Carbon border adjustment mechanism intelligent auxiliary customs declaration method and system based on natural language processing

Abstract

The invention relates to the field of intelligent information processing, in particular to a carbon border adjustment mechanism intelligent auxiliary customs declaration method and system based on natural language processing, and the method comprises the steps: carrying out the real-time translation and key information extraction of CBAM related rule files and unstructured documents; a dynamic CBAM knowledge graph is constructed based on related rule texts, rule updating is monitored in real time through a text classification and event extraction technology, and calculation rules and declaration logic are dynamically adjusted; performing standardization processing on supplier data in different formats, integrating an industry emission factor library, automatically matching a calculation model according to a product type, generating a carbon emission result meeting a CBAM requirement, and mapping the data to a corresponding position of a CBAM declaration form; based on historical declaration data and related rule texts, analysis and compliance risk prediction are carried out by using a natural language processing technology, so that the declaration data are ensured to accord with latest rules, and violation risks caused by rule changes are avoided.

Description

Technical Field

[0001] This application relates to the field of intelligent information processing, specifically to an intelligent auxiliary customs declaration method and system based on a carbon border adjustment mechanism using natural language processing. Background Technology

[0002] The Carbon Border Adjustment Mechanism (CBAM) aims to promote the compliance of imported goods with specific carbon emission standards by imposing carbon tariffs. Against this backdrop, businesses face increasingly complex CBAM customs clearance processes, requiring them to handle large amounts of carbon emission data and related documentation.

[0003] From a technological perspective, existing solutions mainly fall into two categories: one is general-purpose carbon management software, which can record and manage enterprise energy consumption data or calculate the carbon footprint of enterprise products, but it is not deeply adapted to CBAM rules, lacks a CBAM knowledge base, cannot automatically match product HS codes with carbon emission calculation rules, and does not have the ability to parse unstructured documents; the other is the reporting and uploading model. Although relevant agencies provide standardized electronic form uploading functions, they lack intelligent data parsing support. Enterprises need to collect data from suppliers and themselves independently, and perform calculations and filling in according to complex reporting rules. This method is difficult to automatically track continuously updated rule content.

[0004] Based on existing products and technical solutions, the following technical constraints currently exist in the field of CBAM compliance management: First, traditional carbon management systems struggle to automatically extract key information; second, dynamic compliance risks are significant, as CBAM rules are constantly evolving, and existing technologies rely on manual tracking of rule changes, resulting in severe lags in response; third, data integration efficiency is low, with carbon data from enterprises and their supply chains scattered across various unstructured documents, and existing data entry methods are prone to introducing data quality issues, seriously affecting the efficiency and accuracy of reporting. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an intelligent assisted customs declaration method and system based on a carbon border adjustment mechanism using natural language processing.

[0006] In a first aspect, the technical solution of the present invention provides an intelligent assisted customs declaration method based on a carbon border adjustment mechanism using natural language processing, comprising the following steps: S1. Real-time translation and key information extraction of CBAM-related rule files and unstructured documents are performed through a multilingual intelligent parsing engine. Key information includes product codes and carbon emission data. S2. Construct a dynamic CBAM knowledge graph based on relevant rule texts. The knowledge graph includes product classification, emission factors and calculation method entities and their relationships. Real-time monitoring of rule updates is conducted through text classification and event extraction technology to dynamically adjust calculation rules and declaration logic. S3. Through intelligent data integration and computing engine, it collaboratively collects data from the enterprise itself and its supply chain, standardizes supplier data in different formats, integrates industry emission factor library, automatically matches calculation model according to product type, generates carbon emission results that meet CBAM requirements, and maps the data to the corresponding position in the CBAM declaration form. S4. Based on historical declaration data and relevant rule texts, use natural language processing technology to analyze and predict compliance risks to ensure that the declaration data complies with the latest rules.

[0007] Utilizing natural language processing technology to analyze historical application data and related rule texts, and to predict compliance risks, ensures that application data complies with the latest rules. This helps improve the accuracy and reliability of application data, reduces the review risk caused by non-compliant application data, and increases the success rate of applications.

[0008] As a further limitation of the technical solution of the present invention, step S1 includes: S11. Obtain CBAM rule files and unstructured documents, the document formats including PDF, scanned images, email attachments or Word documents; for image documents, use optical character recognition technology to perform text recognition and extraction, and convert the documents into a machine-readable text format; S12. A multilingual pre-trained model based on the Transformer architecture is used to translate non-Chinese text into Chinese in real time, while preserving the correspondence between the original text format and key terms. S13. Based on a pre-trained named entity recognition model, sequence labeling and parsing are performed on the translated text to identify and extract key information entities. The key information entities include at least: product name, product HS code, carbon emission factor, calculation rule clauses, geographical boundary definition, and reporting obligation description. S14. Perform structured processing on the extracted key information entities and output structured data.

[0009] By utilizing optical character recognition technology and a multilingual pre-trained model based on the Transformer architecture, image-based documents can be converted into machine-readable text formats, and non-Chinese text can be translated in real time. This further enhances the system's ability to process documents in different languages ​​and formats, enabling enterprises to process CBAM-related documents from various sources more efficiently.

[0010] Based on a pre-trained named entity recognition model, the translated text undergoes sequence labeling and parsing, accurately identifying and extracting key information entities such as product names, HS codes, and carbon emission factors. This provides an accurate and reliable data foundation for subsequent data processing and declaration, helping to improve the accuracy and completeness of the declaration data. The extracted key information entities are then structured to output structured data, facilitating further analysis and application by the system. This improves data usability and operability, providing strong support for realizing an intelligent CBAM customs declaration process.

[0011] As a further limitation of the technical solution of the present invention, step S2 includes: S21. Based on relevant rule texts, define the core ontology structure of the knowledge graph. The ontology structure includes entity types, as well as the applicability, inclusion, and computational dependencies between entity types. Entity types include product categories, emission factors, and calculation methods. S22. Using natural language processing technology, extract entities and their relationship pairs in batches from relevant rule texts to form a dynamic CBAM knowledge graph; S23. Receive the latest relevant rule text, filter the acquired text using a text classification model, and identify valid documents related to CBAM; then use event extraction technology to identify rule change events from valid documents and extract event elements, including event type, affected entities, new and old rule content, and effective date. S24. Based on the extracted rule change events, perform add, delete, and modify operations on the corresponding entities and relationships in the knowledge graph; the update of the knowledge graph triggers the system to dynamically adjust the calculation rules and reporting logic, including: updating the emission factor library, correcting the calculation method formula, expanding or modifying the product classification list, and changing the data filling requirements.

[0012] By leveraging text classification models and event extraction technology, valid documents and rule change events related to CBAM can be quickly identified, and event elements can be extracted. This allows for timely detection of rule updates and provides accurate data for subsequent knowledge graph updates and adjustments to calculation rules and reporting logic, further improving the system's response speed and adaptability to rule updates. Based on rule change events, corresponding entities and relationships in the knowledge graph can be added, deleted, or modified, enabling dynamic adjustment and updates of the knowledge graph. This ensures that the knowledge graph remains consistent with the latest CBAM rules, providing accurate knowledge support for enterprise compliance reporting and reducing compliance risks caused by rule changes.

[0013] As a further limitation of the technical solution of the present invention, step S22 includes: S221. Clean, segment, and sentence-level divide the acquired relevant rule texts; construct training corpus using back-translation and term substitution techniques to enhance the model's ability to recognize CBAM domain terms; S222. A joint learning framework based on a pre-trained language model is adopted, wherein the joint learning framework shares a text encoding layer and performs named entity recognition and relation extraction subtasks in parallel; the model is fine-tuned using training corpus. S223. Input the relevant rule text to be processed into the fine-tuned model. The model synchronously outputs the identified entities and their relationships, forming a knowledge triple with a head entity-relationship-tail entity structure. S224. Perform entity alignment and cleaning on the knowledge triples extracted from different document sources; import the cleaned triples into the graph database to form a dynamic CBAM knowledge graph.

[0014] By employing back-translation and terminology substitution techniques to construct the training corpus, the model's ability to recognize CBAM domain terms is enhanced, improving its accuracy and reliability in processing CBAM-related text. This facilitates more precise extraction of entities and their relation pairs, further improving the quality and completeness of the knowledge graph. A joint learning framework based on a pre-trained language model, sharing a text encoding layer, and executing named entity recognition and relation extraction subtasks in parallel, along with fine-tuning the model using the training corpus, improves the efficiency and accuracy of knowledge extraction. It quickly and accurately extracts valuable knowledge information from large amounts of rule-based text, forming knowledge triples, providing strong support for constructing a dynamic CBAM knowledge graph.

[0015] The knowledge triples extracted from different document sources are entity aligned and cleaned, and the cleaned triples are imported into a graph database to form a dynamic CBAM knowledge graph. This process can effectively remove duplicate, erroneous or inconsistent knowledge information, ensuring the quality and consistency of the knowledge graph.

[0016] As a further limitation of the technical solution of the present invention, step S24 includes: S241. Analyze the extracted rule change events and automatically assess the scope and level of the impact of the change on the existing knowledge graph structure and downstream computing and reporting logic based on the event type, affected entities, and effective date. S242. Based on the impact assessment results, automatically generate a set of operation instructions for adding, deleting, and modifying the knowledge graph; the operation instruction set includes: Create new entities or relationships; Deprecate or archive entities or relationships that have become invalid; Modify the attributes of an existing entity or relationship; S243. At the set time point, execute the operation instruction set to perform an atomic update operation on the knowledge graph; at the same time, perform version management on the knowledge graph, and record the version number, change content, basis and effective time of each change; S244. The update of the knowledge graph serves as a triggering event, automatically initiating the system to dynamically adjust the calculation rules and reporting logic synchronously, including: Based on the latest version of the knowledge graph, the emission factor library, calculation model formula, and product type matching rules within the intelligent data integration and computing engine module have been adjusted synchronously. The logic verification rule base and risk prediction model of the risk warning and compliance inspection module are updated based on the latest version of the knowledge graph. The entity library in the named entity recognition model of the multilingual intelligent parsing engine has been updated based on the latest version of the knowledge graph.

[0017] It helps enterprises accurately understand the specific impact of rule changes on their compliance reporting, enabling them to prepare in advance and reduce the risks associated with rule changes. It automates the updating and management of the knowledge graph, improving its efficiency and accuracy, ensuring the knowledge graph is always up-to-date, and providing timely and accurate knowledge support to enterprises. It ensures consistency and collaboration between all system modules, improving the overall system performance and adaptability, enabling enterprises to respond to CBAM rule changes more efficiently and accurately.

[0018] As a further limitation of the technical solution of the present invention, step S3 includes: S31. By distributing standardized data collection templates to supply chain suppliers through the system, the system collaboratively collects data from the enterprise itself and its supply chain; this improves the efficiency and accuracy of data collection, ensures the consistency and integrity of the data, and provides a more reliable data foundation for enterprises to conduct carbon emission accounting and reporting.

[0019] S32. Perform parsing operations on the collected multi-format raw data; for unstructured documents, call the multi-language intelligent parsing engine to extract key numerical and textual information; convert heterogeneous data into standardized data objects; improve the efficiency and accuracy of data processing, reduce the workload and error rate of data processing, speed up data processing, and save time for subsequent carbon emission accounting and reporting work.

[0020] S33. Based on the product information in the standardized data object, the system calls the dynamic CBAM knowledge graph to automatically determine the CBAM category to which the product belongs and the corresponding accounting rules; it matches the precise emission factors from the integrated industry emission factor library according to the product type and loads the corresponding calculation method model; it executes the calculation model to generate carbon emission results that meet CBAM requirements and include direct emissions, indirect emissions, and implicit emissions; it realizes the automation of carbon emission accounting, improves the efficiency and accuracy of accounting, reduces errors, ensures the reliability of carbon emission results, and provides accurate declaration data support for enterprises.

[0021] S34. Based on the mapping rules defined in the dynamic CBAM knowledge graph, automatically fill the generated carbon emission results, product information, and supplier information into the corresponding fields of the CBAM declaration form; generate a complete pre-filled report form that can be reviewed and submitted.

[0022] The process of filling out the application form has been simplified, the application efficiency has been improved, and companies can complete the CBAM application process more quickly.

[0023] As a further limitation of the technical solution of the present invention, step S33 includes: S331. The product description information in the standardized data object is semantically matched with the product classification entity in the dynamic CBAM knowledge graph. If the match is successful, the CBAM classification code to which the product belongs and its corresponding unique accounting rule path are automatically bound. If the match fails, a manual intervention process is initiated for confirmation and mapping. This improves the accuracy of product classification, ensures the accuracy of carbon emission accounting, and provides accurate declaration data support for enterprises.

[0024] S332. Based on the bound product CBAM classification, query and match the most accurate default emission factor or specific facility-level emission factor from the integrated industry emission factor library; the emission factor library is associated with emission factor entities in the dynamic CBAM knowledge graph; this further improves the accuracy of carbon emission accounting, ensures the reliability of carbon emission results, and provides accurate reporting data support for enterprises.

[0025] S333. Load the corresponding calculation method model according to the bound accounting rule path; the calculation method model is a predefined calculation function or algorithm module that conforms to the CBAM accounting rules; inject the activity data in the standardized data object and the matched emission factors as input parameters into the calculation model; S334. Execute the calculation model to calculate the direct, indirect, and implicit emissions of the product; synthesize and summarize the emission results according to CBAM requirements to generate the final total carbon emission intensity and total emission data of the product. This achieves automation and refinement of carbon emission accounting, improves the efficiency and accuracy of accounting, reduces accounting errors, and provides accurate reporting data support for enterprises.

[0026] As a further limitation of the technical solution of the present invention, step S4 includes: S41. Continuously collect CBAM-related rule texts and adjudication cases as external corpus; at the same time, use the system's historical application data and its corresponding review status as internal corpus; together, they form a corpus for training risk prediction models; this can help enterprises identify potential compliance risks in advance, reduce compliance risks caused by misunderstandings of rules or data errors, and improve the success rate of applications.

[0027] S42. Perform deep analysis on the corpus using natural language processing (NLP) techniques to extract text features; the NLP techniques include: The key industries and risk issues identified by the rules are based on thematic models; Based on sequence labeling models, common data error patterns and logical conflict rules that lead to application failures are learned from case studies. Based on the extracted features, a risk prediction model for CBAM compliance risk prediction is trained, which further improves the accuracy of risk prediction, provides enterprises with more accurate compliance risk warnings, and enables enterprises to take measures in advance to prevent and rectify risks and reduce compliance risks.

[0028] S43. After the user completes the filling or mapping of the current declaration data, the risk prediction model is invoked to perform the following operations: By comparing the current declaration data with the features of the rule text, risks caused by misunderstandings of the rules can be predicted; Match current data with learned historical error patterns to identify potential data anomalies and logical contradictions; Output the predicted risk points and risk levels; S44. Perform one-click global logic verification; based on all currently effective calculation and declaration logic rules in the dynamic CBAM knowledge graph, automatically verify all data fields in the declaration form and their inherent consistency relationships; for logical errors or inconsistencies detected, automatically generate specific correction suggestions and highlight them. Automatically generating and highlighting specific correction suggestions can promptly identify and correct logical errors or inconsistencies in the declaration data, improving the quality and accuracy of the declaration data, ensuring that the declaration data complies with the latest rules, and increasing the success rate of the declaration.

[0029] As a further limitation of the technical solution of the present invention, step S44 includes: S441. Extract all currently effective calculation rules, declaration logic rules, and constraint relationships between data fields from the dynamic CBAM knowledge graph; formalize the rules into machine-executable verification logic, which includes mathematical relationship verification, logical consistency verification, and scope compliance verification. S442. In response to the user's one-click verification command, perform a global scan of all data fields in the CBAM declaration form; match the value, type, and context information of each field with the verification logic to trigger all applicable verification rules; S443. Execute all triggered validation rules in parallel and calculate the degree of conformity between each data field and the expected value of the rule; when data is detected as not meeting the validation logic, mark it as a logical error point; S444. Summarize all verification results and generate a structured verification report; for each identified logical error, automatically associate it with the source rule clause in the knowledge graph and generate correction suggestions; highlight the error data, associated fields, and correction suggestions on the user interface. Highlighting error data, associated fields, and correction suggestions on the user interface enables quick location of error data and provides users with clear correction directions and suggestions, helping companies quickly correct errors and improve application efficiency and success rate.

[0030] Secondly, the present invention also provides an intelligent assisted customs declaration system based on a carbon border adjustment mechanism using natural language processing, comprising: The multilingual intelligent parsing engine module is used to translate CBAM-related rule files and unstructured documents in real time and extract key information, including product codes and carbon emission data. The dynamic CBAM knowledge graph module is used to construct a knowledge graph based on relevant rule texts. The knowledge graph includes product classification, emission factors and calculation methods, entities and their relationships. It also monitors rule updates in real time through text classification and event extraction technology, and dynamically adjusts calculation rules and declaration logic. The intelligent data integration and computing engine module is used to collaboratively collect data from the enterprise itself and its supply chain, standardize supplier data in different formats, integrate industry emission factor libraries, automatically match calculation models according to product type, generate carbon emission results that meet CBAM requirements, and map the data to the corresponding positions on the CBAM declaration form. The risk warning and compliance check module is used to analyze and predict compliance risks based on historical declaration data and relevant rule texts using natural language processing technology, ensuring that the declaration data complies with the latest rules.

[0031] As can be seen from the above technical solutions, this application has the following advantages: Through a multilingual intelligent parsing engine, it can translate CBAM-related rule files and unstructured documents in real time and extract key information. It supports multiple languages ​​and multiple document formats, effectively solving the problems of language barriers and inconsistent document formats faced by enterprises, enabling enterprises to obtain and process CBAM-related information more conveniently.

[0032] The intelligent data integration and computing engine can collaboratively collect data from the enterprise itself and its supply chain, standardize supplier data in different formats, integrate industry emission factor libraries, and automatically match calculation models to generate carbon emission results. This greatly improves the efficiency and accuracy of data integration, reduces the time and cost of data processing, and mitigates the impact of data quality issues on the efficiency and accuracy of declarations.

[0033] The dynamic CBAM knowledge graph, built based on relevant rule texts, can monitor rule updates in real time through text classification and event extraction technologies, and dynamically adjust calculation rules and declaration logic. This enables enterprises to respond promptly to changes in CBAM rules, maintain the compliance of their declarations, effectively address dynamic compliance risks, and avoid violations caused by rule changes. Attached Figure Description

[0034] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating the method provided in an embodiment of the present invention.

[0036] Figure 2 A block diagram of a system provided in an embodiment of the present invention. Detailed Implementation

[0037] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0039] like Figure 1 As shown, this embodiment of the invention provides an intelligent assisted customs declaration method based on a carbon border adjustment mechanism using natural language processing, comprising the following steps: S1. Real-time translation and key information extraction of CBAM-related rule files and unstructured documents are performed through a multilingual intelligent parsing engine. Key information includes product codes and carbon emission data. The system first receives CBAM-related rule files and unstructured documents uploaded or automatically captured by enterprises. The document formats include PDF, scanned copies, Word documents, email attachments, and image files.

[0040] Multilingual pre-trained models based on the Transformer architecture (such as MBERT and XLM-R) are used to translate non-Chinese rule texts (such as English and French rules) in real time, while preserving the correspondence between the original terminology and the translation (such as establishing a terminology lookup table for "HS Code - Harmonized System Code"). Optical Character Recognition (OCR) technology (such as integrating the Tesseract OCR engine) is used to recognize text in scanned documents, images, and other image-based documents, converting them into machine-readable text. Then, a pre-trained Named Entity Recognition (NER) model is used to automatically extract key information such as product codes (such as HS codes), carbon emission data (such as emissions per unit of product), and calculation rule clauses from the translated text.

[0041] S2. Construct a dynamic CBAM knowledge graph based on relevant rule texts. The knowledge graph includes product classification, emission factors and calculation method entities and their relationships. Real-time monitoring of rule updates is conducted through text classification and event extraction technology to dynamically adjust calculation rules and declaration logic. Based on relevant rule texts, a core ontology structure for the knowledge graph is defined, covering three main entity types: product categories (e.g., steel, aluminum, cement), emission factors (e.g., carbon emission factor per unit of energy consumption), and calculation methods (e.g., life cycle assessment). Relationships between entities include applicability relationships (e.g., "electricity emission factors are applicable to indirect emission calculations across all industries"), inclusion relationships (e.g., "steel products include hot-rolled steel plates and cold-rolled steel plates"), and computational dependency relationships (e.g., "total emission calculations depend on both direct and indirect emission factors"). The final constructed knowledge graph contains no fewer than 5,000 entities and 15,000 relationship pairs.

[0042] The real-time rule update text is filtered using a text classification model (such as a binary classification model based on BERT fine-tuning) to identify valid CBAM-related documents. Event extraction techniques (such as an event extraction model based on BERT) are used to extract rule change events from valid documents, including event type (e.g., "industry coverage expansion," "calculation method adjustment"), affected entities (e.g., "hydrogen products added to coverage"), new and old rule content (e.g., "original emission factor calculation baseline was 2021, updated to 2023"), and effective date. The knowledge graph is automatically updated based on the change events: new entities are added (e.g., "hydrogen product classification"), invalid relationships are deleted (e.g., abandoning the applicability relationship between old emission factors and products), attributes are modified (e.g., updating emission factor values), and adjustments to calculation rules and reporting logic are triggered simultaneously (e.g., expanding the product classification list, correcting calculation model formulas).

[0043] S3. Through intelligent data integration and computing engine, it collaboratively collects data from the enterprise itself and its supply chain, standardizes supplier data in different formats, integrates industry emission factor library, automatically matches calculation model according to product type, generates carbon emission results that meet CBAM requirements, and maps the data to the corresponding position in the CBAM declaration form. The system distributes standardized data collection templates (supporting Excel, online forms, and other formats) to enterprises and their supply chain suppliers. The templates include required fields (such as product name, output, energy consumption data, raw material usage, and supplier ID). The collected data formats include structured data (Excel spreadsheets), semi-structured data (email forms), and unstructured data (images of handwritten energy consumption records). For unstructured data, a multilingual intelligent parsing engine is used to extract key values ​​(e.g., recognizing handwritten "monthly output 500 tons" using OCR and converting it to the value 500). Heterogeneous data is standardized and uniformly converted into JSON format data objects containing "product ID, carbon emission source, value, unit, and data source."

[0044] Based on product information (e.g., "hot-rolled steel plate") in standardized data, the system automatically matches corresponding factors from an integrated industry emission factor library (e.g., the direct emission factor for the steel industry is 0.3 tCO2 / ton) by semantic similarity matching (e.g., text similarity calculation based on Sentence-BERT) to product categories in the knowledge graph. It then loads a predefined calculation model based on the product type (e.g., "Life Cycle Carbon Emission Calculation Model v2.0" for steel products), inputs activity data (e.g., production volume of 500 tons) and emission factors, calculates direct emissions (emissions from the production process), indirect emissions (emissions from electricity consumption), and implicit emissions (carbon emissions from raw materials), generating a total carbon emission result (e.g., total emissions = direct emissions + indirect emissions + implicit emissions = 150 tCO2).

[0045] Based on the data-form field mapping rules defined in the knowledge graph (such as total emissions corresponding to CBAM form Section B Item 3.2), the calculation results, product information, and supplier information are automatically filled into the corresponding positions in the CBAM electronic form (EXCEL / XML format) to generate a pre-filled report.

[0046] S4. Based on historical declaration data and relevant rule texts, use natural language processing technology to analyze and predict compliance risks to ensure that the declaration data complies with the latest rules.

[0047] A corpus is constructed, containing historical enterprise application data (including approval / rejection records), CBAM rule texts, and adjudication cases. Key industries (e.g., steel, aluminum) and risk issues (e.g., changes in calculation methods) are identified using topic analysis (LDA). Error patterns (e.g., "misuse of kg instead of t") are learned from cases using sequence labeling modeling (BERT-CRF). A risk prediction model (e.g., random forest classifier) ​​is trained based on the extracted features. After users submit application data, the model compares the current data with the rule text features (e.g., checking compliance with the latest emission factor requirements in 2025), matches historical error patterns, and outputs risk points (e.g., "emission factor unit mismatch") and risk levels (high / medium / low).

[0048] Based on the currently active rules in the dynamic CBAM knowledge graph (such as "Total emissions = direct emissions + indirect emissions + implicit emissions" and "HS code 6 digits must be consistent with product classification"), the rules are converted into machine-executable verification logic (mathematical relationship verification, logical consistency verification, and scope compliance verification). After the user triggers "one-click verification", the system scans all fields in the form, executes the verification rules in parallel, marks logical errors (such as "indirect emissions calculation omits grid emission factor"), associates the source rule clauses in the knowledge graph (such as referencing CBAM Article 5.3), generates correction suggestions (such as "suggesting to supplement the 2025 grid emission factor of 0.25tCO2 / kWh"), and highlights the erroneous fields.

[0049] In some embodiments, step S1 includes: S11. Obtain CBAM rule files and unstructured documents, the document formats including PDF, scanned images, email attachments or Word documents; for image documents, use optical character recognition technology to perform text recognition and extraction, and convert the documents into a machine-readable text format; S12. A multilingual pre-trained model based on the Transformer architecture is used to translate non-Chinese text into Chinese in real time, while preserving the correspondence between the original text format and key terms. S13. Based on a pre-trained named entity recognition model, sequence labeling and parsing are performed on the translated text to identify and extract key information entities. The key information entities include at least: product name, product HS code, carbon emission factor, calculation rule clauses, geographical boundary definition, and reporting obligation description. S14. Perform structured processing on the extracted key information entities and output structured data.

[0050] In some embodiments, step S2 includes: S21. Based on relevant rule texts, define the core ontology structure of the knowledge graph. The ontology structure includes entity types, as well as the applicability, inclusion, and computational dependencies between entity types. Entity types include product categories, emission factors, and calculation methods. S22. Using natural language processing technology, extract entities and their relationship pairs in batches from relevant rule texts to form a dynamic CBAM knowledge graph; S23. Receive the latest relevant rule text, filter the acquired text using a text classification model, and identify valid documents related to CBAM; then use event extraction technology to identify rule change events from valid documents and extract event elements, including event type, affected entities, new and old rule content, and effective date. S24. Based on the extracted rule change events, perform add, delete, and modify operations on the corresponding entities and relationships in the knowledge graph; the update of the knowledge graph triggers the system to dynamically adjust the calculation rules and reporting logic, including: updating the emission factor library, correcting the calculation method formula, expanding or modifying the product classification list, and changing the data filling requirements.

[0051] In some embodiments, step S22 includes: S221. Clean, segment, and sentence-level divide the acquired relevant rule texts; construct training corpus using back-translation and term substitution techniques to enhance the model's ability to recognize CBAM domain terms; S222. A joint learning framework based on a pre-trained language model is adopted, wherein the joint learning framework shares a text encoding layer and performs named entity recognition and relation extraction subtasks in parallel; the model is fine-tuned using training corpus. S223. Input the relevant rule text to be processed into the fine-tuned model. The model synchronously outputs the identified entities and their relationships, forming a knowledge triple with a head entity-relationship-tail entity structure. S224. Perform entity alignment and cleaning on the knowledge triples extracted from different document sources; import the cleaned triples into the graph database to form a dynamic CBAM knowledge graph.

[0052] In some embodiments, step S24 includes: S241. Analyze the extracted rule change events and automatically assess the scope and level of the impact of the change on the existing knowledge graph structure and downstream computing and reporting logic based on the event type, affected entities, and effective date. Analyze the elements of rule change events and construct an impact assessment matrix: Event type weights: "Industry coverage expansion" weight 0.8, "Calculation method adjustment" weight 0.7, "Emission factor update" weight 0.6; Number of affected entities: ≥100 affected entities, weight 0.9; 50-99 affected entities, weight 0.6; <50 affected entities, weight 0.3. Urgency of effectiveness: 0.9 for effectiveness within 1 month, 0.6 for effectiveness within 3 months, and 0.3 for effectiveness more than 6 months.

[0053] Impact level calculation: Impact score = Event type weight × 0.5 + Number of affected entities weight × 0.3 + Urgency of implementation weight × 0.2. A score ≥ 0.7 is "High", 0.4-0.6 is "Medium", and < 0.4 is "Low". For example, the event score for "New plastics industry coverage" = 0.8 × 0.5 + 0.9 × 0.3 + 0.6 × 0.2 = 0.79, which is "High".

[0054] S242. Based on the impact assessment results, automatically generate a set of operation instructions for adding, deleting, and modifying the knowledge graph; the operation instruction set includes: Create new entities or relationships; Deprecate or archive entities or relationships that have become invalid; Modify the attributes of an existing entity or relationship; Based on the impact assessment results, automatically generate knowledge graph update operation instructions: New operations: For "industry coverage expansion" events with a "high" impact level, create new entities (such as "plastic product classification" and "plastic emission factor library") and add new relationships (such as "plastic products → applicable → life cycle assessment method"). Deprecation operation: For the "Emission Factor Update" event, mark the status of the old emission factor entity (such as "2023 Steel Emission Factor") as "invalid" and record the reason for invalidation and the effective date; Modification operation: For the "Calculation method adjustment" event, update the properties of the calculation method entity (e.g., change "formula parameter" from "a=0.8" to "a=0.85"), and record the comparison before and after the modification.

[0055] S243. At the set time point, execute the operation instruction set to perform an atomic update operation on the knowledge graph; at the same time, perform version management on the knowledge graph, and record the version number, change content, basis and effective time of each change; S244. The update of the knowledge graph serves as a triggering event, automatically initiating the system to dynamically adjust the calculation rules and reporting logic synchronously, including: Based on the latest version of the knowledge graph, the emission factor library, calculation model formula, and product type matching rules within the intelligent data integration and computing engine module have been adjusted synchronously. The logic verification rule base and risk prediction model of the risk warning and compliance inspection module are updated based on the latest version of the knowledge graph. The entity library in the named entity recognition model of the multilingual intelligent parsing engine has been updated based on the latest version of the knowledge graph.

[0056] In some embodiments, step S3 includes: S31. Distribute standardized data collection templates to supply chain suppliers through the system to collaboratively collect data from the enterprise itself and its supply chain. S32. Perform parsing operations on the collected multi-format raw data; for unstructured documents, call the multi-language intelligent parsing engine to extract key numerical and textual information; convert heterogeneous data into standardized data objects; S33. Based on the product information in the standardized data object, call the dynamic CBAM knowledge graph to automatically determine the CBAM category to which the product belongs and the corresponding accounting rules; match the precise emission factors from the integrated industry emission factor library according to the product type, and load the corresponding calculation method model; execute the calculation model to generate carbon emission results that meet CBAM requirements and include direct emissions, indirect emissions and implicit emissions. S34. Based on the mapping rules defined in the dynamic CBAM knowledge graph, automatically fill the generated carbon emission results, product information, and supplier information into the corresponding fields of the CBAM declaration form; generate a complete pre-filled report form that can be reviewed and submitted.

[0057] In some embodiments, step S33 includes: S331. Perform semantic similarity matching between the product description information in the standardized data object and the product classification entity in the dynamic CBAM knowledge graph. If the match is successful, automatically bind the CBAM classification code to which the product belongs and its corresponding unique accounting rule path; if the match fails, initiate a manual intervention process for confirmation and mapping. S332. Based on the bound product CBAM classification, query and match the most accurate default emission factor or specific facility-level emission factor from the integrated industry emission factor library; the emission factor library is associated with emission factor entities in the dynamic CBAM knowledge graph; S333. Load the corresponding calculation method model according to the bound accounting rule path; the calculation method model is a predefined calculation function or algorithm module that conforms to the CBAM accounting rules; inject the activity data in the standardized data object and the matched emission factors as input parameters into the calculation model; S334. Execute the calculation model to calculate the direct emissions, indirect emissions, and hidden emissions of the product respectively; synthesize and summarize the emission results according to the CBAM requirements to generate the final total carbon emission intensity and total emissions data of the product.

[0058] In some embodiments, step S4 includes: S41. Continuously collect CBAM-related rule texts and adjudication cases as external corpus; at the same time, collect historical application data and their corresponding review status as internal corpus; together, they form a corpus for training the risk prediction model. S42. Perform deep analysis on the corpus using natural language processing (NLP) techniques to extract text features; the NLP techniques include: The key industries and risk issues identified by the rules are based on thematic models; Based on sequence labeling models, common data error patterns and logical conflict rules that lead to application failures are learned from case studies. Based on the extracted features, a risk prediction model for CBAM compliance risk prediction is trained. S43. After the user completes the filling or mapping of the current declaration data, the risk prediction model is invoked to perform the following operations: By comparing the current declaration data with the features of the rule text, risks caused by misunderstandings of the rules can be predicted; Match current data with learned historical error patterns to identify potential data anomalies and logical contradictions; Output the predicted risk points and risk levels; S44. Perform one-click global logic verification; based on all currently effective calculation and declaration logic rules in the dynamic CBAM knowledge graph, automatically verify all data fields in the declaration form and their inherent consistency relationships; automatically generate specific correction suggestions and highlight them for logical errors or inconsistencies found during verification.

[0059] Verification rule extraction: Extract all currently effective calculation rules (such as "total emissions = direct + indirect + implicit emissions"), declaration logic rules (such as "HS code 6 digits must match product classification"), and data constraint relationships (such as "emission factor value range 0.1-0.5tCO2 / unit") from the dynamic CBAM knowledge graph to form a verification rule base.

[0060] Verification execution: After the user clicks the "One-click Verification" button, the system iterates through all fields of the CBAM declaration form (no less than 50 fields), matches the value, unit, and context information of each field with the rules in the verification rule base, and triggers the applicable verification rules (such as triggering mathematical relationship verification for the "Total Emissions" field and logical consistency verification for the "HS Code" field).

[0061] Results Feedback: Fields that pass validation are marked "Pass"; fields that do not meet the rules are marked as errors, and the error type is recorded (e.g., "Mathematical relationship not valid," "Range exceeded"), linked to the source rule clause in the knowledge graph (e.g., "Validation basis: CBAM Implementation Rules Annex III"), and correction suggestions are generated (e.g., "Total emissions should be 280+75+200=555, currently entered as 500, correction recommended"). Error fields are highlighted in red on the user interface, and hovering the mouse displays the error reason, rule basis, and correction suggestions.

[0062] In some embodiments, step S44 includes: S441. Extract all currently effective calculation rules, declaration logic rules, and constraint relationships between data fields from the dynamic CBAM knowledge graph; formalize the rules into machine-executable verification logic, which includes mathematical relationship verification, logical consistency verification, and scope compliance verification. S442. In response to the user's one-click verification command, perform a global scan of all data fields in the CBAM declaration form; match the value, type, and context information of each field with the verification logic to trigger all applicable verification rules; S443. Execute all triggered validation rules in parallel and calculate the degree of conformity between each data field and the expected value of the rule; when data is detected as not meeting the validation logic, mark it as a logical error point; S444. Summarize all verification results and generate a structured verification report; for each identified logical error point, automatically associate it with the source rule clause in the knowledge graph and generate a correction suggestion; highlight the error data, associated fields and correction suggestions on the user interface.

[0063] like Figure 2As shown, this embodiment of the invention also provides an intelligent assisted customs declaration system based on a carbon border adjustment mechanism using natural language processing, comprising: The multilingual intelligent parsing engine module is used to translate CBAM-related rule files and unstructured documents in real time and extract key information, including product codes and carbon emission data. The dynamic CBAM knowledge graph module is used to construct a knowledge graph based on relevant rule texts. The knowledge graph includes product classification, emission factors and calculation methods, entities and their relationships. It also monitors rule updates in real time through text classification and event extraction technology, and dynamically adjusts calculation rules and declaration logic. The intelligent data integration and computing engine module is used to collaboratively collect data from the enterprise itself and its supply chain, standardize supplier data in different formats, integrate industry emission factor libraries, automatically match calculation models according to product type, generate carbon emission results that meet CBAM requirements, and map the data to the corresponding positions on the CBAM declaration form. The risk warning and compliance check module is used to analyze and predict compliance risks based on historical declaration data and relevant rule texts using natural language processing technology, ensuring that the declaration data complies with the latest rules.

[0064] In some embodiments, the multilingual intelligent parsing engine module includes: The document acquisition and preprocessing unit is used to acquire CBAM rule files and unstructured documents, the document formats of which include PDF, scanned images, email attachments or Word documents; for image documents, optical character recognition technology is used to perform text recognition and extraction, and the documents are uniformly converted into machine-readable text format; The real-time translation unit is used to translate non-Chinese text into Chinese in real time using a multilingual pre-trained model based on the Transformer architecture, while preserving the correspondence between the original text format and key terms. The key information entity identification and extraction unit is used to perform sequence labeling and parsing on the translated text based on a pre-trained named entity recognition model, and to identify and extract key information entities therein. The key information entities include at least: product name, product HS code, carbon emission factor, calculation rule clauses, geographical boundary definition, and reporting obligation description. The structured output unit is used to process the extracted key information entities into a structured form and output structured data.

[0065] In some embodiments, the dynamic CBAM knowledge graph module includes: An ontology construction unit is used to define the core ontology structure of a knowledge graph based on relevant rule texts. The ontology structure includes entity types and the applicability, inclusion, and computational dependencies between entity types. Entity types include product categories, emission factors, and calculation methods. The knowledge extraction and graph construction unit is used to extract entities and their relationship pairs in batches from relevant rule texts using natural language processing technology, forming a dynamic CBAM knowledge graph. The rule monitoring and event extraction unit is used to receive the latest relevant rule texts, filter the acquired texts using a text classification model, and identify valid documents related to CBAM. Then, it uses event extraction technology to identify rule change events from the valid documents and extract event elements, including event type, affected entities, new and old rule content, and effective date. The knowledge graph update and logic adjustment unit is used to add, delete, and modify the corresponding entities and relationships in the knowledge graph based on the extracted rule change events. The knowledge graph update triggers the system to dynamically adjust the calculation rules and reporting logic. The adjustments include: updating the emission factor library, correcting the calculation method formula, expanding or modifying the product classification list, and changing the data filling requirements.

[0066] In some embodiments, the knowledge extraction and graph construction unit is specifically used for: The acquired rule texts are cleaned, segmented, and divided into sentences; back-translation and term substitution techniques are used to construct training corpora to enhance the model's ability to recognize CBAM domain terms; A joint learning framework based on a pre-trained language model is adopted, which shares a text encoding layer and performs named entity recognition and relation extraction subtasks in parallel; the model is fine-tuned using training corpus. The relevant rule text to be processed is input into the fine-tuned model, and the model outputs the identified entities and their relationships in a synchronized manner, forming a knowledge triple with a head entity-relationship-tail entity structure. The knowledge triples extracted from different document sources are subjected to entity alignment and cleaning; the cleaned triples are then imported into a graph database to form a dynamic CBAM knowledge graph.

[0067] In some embodiments, the map update and logic adjustment unit is specifically used for: The system analyzes extracted rule change events and automatically assesses the scope and level of impact of the change on the existing knowledge graph structure and downstream computation and reporting logic based on the event type, affected entities, and effective date. Based on the impact assessment results, an automatic set of operation instructions for adding, deleting, and modifying the knowledge graph is generated; the operation instruction set includes: creating new entities or relations; deprecating or archiving expired entities or relations; and modifying the attributes of existing entities or relations. At the set time points, the operation instruction set is executed to perform atomic update operations on the knowledge graph; at the same time, version management of the knowledge graph is performed, recording the version number, content of the change, basis and effective time of each change; The update of the knowledge graph serves as a triggering event, automatically initiating the system to dynamically adjust the calculation rules and reporting logic in sync.

[0068] In some embodiments, the intelligent data integration and computing engine module includes: The collaborative data collection unit is used to distribute standardized data collection templates to supply chain suppliers through the system, and to collaboratively collect data from the enterprise itself and its supply chain. The data standardization preprocessing unit is used to perform parsing operations on the collected multi-format raw data; for unstructured documents, it calls a multi-language intelligent parsing engine to extract key numerical and textual information; and it converts heterogeneous data into standardized data objects. The carbon emission accounting unit is used to automatically determine the CBAM category of a product and the corresponding accounting rules by calling the dynamic CBAM knowledge graph based on product information in standardized data objects; match the precise emission factors from the integrated industry emission factor library according to product type and load the corresponding calculation method model; execute the calculation model to generate carbon emission results that meet CBAM requirements and include direct emissions, indirect emissions and implicit emissions. The form mapping and generation unit is used to automatically fill the corresponding fields of the CBAM declaration form with the generated carbon emission results, product information and supplier information according to the mapping rules defined in the dynamic CBAM knowledge graph; and generate a complete pre-filled form that can be reviewed and submitted.

[0069] In some embodiments, the carbon emission accounting unit is specifically used for: The product description information in the standardized data object is semantically matched with the product classification entity in the dynamic CBAM knowledge graph. If the match is successful, the CBAM classification code to which the product belongs and its corresponding unique accounting rule path are automatically bound. If the match fails, a manual intervention process is initiated to confirm and map the data. Based on the bound product CBAM classification, the most accurate default emission factor or specific facility-level emission factor is queried and matched from the integrated industry emission factor library; the emission factor library is associated with emission factor entities in the dynamic CBAM knowledge graph; According to the bound accounting rule path, the corresponding calculation method model is loaded; the calculation method model is a predefined calculation function or algorithm module that conforms to the CBAM accounting rules; the activity data in the standardized data object and the matched emission factors are used as input parameters and injected into the calculation model; The calculation model is executed to calculate the direct emissions, indirect emissions, and hidden emissions of the product. The emission results are then synthesized and summarized according to the CBAM requirements to generate the final data on the product's total carbon emission intensity and total emissions.

[0070] In some embodiments, the risk warning and compliance check module includes: The multi-source corpus construction unit is used to continuously collect CBAM-related rule texts and adjudication cases as external corpus; at the same time, the system's historical application data and its corresponding review status are used as internal corpus; together, they constitute the corpus used for training the risk prediction model. The risk prediction model building unit is used to perform in-depth analysis of the corpus using natural language processing technology to extract text features; based on the extracted features, a risk prediction model for CBAM compliance risk prediction is trained; the natural language processing technology includes: identifying key industries and risk issues of concern based on topic modeling; and learning common data error patterns and logical conflict rules that lead to declaration failure from cases based on sequence labeling modeling. The real-time risk prediction unit is used to call the risk prediction model to perform the following operations after the user completes the filling or mapping of the current declaration data: compare the current declaration data with the rule text features to predict the risk caused by the misunderstanding of the rules; match the current data with the learned historical error patterns to identify potential data anomalies and logical contradictions; and output the predicted risk points and risk levels. The one-click logic verification unit is used to perform one-click global logic verification; based on all currently effective calculation and declaration logic rules in the dynamic CBAM knowledge graph, it automatically verifies all data fields in the declaration form and their inherent consistency relationships; for logical errors or inconsistencies found during verification, it automatically generates specific correction suggestions and highlights them.

[0071] In some embodiments, the one-click logic verification unit is specifically used for: Extract all currently effective calculation rules, declaration logic rules, and constraint relationships between data fields from the dynamic CBAM knowledge graph; formalize the rules into machine-executable verification logic, which includes mathematical relationship verification, logical consistency verification, and scope compliance verification; In response to the user's one-click verification command, a global scan of all data fields in the CBAM declaration form is performed; the value, type, and context information of each field are matched with the verification logic to trigger all applicable verification rules; All triggered validation rules are executed in parallel, and the degree of conformity between each data field and the expected value of the rule is calculated; when data is detected that does not meet the validation logic, it is marked as a logical error point. Summarize all validation results and generate a structured validation report; for each identified logical error, automatically associate it with the source rule clause in the knowledge graph and generate a correction suggestion; highlight the error data, associated fields, and correction suggestions on the user interface.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A natural language processing-based carbon border adjustment mechanism intelligent auxiliary declaration method, characterized in that, The method comprises the following steps: S1, real-time translation and key information extraction of CBAM related rule files and unstructured documents through a multilingual intelligent analysis engine, the key information including product code and carbon emission data; S2, constructing a dynamic CBAM knowledge graph based on related rule texts, the knowledge graph including product classification, emission factor and calculation method entities and their relationships, and real-time monitoring of rule updates through text classification and event extraction technology, dynamically adjusting calculation rules and declaration logic; S3, through an intelligent data integration and calculation engine, cooperatively collecting enterprise itself and supply chain data, standardizing different formats of supplier data, integrating an industry emission factor library, automatically matching calculation models according to product types, generating carbon emission results meeting CBAM requirements, and mapping the data to corresponding positions of CBAM declaration forms; S4, based on historical declaration data and related rule texts, analyzing and predicting compliance risks using natural language processing technology to ensure that the declaration data meets the latest rules.

2. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 1, characterized in that, The steps of S1 include: S11, obtaining CBAM rule files and unstructured documents, the document formats including PDF, scanned images, email attachments or Word documents; for image documents, using optical character recognition technology for character recognition and extraction, and converting the documents into machine-readable text format; S12, using a multilingual pre-training model based on the Transformer architecture to translate non-Chinese texts into Chinese in real time while retaining the correspondence between the original text format and key terms; S13, based on a pre-trained named entity recognition model, performing sequence labeling and analysis on the translated text to identify and extract key information entities, the key information entities including at least product name, product HS code, carbon emission factor, calculation rule clause, geographical boundary definition and reporting obligation description; S14, structuring the extracted key information entities and outputting structured data.

3. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 2, characterized in that, The steps of S2 include: S21, based on related rule texts, defining the core ontology structure of the knowledge graph, the ontology structure including entity types and applicable relationships, inclusion relationships and calculation dependency relationships between the entity types; the entity types including product classification, emission factor and calculation method; S22, using natural language processing technology to batch extract entities and their relationship pairs from related rule texts to form a dynamic CBAM knowledge graph; S23, receiving the latest related rule texts, filtering the obtained texts using a text classification model to identify effective documents related to CBAM; then using event extraction technology to identify rule change events from the effective documents and extract event elements, the event elements including event type, affected entity, new and old rule content and effective date; S24, according to the extracted rule change event, the corresponding entity and relationship in the knowledge graph are operated by adding, deleting and modifying; the update of the knowledge graph triggers the system to dynamically adjust the calculation rules and the declaration logic, and the adjustment includes: updating the emission factor library, correcting the calculation method formula, expanding or modifying the product classification list, and changing the data reporting requirements.

4. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 3, characterized in that, The steps of S22 include: S221, clean, segment and sentence-level segment the obtained related rule text; use back-translation and term replacement technology to construct training corpus to enhance the model's recognition ability of CBAM domain terms; S222, use a joint learning framework based on a pre-trained language model, which shares a text encoding layer and performs the subtasks of named entity recognition and relation extraction in parallel; use the training corpus to fine-tune the model; S223, input the related rule text to be processed into the fine-tuned model, and the model synchronously outputs the recognized entities and their relationships to form knowledge triples with the structure of head entity-relation-tail entity; S224, align and clean the knowledge triples extracted from different document sources; import the cleaned triples into a graph database to form a dynamic CBAM knowledge graph.

5. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 4, characterized in that, The steps of S24 include: S241, analyze the extracted rule change event, and automatically evaluate the influence range and level of the change on the existing knowledge graph structure and downstream calculation and declaration logic according to the event type, affected entity and effective date; S242, according to the influence evaluation result, automatically generate a set of operation instructions for adding, deleting and modifying the knowledge graph; the operation instruction set includes: create a new entity or relationship; discard or archive an invalid entity or relationship; modify the attributes of an existing entity or relationship; S243, at a set time point, execute the operation instruction set to perform an atomic update operation on the knowledge graph; at the same time, version management is performed on the knowledge graph, and the version number, change content, basis and effective time of each change are recorded; S244, the update of the knowledge graph as a trigger event automatically triggers the system to synchronously and dynamically adjust the calculation rules and the declaration logic, including: According to the latest version of the knowledge graph, synchronously adjust the emission factor library, calculation model formula and product type matching rule in the intelligent data integration and calculation engine module; According to the latest version of the knowledge graph, update the logic verification rule library and risk prediction model of the risk early warning and compliance checking module; According to the latest version of the knowledge graph, update the entity library in the named entity recognition model of the multi-language intelligent parsing engine.

6. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 5, characterized in that, The steps of S3 include: S31, distribute standardized data collection templates to supply chain suppliers through the system to cooperatively collect enterprise's own and supply chain data; S32, perform parsing operation on the collected multi-format raw data; for unstructured documents, call the multi-language intelligent parsing engine to extract key numerical and text information; convert heterogeneous data into standardized data objects; S33, according to the product information in the standardized data object, calling the dynamic CBAM knowledge graph, automatically determining the CBAM classification to which the product belongs and the corresponding accounting rules; according to the product type, matching the accurate emission factor from the integrated industry emission factor library, and loading the corresponding calculation method model; executing the calculation model to generate the carbon emission results conforming to the CBAM requirements, including direct emission, indirect emission and implied emission; S34, according to the mapping rules defined in the dynamic CBAM knowledge graph, the generated carbon emission results, product information and supplier information are automatically filled into the corresponding fields of the CBAM declaration form; a complete pre-reporting form is generated, which can be audited and submitted.

7. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 6, characterized in that, The steps of S33 include: S331, the product description information in the standardized data object is matched with the product classification entity in the dynamic CBAM knowledge graph in terms of semantic similarity, if the matching is successful, the CBAM classification code to which the product belongs and the corresponding unique accounting rule path are automatically bound; if the matching fails, manual intervention process is started for confirmation and mapping; S332, according to the bound product CBAM classification, the most accurate default emission factor or specific facility level emission factor is queried and matched from the integrated industry emission factor library; the emission factor library is associated with the emission factor entity in the dynamic CBAM knowledge graph; S333, according to the bound accounting rule path, the corresponding calculation method model is loaded; the calculation method model is a predefined calculation function or algorithm module conforming to the CBAM accounting rules; the activity data in the standardized data object and the matched emission factor are input into the calculation model as input parameters; S334, execute the calculation model to account for the direct emission, indirect emission and implied emission of the product respectively; according to the CBAM requirements, the emission results are synthesized and summarized to generate the final product total carbon emission intensity and total emission data.

8. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 7, characterized in that, The steps of S4 include: S41, continuously collect CBAM related rule texts and adjudication cases as external corpus, at the same time, collect system historical declaration data and their corresponding audit status as internal corpus, together constitute a corpus library for risk prediction model training; S42, use natural language processing technology to deeply analyze the corpus library and extract text features; the natural language processing technology includes: Based on the topic model, identify the key industries and risk issues concerned by the rules; Based on the sequence labeling model, learn the common data error patterns and logical conflict rules that lead to declaration failure from the cases; Based on the extracted features, train a risk prediction model for CBAM compliance risk prediction; S43, after the user completes the filling or mapping of the current declaration data, the risk prediction model is called to perform the following operations: Compare the current declaration data with the rule text features to predict the risks caused by rule understanding deviation; Match the current data with the learned historical error patterns to identify potential data abnormalities and logical contradictions; Output the predicted risk points and risk levels; S44, performing one-key global logical verification; based on all currently effective calculation and declaration logic rules in the dynamic CBAM knowledge graph, automatically verifying all data fields in the declaration form and their internal cross-checking relationships; for the logical errors or inconsistencies found, automatically generating specific correction suggestions and highlighting them.

9. The natural language processing based intelligent assistant customs declaration method for carbon border adjustment mechanism according to claim 8, characterized in that, The steps of S44 include: S441, extracting all currently effective calculation rules, declaration logic rules and constraint relationships between data fields from the dynamic CBAM knowledge graph; formalizing the rules into machine executable verification logic, which includes mathematical relationship verification, logical consistency verification, and range compliance verification; S442, in response to a one-key verification instruction from the user, globally scanning all data fields in the CBAM declaration form; matching the value, type and context information of each field with the verification logic to trigger all applicable verification rules; S443, executing all triggered verification rules in parallel, calculating the degree of compliance of each data field with the expected value of the rule; when it is detected that the data does not meet the verification logic, marking it as a logical error point; S444, aggregating all verification results to generate a structured verification report; for each identified logical error point, automatically associating the source rule clause in the knowledge graph and generating a correction suggestion; highlighting the error data, associated fields and correction suggestions on the user interface.

10. A natural language processing based carbon border adjustment mechanism intelligent auxiliary declaration system, characterized in that, It includes: A multilingual intelligent analysis engine module for real-time translation and key information extraction of CBAM-related rule files and unstructured documents, including product codes and carbon emission data; A dynamic CBAM knowledge graph module for building a knowledge graph based on related rule texts, including product classification, emission factors and calculation method entities and their relationships, and real-time monitoring of rule updates through text classification and event extraction technology to dynamically adjust calculation rules and declaration logic; An intelligent data integration and calculation engine module for collecting enterprise and supply chain data, standardizing different formats of supplier data, integrating industry emission factor libraries, automatically matching calculation models based on product types, generating carbon emission results that meet CBAM requirements, and mapping data to the corresponding positions of the CBAM declaration form; A risk warning and compliance checking module for analyzing and predicting compliance risks based on historical declaration data and related rule texts using natural language processing technology to ensure that the declaration data meets the latest rules.

Images