ; in, For user-document interaction matrix, For the user latent factor matrix, It is a singular value matrix. This is the document's latent factor matrix. The user-document interaction matrix data comes from access logs and operation records in the DOC table. Through SVD decomposition, the system can identify potential user interest groups and document topic clusters, laying the foundation for accurate recommendations.

[0040] Furthermore, in step S3-3, the personalized recommendation strategy is generated based on association rule mining, and the recommendation results include a set of actual engineering documents with high probability of association. The formula for calculating the support of association rules is: ; in, and This is a collection of actual engineering documents. To contain simultaneously and The actual number of engineering documents For inclusion The actual number of engineering documents.

[0041] The results of association rule mining can be directly used to generate recommendation units. For example, when a user views a "wind turbine foundation construction drawing", the system can automatically recommend related documents with high support, such as "pile foundation test report" or "concrete mix design", thereby optimizing the user's document search experience.

[0042] A heterogeneous data governance system based on a rule engine and dynamic classification coding includes the following steps: The dynamic classification system construction module is used to construct a rule-based and evolving dynamic classification system based on the IEC 61355 classification table framework by analyzing the multi-dimensional tags and natural language processing semantic features of actual engineering documents. The unique identifier generation module, connected to the dynamic classification system construction module, is used to automatically generate unique identifiers that conform to the IEC code structure by combining a rule engine with a deep learning model, and to dynamically adjust the classification rules based on historical data analysis. The heterogeneous system governance module, connected to the unique identifier generation module, is used to uniformly govern the unique identifier among heterogeneous systems such as Material Requirements System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA), and to achieve intelligent recommendations to optimize document management. The rule engine optimization module, connected to the unique identifier generation module, is used to perform the reinforcement learning algorithm optimization based on the Q-learning framework.

[0043] This system architecture corresponds entirely to the aforementioned method, with each module achieving data interoperability through a shared core data table (DOC, RULE, TAXONOMY, FEATURE_STORE). The dynamic classification system construction module is responsible for populating FEATURE_STORE and the initial LSM_path; the unique identifier generation module reads this data and writes it to DOC.uid_code; the heterogeneous system governance module uses the DOC.source_system field to interface with MRS / PLM / ERP / SCADA; and the rule engine optimization module continuously updates the priority field in the RULE table, forming a closed-loop optimization.

[0044] Furthermore, the dynamic classification system construction module includes a semantic feature extraction unit and a feature fusion unit, wherein the semantic feature extraction unit performs the deep vectorization processing of the bidirectional encoder representation transformation model. The output of the semantic feature extraction unit is the FEATURE_STORE.embedding field, while the feature fusion unit uses an adaptive weighting mechanism to fuse the embedding vector with the structured features (such as project_code, system_code, etc.) from the multidimensional label extraction unit to generate the final dynamic classification feature space.

[0045] Furthermore, the unique identifier generation module includes an encoding prediction unit and a rule optimization unit, wherein the encoding prediction unit performs the prediction of the bidirectional long short-term memory network model.

[0046] The encoding prediction unit is responsible for executing step S2-1. Its BiLSTM model takes the dynamic classification feature space as input and outputs the probability distribution of each encoding segment. The rule optimization unit is responsible for executing step S2-2, which continuously optimizes the rule priority in the RULE table through reinforcement learning algorithm.

[0047] Furthermore, the heterogeneous system governance module includes a behavior analysis unit and a recommendation generation unit, wherein the behavior analysis unit performs the user's operation behavior data processing on the actual engineering documents.

[0048] The behavior analysis unit is responsible for collecting and processing user operation logs, constructing a user-document interaction matrix, and performing SVD decomposition. The recommendation generation unit, based on the decomposition results and association rule mining, provides users with personalized document classification suggestions and related document recommendations, and finally feeds the optimization results back into the document management process.

[0049] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.< / subtype> < / dcc> < / equip> < / system>

Claims

1. A heterogeneous data governance method based on rule engine and dynamic classification coding, characterized in that, Includes the following steps: S1. Based on the IEC 61355 classification table framework, a dynamic classification system that can be evolved according to rules is constructed by analyzing the multi-dimensional tags and semantic features of natural language processing in actual engineering documents. S2. By combining a rule engine with a deep learning model, unique identifiers that conform to the IEC code structure are automatically generated, and classification rules are dynamically adjusted based on historical data analysis. S3. Unify the management of the unique identifier among heterogeneous systems such as Material Requirements System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA), and implement intelligent recommendations to optimize document management.

2. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 1, characterized in that, In step S1, the construction of the dynamic classification system includes the following steps: S1-1. Extract multi-dimensional tags from actual engineering documents, including equipment manuals, design drawings, test reports, and other engineering documents. The multi-dimensional tags include project stage, unit number, system, equipment, and document type. S1-2. Perform entity recognition, keyword extraction, and theme modeling on the title, body text, header, footer, or label of the actual engineering document; S1-3. Use a bidirectional encoder to represent the transformation model to encode the semantics of the actual engineering document and generate a 768-dimensional semantic embedding vector. S1-4. The 768-dimensional semantic embedding vector and the multi-dimensional label are weighted and fused to construct a dynamic classification feature space.

3. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 1, characterized in that, In step S2, the generation of the unique identifier includes the following steps: S2-1. Based on the dynamic classification feature space, predict the generation probability distribution of each coding segment using a bidirectional long short-term memory network model; S2-2. The decision path of the rule engine is optimized by using a reinforcement learning algorithm based on the Q-learning framework, and the rule priority is dynamically adjusted according to the historical classification accuracy. S2-3. Generate an encoding structure containing project code, stage code, unit code, system code, device code, file type code, sub-file type code, sequence code, and check bit; S2-4. Verify the integrity of the encoding structure using a cyclic redundancy check algorithm to ensure compliance with the IEC 61355 standard.

4. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 1, characterized in that, In step S3, the heterogeneous system governance includes the following steps: S3-1. Establish a unified identifier mapping table between Material Requirements Management System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA). S3-2. Real-time analysis of user operation behavior data on actual engineering documents, wherein the user operation behavior data includes access frequency, modification history and associated documents; S3-3. Generate a personalized recommendation strategy based on the user operation behavior data to provide users with the best document classification suggestions; S3-4. Dynamically adjust the governance strategy according to the system load, and prioritize the processing of critical system documents when the load is high.

5. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 2, characterized in that, In steps S1-3, the bidirectional encoder represents the input of the transformation model used to encode the semantics of the actual engineering document as a text sequence, and the output is a 768-dimensional semantic embedding vector.

6. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 2, characterized in that, In steps S1-4, the weighted fusion adopts an adaptive weight allocation mechanism. The weight coefficients are dynamically calculated based on the correlation strength between the document type and the project stage. The calculation formula is as follows: ； in, For the first The weight coefficient of each label, The semantic distance between the tag and the document. The attenuation coefficient is... This represents the total number of tags.

7. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 3, characterized in that, In step S2-1, the bidirectional long short-term memory network model takes the dynamic classification feature space as its input and outputs the conditional probability distribution of each coding segment.

8. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 3, characterized in that, In step S2-2, the state space of the reinforcement learning algorithm based on the Q-learning framework includes historical classification accuracy, document type distribution, and system load; the action space includes rule priority adjustment strategy; and the reward function is defined as: ； in, For the first The reward value of each step To improve classification accuracy, For rule coverage, For system latency, , , These are the weighting coefficients.

9. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 4, characterized in that, In step S3-2, the analysis of user behavior data related to actual engineering documents includes constructing a user-document interaction matrix and identifying high-frequency association patterns using singular value decomposition (SVD). The matrix decomposition formula is as follows: ； in, For user-document interaction matrix, For the user latent factor matrix, It is a singular value matrix. This is the document's latent factor matrix.

10. The heterogeneous data governance method based on rule engine and dynamic classification coding according to claim 4, characterized in that, In step S3-3, the personalized recommendation strategy is generated based on association rule mining. The recommendation results include a set of real-world engineering documents with high probability of association. The formula for calculating the support of association rules is: ； in, and This is a collection of actual engineering documents. To contain simultaneously and The actual number of engineering documents For inclusion The actual number of engineering documents.

11. A heterogeneous data governance system based on a rule engine and dynamic classification coding, characterized in that, The heterogeneous data governance method based on rule engine and dynamic classification coding according to any one of claims 1-10 is characterized by comprising the following steps: The dynamic classification system construction module is used to construct a rule-based and evolving dynamic classification system based on the IEC 61355 classification table framework by analyzing the multi-dimensional tags and natural language processing semantic features of actual engineering documents. The unique identifier generation module, connected to the dynamic classification system construction module, is used to automatically generate unique identifiers that conform to the IEC code structure by combining a rule engine with a deep learning model, and to dynamically adjust the classification rules based on historical data analysis. The heterogeneous system governance module, connected to the unique identifier generation module, is used to uniformly govern the unique identifier among heterogeneous systems such as Material Requirements System (MRS), Product Lifecycle Management System (PLM), Enterprise Resource Planning System (ERP), or Supervisory Control System (SCADA), and to achieve intelligent recommendations to optimize document management. The rule engine optimization module, connected to the unique identifier generation module, is used to perform the reinforcement learning algorithm optimization based on the Q-learning framework.

12. The heterogeneous data governance system based on rule engine and dynamic classification coding according to claim 11, characterized in that, The dynamic classification system construction module includes a semantic feature extraction unit and a feature fusion unit, wherein the semantic feature extraction unit performs the deep vectorization processing of the bidirectional encoder representation transformation model.

13. The heterogeneous data governance system based on rule engine and dynamic classification coding according to claim 11, characterized in that, The unique identifier generation module includes an encoding prediction unit and a rule optimization unit, wherein the encoding prediction unit performs the prediction of the bidirectional long short-term memory network model.

14. The heterogeneous data governance system based on rule engine and dynamic classification coding according to claim 11, characterized in that, The heterogeneous system governance module includes a behavior analysis unit and a recommendation generation unit, wherein the behavior analysis unit performs the processing of user operation behavior data on actual engineering documents.

Images