A cloud-based intelligent green product auditing and verification method

Abstract

This invention provides an intelligent green product review and verification method based on a cloud platform, belonging to the field of cloud platform technology. The method includes: obtaining a green review specification process based on the green attributes of the product to be reviewed and in conjunction with an attribute-specification database; retrieving uploaded materials for each review specification in the green review specification process from the cloud platform; mapping the uploaded materials one-to-one with each review condition in the review specification, and standardizing the corresponding sub-materials according to the material requirements of each review condition, and annotating the converted sub-materials according to the execution logic of the review standard of the review condition to obtain an review set; performing green verification and anti-counterfeiting verification on each production stage of the product to be reviewed to obtain a verification set; and comprehensively analyzing the review set and the verification set to obtain the output result of the product to be reviewed. This improves review efficiency and accuracy, ensuring the authority and credibility of green product certification.

Description

Technical Field

[0001] This invention relates to the field of cloud platform technology, and in particular to an intelligent green product auditing and verification method based on a cloud platform. Background Technology

[0002] With the deepening of the green development concept, green product certification has become an important means of regulating market order and guiding green consumption. As an authoritative certification system, the national green product certification process covers multiple stages, including basic information verification, review of technical documents, and on-site inspections. It also includes specific clauses on controlling key factors affecting the green attributes of products, verifying compliance with design, development, and green evaluation requirements. Common audit methods involve document review, on-site inspections, data verification, and interviews.

[0003] However, the existing green product audit and verification process has many limitations: First, the audit process relies on manual operation. For example, document verification requires manual comparison of documents such as business licenses and environmental testing reports submitted by enterprises, and on-site inspections require inspectors to track the entire production process. This is not only inefficient but also prone to omissions due to human error. Second, there is insufficient consistency in the implementation of audit standards. Different inspectors have different grasps of the standard comparison criteria in document review and the key points of process tracking in on-site inspections. For example, there is a lack of unified standards for determining the depth of production record tracing and the sample size for sampling and testing, which may lead to deviations in audit results. Third, information management is fragmented, with most of the materials uploaded by enterprises being paper-based. The data is stored in physical or local file formats, lacking centralized management based on a cloud platform, making it difficult to retrieve data and trace historical data, and information distortion is prone to occur during data transmission; fourth, the verification process has a low level of intelligence, and the statistical analysis of green attribute-related data (such as energy consumption and pollutant emissions) relies on manual calculation, and anti-counterfeiting verification mostly relies on manual verification of production records, making it difficult to achieve real-time dynamic monitoring of the production process and posing a risk of false data; fifth, the comprehensive analysis of audit and verification results lacks a systematic mechanism, and the integration of "compliant", "partially compliant", and "non-compliant" labeling results and the correlation analysis with production process labels are insufficient, making it difficult to form accurate green product evaluation conclusions.

[0004] Therefore, this invention proposes an intelligent green product auditing and verification method based on a cloud platform. Summary of the Invention

[0005] This invention provides an intelligent green product review and verification method based on a cloud platform, which improves review efficiency and accuracy through standardized process design, intelligent data processing and comprehensive analysis, and ensures the authority and credibility of green product certification.

[0006] This invention provides an intelligent green product auditing and verification method based on a cloud platform, comprising:

[0007] Step 1: Based on the green attributes of the product to be reviewed and in conjunction with the attribute-specification database, obtain the green review specification process, and retrieve the uploaded materials based on each review specification in the green review specification process from the cloud platform;

[0008] Step 2: Match the uploaded materials with each review condition in the review specifications, and perform standard conversion on the corresponding sub-materials according to the material requirements of each review condition. Then, mark the converted sub-materials according to the execution logic of the review standards of the review conditions to obtain the review set.

[0009] Step 3: Perform green verification and anti-counterfeiting verification on each production stage of the product to be audited to obtain a verification set;

[0010] Step 4: Perform a comprehensive analysis on the audit set and the verification set to obtain the output results of the product to be audited.

[0011] Preferably, based on the green attributes of the product to be reviewed and in conjunction with the attribute-specification database, a green review specification process is derived, including:

[0012] Construct a feature vector for the green attribute of the product to be reviewed, wherein the feature vector includes first-level dimension parameters and second-level dimension parameters;

[0013] Based on the feature vector, a hierarchical retrieval is performed on the attribute-norm database, and the association relationship between norms at each layer is constructed through knowledge graph technology. The hierarchical retrieval includes: basic norm layer retrieval, industry norm layer retrieval, and dynamic norm layer retrieval.

[0014] Based on the full lifecycle timeline logic of the product to be reviewed, the initial conditions for review under each timeline logic are extracted, and an initial review specification process is generated.

[0015] Based on the aforementioned relationships and the product scale of the products to be reviewed, the initial review process is adjusted to obtain a green review process.

[0016] Preferably, the corresponding sub-data is standardized according to the data requirements of each audit condition, including:

[0017] Based on the review criteria, a dynamic conversion rule base is generated for the required materials. At the same time, the metadata and content features of the sub-materials are analyzed using natural language processing technology to extract material features.

[0018] The data features are matched with a dynamic conversion rule base to generate converted sub-data.

[0019] Preferably, the converted sub-data is labeled according to the execution logic of the review criteria of the aforementioned review conditions to obtain a review set, including:

[0020] The execution logic of the audit standard is broken down into logical nodes and node relationships, wherein the logical nodes include: judgment nodes, jump nodes and / or condition nodes;

[0021] The transformed sub-data is feature-mapped, and matching features with logical nodes are extracted to generate a feature matching matrix;

[0022] A labeling decision tree is established based on node relationships, and decisions are made based on the feature matching matrix. All labeling results are integrated to form an audit set. The root node of the labeling decision tree is the overall audit standard target, the leaf nodes are specific labeling results, and the non-leaf nodes are logical judgment conditions.

[0023] Preferably, anti-counterfeiting verification is performed, including:

[0024] Based on the permissions of each production stage, determine the production cycle of the corresponding production stage and obtain the production status under the most recent production cycle.

[0025] When the production status matches the verification indicators of the green verification results, a true label is assigned to the corresponding production process.

[0026] Otherwise, a false label is assigned to the corresponding production stage.

[0027] Preferably, a comprehensive analysis is performed on the audit set and the verification set to obtain the output result of the product to be audited, including:

[0028] The audit set is divided into static features and dynamic features, and the verification set is divided into process features and time sequence features.

[0029] Extract the first feature containing non-conformities from the static and dynamic features, and extract the second feature belonging to pseudo-labels from the link features and time-series features to determine the co-occurrence risk vector;

[0030] Based on the remaining features of the static and dynamic features, and the remaining features based on the link features and time sequence features, a defect risk vector is determined.

[0031] The co-occurrence risk vector and the defect risk vector are input into the two-vector analytical model to obtain the output results of the product to be audited.

[0032] Preferably, before retrieving the uploaded materials based on each review standard in the green review standard process from the cloud platform, the process further includes:

[0033] When a third party uploads initial data to the cloud platform, the data types involved in the initial data are traversed and recorded, and an encrypted structure is created for each data type based on the green attribute. The data types are pure image type, pure text type, and mixed image and text type.

[0034] The system obtains the environmental difference between the location environment of the data file based on the third-party storage location and the preset environment. When the environmental difference is greater than the preset difference, a dual detection mechanism is activated to perform content detection and integrity detection on the data file, and a new file is constructed and saved.

[0035] Otherwise, retain the aforementioned data file;

[0036] The transmission path to the third-party sending interface is determined based on the storage location of each file.

[0037] The input and output trajectory logs of each path node and its adjacent nodes in the transmission path are retrieved from the historical database, and the input and output trajectory logs are analyzed and a trajectory matrix is ​​constructed based on the analysis indicators consistent with the input and output.

[0038] The feature vectors of the trajectory matrix are obtained and input into the vector analysis model to obtain the safety coefficients of the corresponding path nodes;

[0039] Select the smallest security factor from all security factors under the transmission path as the reference factor;

[0040] When the storage location of the corresponding retained file is a separate location, the corresponding reference coefficient is assigned to the corresponding retained file. Here, the separate location means that there is only one retained file on the longest path to which the corresponding retained file belongs.

[0041] When the storage location of the corresponding retain file is multiple locations, the reference coefficients are sorted sequentially from front to back to obtain a coefficient array. If the first coefficient in the coefficient array is the minimum coefficient, the minimum coefficient is assigned to the retain file at each of the multiple locations. If the last coefficient in the coefficient array is the minimum coefficient, the original reference coefficients at the multiple locations are kept unchanged and assigned to the corresponding retain file. Otherwise, the minimum coefficient in the coefficient array is locked, and the reference coefficients before the minimum coefficient are kept unchanged. The reference coefficients after the minimum coefficient are replaced with the corresponding minimum coefficient and assigned to the corresponding retain file.

[0042] Determine whether each stored file has undergone third-party encryption. If so, establish a structural mapping relationship between the encryption operation and the encryption structure, as well as an attribute mapping relationship between the basic attributes of the stored file and the structural attributes of the encryption structure. Combined with the assigned coefficients, perform a second encryption operation on the data file.

[0043] If it does not exist, perform a new encryption operation on the data file according to the encryption structure and the corresponding assigned coefficient;

[0044] The encrypted files are stored on a cloud platform.

[0045] Preferably, after obtaining the output result of the product to be reviewed, the method further includes:

[0046] The output results of all products to be reviewed are classified according to the green attribute, and abnormal and normal features are extracted from the process set of each review specification in the classification results to determine the first occurrence state of different abnormal features and the second occurrence state of different normal features.

[0047] Based on the first occurrence state, the second occurrence state and the corresponding review specifications, hidden associations are mined, and content not covered by the current review specifications is listed as initial items;

[0048] Determine the expansion probability of each initial item, and regard initial items with expansion probabilities greater than preset probabilities as expansion items, so as to update the corresponding audit specifications.

[0049] Compared with the prior art, the beneficial effects of this application are as follows:

[0050] By constructing quantified green attribute feature vectors, combining hierarchical retrieval with knowledge graphs to associate each layer of specifications, and then dynamically adjusting according to the product's full lifecycle timeline logic and scale, the resulting green audit specification process can accurately cover basic, industry, and dynamic requirements, while also adapting to the product's own characteristics. This significantly improves the relevance and compliance of the audit process, reduces redundant steps, and provides a scientific and orderly execution framework for subsequent audits and verifications.

[0051] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0052] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 This is a flowchart of an intelligent green product review and verification method based on a cloud platform, as described in an embodiment of the present invention. Detailed Implementation

[0055] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0056] This invention provides an intelligent green product auditing and verification method based on a cloud platform, such as... Figure 1 As shown, it includes:

[0057] Step 1: Based on the green attributes of the product to be reviewed and in conjunction with the attribute-specification database, obtain the green review specification process, and retrieve the uploaded materials based on each review specification in the green review specification process from the cloud platform;

[0058] Step 2: Match the uploaded materials with each review condition in the review specifications, and perform standard conversion on the corresponding sub-materials according to the material requirements of each review condition. Then, mark the converted sub-materials according to the execution logic of the review standards of the review conditions to obtain the review set.

[0059] Step 3: Perform green verification and anti-counterfeiting verification on each production stage of the product to be audited to obtain a verification set;

[0060] Step 4: Perform a comprehensive analysis on the audit set and the verification set to obtain the output results of the product to be audited.

[0061] In this embodiment, the product to be reviewed refers to the product that needs to be reviewed in accordance with regulations, and the green attribute is related to the first-level dimension parameters and the second-level dimension parameters. The attribute-regulation database is a database that stores the mapping relationship between the green attribute and the corresponding review regulations.

[0062] The green audit standard process is a set of orderly audit steps determined based on the green attributes of the product to be audited and the attribute-standard database. It covers aspects such as document review and on-site verification. For example, the green audit standard process for new energy vehicles is: raw material environmental certification audit → battery production energy consumption testing → vehicle pollutant emission verification → recycling and dismantling process evaluation.

[0063] The uploaded materials are relevant documents submitted by third parties through the cloud platform, including qualification certificates, test reports, production records, etc., such as business licenses, environmental protection test reports (such as exhaust gas emission concentration tests), production energy consumption records, and raw material purchase vouchers. The uploaded materials are encrypted.

[0064] Audit criteria are specific judgment indicators or requirements in the audit specifications, and are the smallest unit of audit.

[0065] Sub-data refers to the detailed data in the uploaded data that corresponds to a single review condition. It is the object of standard conversion and annotation.

[0066] Standard conversion is used to convert sub-data into a unified format, unit, or structure that meets the audit requirements, ensuring data comparability. For example, energy consumption data in different units (such as "kWh / unit" or "kWh / unit") can be uniformly converted to "kWh / unit"; and image-format test reports can be converted into text data using OCR.

[0067] The execution logic of the audit standards is the order of audit conditions, the rules of association, and the judgment logic. For example, condition B can only be audited if condition A is met, or if either A or B is met, the standard is met.

[0068] The labeling is based on the execution logic of the audit standards, marking the converted sub-data as "compliant", "partially compliant", or "non-compliant". For example, if the sub-data shows "wastewater discharge of 45 tons / day" (≤50 tons / day) → labeled "compliant"; "renewable materials account for 25%" (≥30% is not up to standard) → labeled "non-compliant".

[0069] In this embodiment, the review set is the collection of annotation results for all sub-data, including sub-data identifier, corresponding review conditions, annotation conclusion and judgment basis. For example, {Sub-data ID:20240601, Review conditions: Wastewater discharge ≤50 tons / day, Annotation: Compliant, Basis: Monitoring daily report data 45 tons / day}.

[0070] In this embodiment, the production process refers to the specific procedures or stages in the product manufacturing process, covering raw material input, processing, assembly, packaging, etc. For example, the "chip welding", "shell injection molding" and "whole machine testing" stages in mobile phone production; and the "fabric dyeing", "cutting" and "sewing" stages in clothing production.

[0071] In this embodiment, green verification checks whether the production process meets green standards. The verification content includes energy consumption, pollutant emissions, resource utilization rate, etc. For example, verifying the "dyeing agent COD value ≤100mg / L" and "water reuse rate ≥80%" in the "fabric dyeing" process. Anti-counterfeiting verification confirms the authenticity of the production process data, prevents false records or tampering, and ensures the credibility of the verification results.

[0072] The verification set is a collection of green verification results and anti-counterfeiting verification labels for all production processes. It includes process identification, green compliance status, and authenticity labels. For example, {process: fabric dyeing, green verification: COD value 90mg / L (compliant), anti-counterfeiting label: genuine}; {process: chip welding, green verification: energy consumption exceeds the standard, anti-counterfeiting label: genuine}.

[0073] In this embodiment, the comprehensive analysis combines the annotation results of the audit set with the green compliance and anti-counterfeiting label status of the verification set to draw a final conclusion. For example, if the audit set has 80% "compliant" items, the verification set has a 90% green compliance rate and no counterfeit labels, the comprehensive judgment is passed; if the audit set has 30% "non-compliant" items and the verification set has counterfeit labels in two stages, the comprehensive judgment is failed.

[0074] The output results are the product green attribute audit conclusions derived from comprehensive analysis, including "passed", "failed", "conditionally passed" and detailed explanations, such as "passed green product audit, meeting the standards for: environmental friendliness of raw materials and energy consumption in production; suggested improvement items: recyclability of packaging materials"; "failed, reason: false labels in 3 production stages and excessive wastewater discharge".

[0075] The beneficial effects of the above technical solution are: by integrating the entire process of review and verification through the cloud platform, it realizes standardized data conversion, intelligent labeling and real-time verification in the production process, reduces manual operation, lowers the review deviation rate, and ensures data credibility through blockchain anti-counterfeiting, ultimately forming accurate green product evaluation results and improving review efficiency and authority.

[0076] This invention provides an intelligent green product review and verification method based on a cloud platform. Based on the green attributes of the product to be reviewed and combined with an attribute-specification database, a green review specification process is derived, including:

[0077] Construct a feature vector for the green attribute of the product to be reviewed, wherein the feature vector includes first-level dimension parameters and second-level dimension parameters;

[0078] Based on the feature vector, a hierarchical retrieval is performed on the attribute-norm database, and the association relationship between norms at each layer is constructed through knowledge graph technology. The hierarchical retrieval includes: basic norm layer retrieval, industry norm layer retrieval, and dynamic norm layer retrieval.

[0079] Based on the full lifecycle timeline logic of the product to be reviewed, the initial conditions for review under each timeline logic are extracted, and an initial review specification process is generated.

[0080] Based on the aforementioned relationships and the product scale of the products to be reviewed, the initial review process is adjusted to obtain a green review process.

[0081] In this embodiment, the feature vector of the green attributes of the product to be audited is to transform the green attributes of the product to be audited into a quantifiable vector form for accurate matching of audit specifications. The vector contains green feature parameters at different levels. For example, the green attribute feature vector is [first-level dimension: environmental friendliness of raw materials (0.8), production energy consumption (0.7), recyclability (0.9); second-level dimension: proportion of non-toxic substances in raw materials (95%), power consumption per unit product (0.5kWh / h), recyclability of plastic parts (80%)], where the values ​​are feature quantification values ​​(0-1 or specific indicators).

[0082] The first-level dimension parameters are the core categories that cover a wide range of green attributes. They are the upper-level classifications of the feature vector and reflect the main aspects of the product's green attributes, such as the environmental friendliness of raw materials, the greenness of the production process, energy consumption during use, recyclability after disposal, and control of pollutant emissions.

[0083] Secondary dimension parameters are specific indicators belonging to the primary dimension. They are a refinement of the primary dimension and are used to quantitatively describe the specific characteristics of green attributes. For example, the secondary dimension parameters corresponding to the primary dimension "environmental friendliness of raw materials" include "content of substances prohibited by EU REACH regulations in raw materials (≤0.1%)", "proportion of renewable materials (≥30%)", and "supplier environmental certification qualifications (with ISO14001 certification)".

[0084] The attribute-specification database stores the mapping relationship between green attributes and corresponding audit specifications. It is used to match applicable audit requirements based on the green attributes of products and contains specification content at different levels. For example, the specifications associated with "first-level dimension: production energy consumption" in the database include "GB12021.3-2010 Energy Efficiency Limits for Washing Machines", "Industry Specification: Unit Product Power Consumption in the Electronics Industry ≤ 5kWh / Unit", and "Dynamic Specification: New Energy Consumption Reduction Policy in a Certain Province in 2024".

[0085] In this embodiment, the hierarchical retrieval is performed by retrieving applicable audit standards from the attribute-standard database step by step according to the hierarchical differences of the standards, ensuring coverage of basic requirements, industry characteristics and the latest policies. For example, basic standards (such as GB / T28382-2012 Electric Vehicle Safety Requirements) are retrieved first, then industry standards (such as Green Factory Evaluation Requirements for the Automotive Industry) are retrieved, and finally dynamic standards (such as the 2024 National New Energy Vehicle Energy Consumption Subsidy Related Audit Standards) are retrieved.

[0086] The basic standard layer consists of general and fundamental standards or regulations applicable to all green products. These are the minimum requirements for auditing and are universal and mandatory. Examples include the basic requirements for the environmental impact of a product throughout its entire life cycle in the GB / T24001 Environmental Management System standard and the General Rules for Green Product Evaluation (GB / T35601-2017).

[0087] The industry standard layer consists of green product audit standards formulated for specific industries (such as electronics, automobiles, textiles, etc.), reflecting the green requirements with industry characteristics. For example, the "Management Measures for Pollution Control of Electronic Information Products" in the electronics industry (requires limits on harmful substances such as lead and mercury in products) and the OEKO-TEX® standard in the textile industry (requires that textiles be non-toxic).

[0088] The dynamic specification layer consists of audit specifications that are dynamically adjusted based on policy updates, technological advancements, or market demands. These include temporary requirements and local pilot standards issued in the past 1-3 years. For example, a province issued "Supplementary Requirements for the Audit of Green Products in High Energy-Consuming Industries" in 2023 (which added a carbon footprint verification item) and the EU's newly issued battery recycling efficiency standard in 2024.

[0089] Knowledge graph technology uses a visual graph format to display the relationships (such as references, supplements, and conflicts) between different levels of specifications, enabling the management of logical relationships between specifications. For example, in a knowledge graph, the "GB / T35601-2017 of the basic specification layer" is connected to the "OEKO-TEX® standard of the industry specification layer" through a "reference" relationship, and is connected to the "local carbon footprint requirements of the dynamic specification layer" through a "supplement" relationship.

[0090] The relationships between different levels of standards refer to the logical connections between them, including references (lower-level standards reference higher-level standards), supplements (dynamic standards refine basic standards), and conflicts (different standards have different requirements for the same indicator). For example, "the 'heavy metal limits' in the electronics industry standards reference the indicators in the basic standard GB / T35601-2017" (reference relationship); "the 'carbon footprint verification' in the dynamic standards supplements the 'energy consumption audit' which is not covered in the basic standard" (supplementary relationship).

[0091] The timeline logic of the entire life cycle of the product to be reviewed is a stage logic formed by the time sequence of the product from raw material procurement, production and processing, warehousing and transportation, use to waste recycling. It reflects the chronological relationship of the entire product life cycle. For example, raw material procurement (pulp) → production and processing (paper making, printing) → warehousing and transportation → consumption and use → waste recycling (pulp recycling).

[0092] Initial conditions are the prerequisites that must be met before the audit is initiated at each stage of the entire life cycle. For example, the initial condition for the raw material procurement stage is that the supplier's environmental qualification certificate has been obtained; the initial condition for the production and processing stage is that the raw material inspection certificate has been uploaded.

[0093] The initial audit standard process is a preliminary audit step arranged in chronological order based on the full life cycle chronological logic and the initial conditions of each stage. It includes the audit standards that need to be performed at each stage, such as raw material environmental certification audit (based on basic standards) → non-toxic testing of the production process (based on industry standards) → degradability verification of packaging materials (based on basic standards).

[0094] In this embodiment, the product size of the product to be audited is a parameter that reflects the production volume of the product to be audited and affects the detailed design of the audit process (such as sampling size and inspection frequency). It usually includes annual output, production batches, market coverage, etc.

[0095] The green audit standard process is the final audit steps formed by adjusting the initial audit standard process. It integrates the relationship between the various levels of standards and the adaptability to the product scale, and is executable. For example, the green audit standard process is: raw material environmental certification audit (basic + industry standard) → production energy consumption test (add 3 samplings, based on scale adjustment) → carbon footprint verification (dynamic standard, supplementing basic standard) → recyclability verification (basic standard).

[0096] The beneficial effects of the above technical solution are as follows: by constructing a quantified green attribute feature vector, combining hierarchical retrieval and knowledge graph association of each layer of specifications, and then dynamically adjusting according to the product's full life cycle time sequence logic and scale, the resulting green audit specification process can accurately cover basic, industry and dynamic requirements, and adapt to the product's own characteristics, greatly improving the relevance and compliance of the audit process, reducing redundant steps, and providing a scientific and orderly execution framework for subsequent audits and verifications.

[0097] This invention provides an intelligent green product review and verification method based on a cloud platform, which standardizes the corresponding sub-data according to the data requirements of each review condition, including:

[0098] Based on the review criteria, a dynamic conversion rule base is generated for the required materials. At the same time, the metadata and content features of the sub-materials are analyzed using natural language processing technology to extract material features.

[0099] The data features are matched with a dynamic conversion rule base to generate converted sub-data.

[0100] In this embodiment, the data requirements of the audit conditions are the specific specifications of the required sub-data in the audit conditions, including file format, content elements, data units, accuracy requirements, etc., which are the basis for data conversion. For example, the data requirements of the "audit conditions for production wastewater discharge" are "to provide a third-party test report (PDF format) within the last 3 months, which must include COD concentration (unit: mg / L, retain 1 decimal place), qualification number of the testing institution, and sampling time".

[0101] The dynamic conversion rule base is a dynamically updated database that stores data conversion rules. The rules cover format conversion, unit conversion, content structuring, etc., and can automatically iterate according to new data types or requirements. For example, the rule base includes rules such as "image format detection report → OCR text conversion", "unit 'ton / day' → 'kg / h' conversion formula (1 ton / day = 41.67kg / h)", and "unstructured text (such as handwritten records) → structured table (including 'detection item-value-date' fields)". When a new "SVG format chart" data type is added, an "SVG → PNG conversion rule" is automatically added.

[0102] Metadata of sub-data describes the attributes of the sub-data itself, without involving the content itself. It is used to identify the basic characteristics of the data, such as file format (PDF / JPG / Excel), file size (2.5MB), creation time (2024-05-01-14:30), author (testing agency A), and hash value (used to verify integrity).

[0103] The content characteristics of sub-data are the key information contained in the sub-data content, including numerical values, keywords, structural features, etc., which reflect the core content of the data. For example, keywords such as "COD concentration 120mg / L" and "exceeding the standard" in the test report text; column headings of "test items" and data distribution patterns in Excel spreadsheets (such as values ​​concentrated in 50-200mg / L); and axis labels of charts in images ("time" and "emissions").

[0104] Data features are a comprehensive set of features formed by integrating the metadata and content features of sub-data. They are used to match with the conversion rule base to determine the applicable conversion rules. For example, the data features of a test report are "Metadata: Format=JPG, Size=1.2MB; Content features: Contains 'COD', '120', and 'mg / L' entities, without structured tables".

[0105] Matching data features with the dynamic conversion rule base involves comparing data features with the triggering conditions of conversion rules to determine the most suitable conversion rule, ensuring that the data is converted according to the review requirements. For example, if a sub-data feature is "format=JPG, content contains detection data", it matches the rule in the rule base "image format and contains text → OCR conversion + structured table generation", triggering the corresponding conversion action.

[0106] The converted sub-data is a standardized sub-data that meets the requirements of the audit criteria after being processed by dynamic conversion rules. It has the characteristics of unified format, structured content, and data compliance. For example, the original sub-data is "JPG format handwritten test record (content: 'May COD120')", which is converted into "PDF format structured table, including 'Test item: COD concentration; value: 120mg / L; test date: 2024-05-XX; remarks: verified by OCR conversion'".

[0107] The beneficial effects of the above technical solution are: by dynamically converting the rule base to adapt to diverse data types and combining natural language processing technology to accurately extract data features, the automatic and standardized conversion of sub-data is achieved, ensuring that the data meets the review requirements and laying a reliable foundation for subsequent annotation and review set construction.

[0108] This invention provides an intelligent green product review and verification method based on a cloud platform. The method involves labeling the converted sub-data according to the execution logic of the review criteria, resulting in a review set, including:

[0109] The execution logic of the audit standard is broken down into logical nodes and node relationships, wherein the logical nodes include: judgment nodes, jump nodes and / or condition nodes;

[0110] The transformed sub-data is feature-mapped, and matching features with logical nodes are extracted to generate a feature matching matrix;

[0111] A labeling decision tree is established based on node relationships, and decisions are made based on the feature matching matrix. All labeling results are integrated to form an audit set. The root node of the labeling decision tree is the overall audit standard target, the leaf nodes are specific labeling results, and the non-leaf nodes are logical judgment conditions.

[0112] In this embodiment, the execution logic of the audit standard is the audit process advancement rules specified in the audit standard, including the indicator judgment order, condition triggering relationship and result derivation logic. It is the core logical framework that guides the audit operation. For example, the execution logic of the "green audit standard for new energy vehicles" is "first audit the environmental protection of battery raw materials, if it meets the standard, then continue to audit the production energy consumption; if it does not meet the standard, it is directly judged as non-compliant".

[0113] In this embodiment, a logical node is an indivisible independent decision unit in the execution logic, used to carry out specific audit operations or condition judgments. It is a basic element in building the audit process. For example, a judgment node: "Is the heavy metal content in the battery raw materials ≤0.01%?" (a binary judgment of whether a specific indicator meets the standard); a jump node: "If the battery raw materials audit meets the standard, jump to the production energy consumption audit node" (pointing to the subsequent node based on the previous result); a condition node: "The production energy consumption audit must meet the conditions of 'raw material audit completed' and 'production records complete'" (prerequisites for triggering node execution).

[0114] In this embodiment, the node association relationship is the connection rule between logical nodes, reflecting the order, dependency relationship or triggering condition of the nodes, to ensure that the review process proceeds in an orderly manner according to the execution logic. For example, the "raw material environmental protection judgment node" and the "production energy consumption review node" have a "compliance → trigger" relationship; the "production energy consumption exceeding the standard judgment node" and the "final non-compliance result node" have a "direct pointing" relationship; and the "packaging material recyclability node" and the "transportation link review node" have a "parallel execution" relationship.

[0115] Feature mapping is the process of associating and matching key information in the transformed sub-data with the core parameters of logical nodes, so as to achieve accurate docking between data content and review nodes. For example, mapping "battery heavy metal content 0.008%" in the sub-data to the "heavy metal content threshold ≤ 0.01%" parameter of the "raw material environmental protection judgment node" is done by: using feature extraction algorithms (such as TF-IDF for text and CNN for images) to extract key information in the sub-data, and establishing the association through semantic mapping tables (such as "heavy metal content" corresponding to "raw material environmental protection node parameter").

[0116] In this embodiment, the matching features are the feature information in the sub-data that is directly related to the logical node parameters. They are the core basis for judging whether the node conditions are met. They include numerical values, keywords, status indicators, etc. For example, the feature that matches the "production energy consumption judgment node" is "unit product power consumption 3.2kWh"; the feature that matches the "packaging material condition node" is "95% of the material is biodegradable".

[0117] The feature matching matrix is ​​a structured representation of the correspondence between matching features and logical nodes in matrix form. The matrix elements represent the matching degree (0-100%) between features and nodes, which is used to intuitively show the degree of coverage of the data to the review nodes, as shown in Table 1.

[0118] Table 1

[0119] Logical Node Matching features (sub-data content) Match Environmental assessment of battery raw materials Heavy metal content: 0.008% 100% Production energy consumption assessment Unit power consumption: 3.2 kWh 90%

[0120] In this embodiment, the labeling decision tree is a hierarchical decision model built on the relationship between nodes. It determines the conformity of matching features and node conditions layer by layer and outputs the final labeling result. It has a tree structure with condition branches and result derivation. For example, the root node in the decision tree is the green attribute compliance of textile products, the first-level non-leaf nodes are the environmental protection of fabric and production pollution discharge indicators, and the leaf nodes are compliant, partially compliant, and non-compliant.

[0121] The root node is the top-level node of the decision tree, representing the overall goal of the audit standard. Specifically, a unique root node is set in the decision tree model, binding the overall goal description of the audit standard (such as conforming to GB / T35601-2017 General Rules for Green Product Evaluation). Leaf nodes are the terminal nodes of the decision tree, representing the final judgment result of the audit. Non-leaf nodes are used to carry logical judgment conditions, and output branch directions (pointing to other non-leaf nodes or leaf nodes) based on the matching features of the input. The logical judgment conditions are the rules used for branch judgment in the non-leaf nodes, which are composed of specific indicator thresholds or state requirements in the audit standard, and are the basis for dividing the branch directions. For example, non-leaf node: the judgment condition for the recyclability of packaging materials is that the proportion of recyclable materials is ≥80% → branch to the compliant direction; <80% → branch to the partially compliant direction.

[0122] In this embodiment, the labeling result is the review conclusion assigned to the sub-data after the decision tree is used for judgment. It includes "compliant", "partially compliant", "non-compliant" and the corresponding explanation (such as compliance indicators and non-compliance items). For example, the sub-data "Battery Raw Material Test Report" is labeled as "compliant, based on: heavy metal content 0.008%≤0.01% threshold" after judgment; the sub-data "Production Energy Consumption Record" is labeled as "partially compliant, based on: unit energy consumption 3.2kWh (compliant), but energy consumption fluctuation coefficient exceeds the standard".

[0123] The review set is a structured collection that integrates the annotation results of all sub-data, as shown in Table 2. It includes sub-data identifiers, corresponding logical node paths, annotation results, and judgment criteria, and serves as the basic data for subsequent comprehensive analysis.

[0124] Table 2

[0125] Sub-data ID Logical node path Annotation results Judgment basis MAT-2024-01 Environmental friendliness of raw materials → Assessment of heavy metal content in batteries conform to 0.008%≤0.01% ELE-2024-02 Production energy consumption → Unit electricity consumption judgment Partially consistent 3.2kWh meets the standard, but the fluctuation coefficient exceeds the standard.

[0126] The beneficial effects of the above technical solution are: by decomposing the execution logic of the audit standard into structured nodes and relationships, and combining feature mapping and decision tree judgment, the intelligent and standardized labeling of sub-data is realized, reducing the bias of manual judgment and ensuring the consistency of the labeling results with the audit standard; at the same time, the structured audit set provides clear and traceable basic data for subsequent comprehensive analysis, improving the accuracy and efficiency of green product audit.

[0127] This invention provides an intelligent green product review and verification method based on a cloud platform for anti-counterfeiting verification, including:

[0128] Based on the permissions of each production stage, determine the production cycle of the corresponding production stage and obtain the production status under the most recent production cycle.

[0129] When the production status matches the verification indicators of the green verification results, a true label is assigned to the corresponding production process.

[0130] Otherwise, a false label is assigned to the corresponding production stage.

[0131] In this embodiment, the production process is an independent procedure or stage divided according to the process of product production. The process permission is the scope of operation permission set for each production process, including data modification permission, parameter adjustment permission, process intervention permission, etc., which is used to standardize process operation and prevent data fraud.

[0132] Within the scope of the process authority, the production cycle refers to the normal operating time of a production process (such as the effective daily production time and batch production cycle). This timeframe is determined by historical production data, equipment capacity, and authority rules, reflecting the routine production rhythm of the process. For example, the process authority for the "chip welding" process stipulates that "the maximum effective daily production time is 12 hours." Based on historical data showing that the average daily production time for this process is 8 hours, its production cycle is determined to be "8:00-16:00 (regular) + 16:00-20:00 (overtime, requiring authority approval)." The production cycle for the "packaging" process is "packaging time per batch of products ≤ 2 hours (based on equipment packaging speed)."

[0133] The production status under the most recent production cycle refers to the actual operating data and status of the production process within the production cycle closest to the audit time, including equipment parameters, material consumption, energy consumption, output, operation records, etc. For example, the most recent production cycle of the "fabric dyeing" process is "2024-06-10 9:00-15:00", and its production status is "dye consumption 120kg, water consumption 500L, dyeing temperature 95℃, output of qualified fabric 300 meters, and no abnormal equipment shutdowns".

[0134] The verification indicators for green verification results are the core judgment criteria set for the production process in the green verification process. These include energy consumption limits, pollutant emission concentrations, material utilization rates, and environmental protection process implementation requirements. They serve as the benchmark for measuring whether a process meets the requirements for green production. For example, the verification indicators for the battery cell packaging process are "energy consumption during packaging ≤ 5 kWh / batch, VOCs content of packaging adhesive ≤ 50 g / L, and yield rate ≥ 98%".

[0135] In this embodiment, the consistency between production status and verification indicators refers to the degree of matching between the actual production status data of the production process and the green verification indicators, including numerical compliance (e.g., energy consumption ≤ limit), process compliance (e.g., using environmentally friendly processes), and complete records (e.g., no missing operation logs). For example, in the "chip welding" process, the production status is "energy consumption 3.2kWh / batch, welding process meets lead-free standards," which is completely consistent with the verification indicator "energy consumption ≤ 4kWh / batch, using lead-free processes." In this case, a genuine label is assigned. In the "packaging" process, the production status is "packaging material degradability rate 90%," while the verification indicator is "≥95%," which is inconsistent. In this case, a fake label is assigned.

[0136] The beneficial effects of the above technical solution are: by controlling the production cycle through process permissions, and by combining the consistency verification of production status and green verification indicators, true / false labels are accurately assigned, thereby realizing anti-counterfeiting and traceability of production process data, improving the accuracy of false label identification, providing real and reliable production process data support for the subsequent construction of verification sets, and ensuring the objectivity of green product review from the source.

[0137] This invention provides an intelligent green product review and verification method based on a cloud platform. The method involves comprehensive analysis of the review set and the verification set to obtain the output result of the product to be reviewed, including:

[0138] The audit set is divided into static features and dynamic features, and the verification set is divided into process features and time sequence features.

[0139] Extract the first feature containing non-conformities from the static and dynamic features, and extract the second feature belonging to pseudo-labels from the link features and time-series features to determine the co-occurrence risk vector;

[0140] Based on the remaining features of the static and dynamic features, and the remaining features based on the link features and time sequence features, a defect risk vector is determined.

[0141] The co-occurrence risk vector and the defect risk vector are input into the two-vector analytical model to obtain the output results of the product to be audited.

[0142] In this embodiment, static features are fixed attributes or results that do not change dynamically with the audit process. They reflect the inherent characteristics of the data itself or a one-time judgment result. For example, the qualification number of the test report (such as SGS-2024-XXX), the validity period of the environmental certification of raw materials (2024.01-2026.01), and the exceeding value of indicators in the non-conformity items (such as heavy metal content of 0.012% (standard ≤0.01%)).

[0143] Dynamic characteristics are attributes or results that change during the review process or over time, reflecting the status changes of materials during the review or the differences in multiple judgments. For example, the number of times sub-materials are reviewed (e.g., 3 submissions, the first 2 were returned due to incomplete information), the change record of the labeling results (initially judged as 'non-compliant,' corrected to 'partially compliant' after supplementing materials), and the update of the reviewers' review comments (initially approved, but calculation errors were found during the review).

[0144] The characteristics of a process are those that reflect the inherent attributes or static verification results of a single production process. They are directly related to the green indicators of a specific process, such as the energy consumption value of the welding process (3.2 kWh / batch), the pollutant emission concentration of the electroplating process (COD 70 mg / L), and the anti-counterfeiting label (genuine label) of that process.

[0145] Time-series characteristics are verification results or trend characteristics that reflect the dynamic changes of multiple production links over time, reflecting the temporal correlation of the production process. For example, the energy consumption time-series curve of a batch of products in the three links of raw material input → processing → assembly (8:00 energy consumption 2kWh → 9:00 energy consumption 5kWh → 10:00 energy consumption 3kWh), and the frequency of the appearance of counterfeit labels in the packaging link in five consecutive production batches (appearing in the 2nd and 4th batches).

[0146] In this embodiment, the first feature is the non-compliance-related feature extracted from the static and dynamic features of the audit set, which reflects the clear non-compliance issues at the data level. For example, the heavy metal content of raw materials exceeds the standard (0.012%) in the static feature, and the test report still fails to pass the audit after three revisions in the dynamic feature.

[0147] The second feature is the feature related to the fake label extracted from the process feature and time series feature of the validation set. It reflects the data fraud or non-compliance issues at the production level. For example, fake labels in the painting process in the process feature (corresponding to abnormal emission data) and time series feature of three consecutive processes with fake labels appearing in the time series feature.

[0148] The co-occurrence risk vector is a vector formed by integrating the first feature and the second feature according to their risk correlation. It is used to quantify the risk level of the co-occurrence of non-conformities at the data level and false labels at the production level. For example, the co-occurrence risk vector of a certain product is [raw material exceeding the standard (first feature), false label in the welding process (second feature), risk correlation 0.8], which reflects the high risk of the two co-occurring. Specifically, the first feature and the second feature are merged by using vector splicing technology (such as feature dimension superposition), and the co-occurrence probability is calculated by using association rule algorithm (such as Apriori) to generate a risk vector (including feature value and correlation).

[0149] The remaining features are those features remaining after removing the first and second features from the audit set and the verification set. These include conforming features and neutral features that are not marked as risk. For example, in static features, the production license is valid; in dynamic features, the audit passed on the first attempt; in process features, the assembly process has a genuine label (energy consumption meets the standard); and in time-series features, the energy consumption fluctuations of each process are normal.

[0150] The defect risk vector is a vector formed from potential defect features extracted from the remaining features (such as indicators that meet the criteria but are close to the threshold). It reflects the hidden risks at the data or production level. For example, the production energy consumption in the remaining features meets the criteria but is close to the upper limit (standard 5kWh / unit, actual 4.8kWh / unit). A certain link is truly labeled, but the energy consumption time series fluctuation coefficient is high (0.3, close to the warning value of 0.35).

[0151] The dual-vector parsing model is an intelligent model used to fuse co-occurrence risk vectors and defect risk vectors, analyze the comprehensive risk level through algorithms, and output the review results. It has the functions of risk weight allocation and feature correlation analysis. Specifically, it is built based on a deep learning framework (such as TensorFlow). The input layer receives dual vectors, the hidden layer assigns higher weights to high-risk features through an attention mechanism (co-occurrence risk weight 0.6, defect risk 0.4), and the output layer outputs the comprehensive score and result label.

[0152] The output is the final conclusion of the dual-vector analytical model for the green attribute review of the product, including "pass", "fail", "conditionally pass" and the corresponding risk description (such as high-risk items, potential defect improvement suggestions). For example, if it fails, the reason is: the co-occurrence risk vector contains 2 high-risk items (excessive raw materials + false labeling), and the comprehensive score is 65 points < 70 points; if it is conditionally pass, the problem of 'energy consumption close to the upper limit' in the defect risk vector needs to be rectified.

[0153] The beneficial effects of the above technical solution are as follows: by splitting the features of the audit set and the verification set and extracting risk-related features, a dual vector of co-occurrence risk and defect risk is constructed. Combined with the analytical model, a comprehensive assessment of explicit and implicit risks is achieved, which significantly improves the comprehensiveness and accuracy of risk identification in green product audits, provides scientific and quantitative decision-making basis for output results, and reduces manual analysis costs and improves audit efficiency.

[0154] This invention provides an intelligent green product review and verification method based on a cloud platform. Before obtaining the uploaded materials based on each review specification in the green review specification process from the cloud platform, the method further includes:

[0155] When a third party uploads initial data to the cloud platform, the data types involved in the initial data are traversed and recorded, and an encrypted structure is created for each data type based on the green attribute. The data types are pure image type, pure text type, and mixed image and text type.

[0156] The system obtains the environmental difference between the location environment of the data file based on the third-party storage location and the preset environment. When the environmental difference is greater than the preset difference, a dual detection mechanism is activated to perform content detection and integrity detection on the data file, and a new file is constructed and saved.

[0157] Otherwise, retain the aforementioned data file;

[0158] The transmission path to the third-party sending interface is determined based on the storage location of each file.

[0159] The input and output trajectory logs of each path node and its adjacent nodes in the transmission path are retrieved from the historical database, and the input and output trajectory logs are analyzed and a trajectory matrix is ​​constructed based on the analysis indicators consistent with the input and output.

[0160] The feature vectors of the trajectory matrix are obtained and input into the vector analysis model to obtain the safety coefficients of the corresponding path nodes;

[0161] Select the smallest security factor from all security factors under the transmission path as the reference factor;

[0162] When the storage location of the corresponding retained file is a separate location, the corresponding reference coefficient is assigned to the corresponding retained file. Here, the separate location means that there is only one retained file on the longest path to which the corresponding retained file belongs.

[0163] When the storage location of the corresponding retain file is multiple locations, the reference coefficients are sorted sequentially from front to back to obtain a coefficient array. If the first coefficient in the coefficient array is the minimum coefficient, the minimum coefficient is assigned to the retain file at each of the multiple locations. If the last coefficient in the coefficient array is the minimum coefficient, the original reference coefficients at the multiple locations are kept unchanged and assigned to the corresponding retain file. Otherwise, the minimum coefficient in the coefficient array is locked, and the reference coefficients before the minimum coefficient are kept unchanged. The reference coefficients after the minimum coefficient are replaced with the corresponding minimum coefficient and assigned to the corresponding retain file.

[0164] Determine whether each stored file has undergone third-party encryption. If so, establish a structural mapping relationship between the encryption operation and the encryption structure, as well as an attribute mapping relationship between the basic attributes of the stored file and the structural attributes of the encryption structure. Combined with the assigned coefficients, perform a second encryption operation on the data file.

[0165] If it does not exist, perform a new encryption operation on the data file according to the encryption structure and the corresponding assigned coefficient;

[0166] The encrypted files are stored on a cloud platform.

[0167] In this embodiment, "third party" refers to the entity that uploads relevant information about the product to be reviewed to the cloud platform, including product manufacturers, testing institutions, supply chain partners, and other non-reviewing parties or non-platform operators.

[0168] Initial data refers to the original data files that a third party uploads to the cloud platform for the first time without any processing by the platform. It is the basic data source for the audit.

[0169] Data types are categories based on the content format of the initial data, used to adapt to different encryption strategies, including pure image type, pure text type, and mixed image and text type.

[0170] The encryption structure is a standardized encryption framework designed for different data types and in conjunction with the green attributes of products. It includes encryption algorithms, key length, data segmentation methods, etc., to ensure the confidentiality and integrity of data.

[0171] Specifically: For pure image types: a hash digest + AES-256 encryption structure is used (the hash is used for integrity verification, AES is used for content encryption, and the key length is adjusted according to the sensitivity level of the green attribute).

[0172] Plain text type: Employs RSA asymmetric encryption + digital signature structure (RSA encrypts the content, and the signature prevents tampering).

[0173] Image and text hybrid type: It adopts a hybrid structure of AES encryption for image segments, SM4 encryption for text segments, and overall checksum.

[0174] In this embodiment, the location environment of the third-party storage location is the physical environment (such as the temperature and humidity of the storage device) or network environment (such as the IP address location, network bandwidth, and access device security level) of the initial data stored by the third party.

[0175] In this embodiment, the preset environment refers to the standard environmental parameters preset by the cloud platform to ensure the security of data storage and transmission, including physical environment thresholds (such as temperature 25±5℃, humidity 40%-60%) and network environment benchmarks (such as trusted IP list, bandwidth fluctuation ≤10%).

[0176] Environmental difference refers to the degree of parameter deviation between the location environment of a third-party storage site and the preset environment. It is measured by quantitative indicators (such as temperature difference and bandwidth fluctuation difference). Specifically, a weighted summation algorithm is used to calculate the comprehensive difference value (e.g., physical environment weight 0.3, network environment weight 0.7). The formula is "Environmental difference = Σ(|actual value - preset value| / preset value × weight)".

[0177] The preset difference is an environmental difference threshold set by the cloud platform to determine whether to activate the dual detection mechanism. It is usually dynamically adjusted according to the data sensitivity level (the threshold is lower for sensitive data).

[0178] The dual-detection mechanism performs two layers of checks on data files when environmental differences exceed preset limits. These checks include content verification (file content compliance) and integrity verification (file completeness). Specifically:

[0179] Content detection: Using NLP technology to identify false data and non-compliant indicators in plain text (such as emission concentration of 100mg / L actually exceeding the standard but labeled as compliant); using image recognition to detect tampering traces in plain images (such as Photoshop modification of test report values);

[0180] Complete inspection: Verify the file hash value (compare with the original hash provided by a third party; a deviation of ≤1% is considered complete), and check whether there are missing pages or data truncation in mixed text and image files.

[0181] Creating a new file involves repairing (e.g., adding missing pages) and correcting (e.g., removing infringing content) a document that meets the platform's requirements when double-checking reveals content violations or incompleteness in the original document. For example, if the original mixed text and image document is missing the stamp page of the detection agency, it can be obtained by sending a request to a third party and then added to form a new document; if a plain text document contains false data, it can be removed and the true parts can be retained to form a new document.

[0182] The retained files are those that have been tested for environmental differences. These include the original data files when the environmental difference is less than or equal to the preset difference, or the new files that have been repaired after double testing when the environmental difference is greater than or equal to the preset difference.

[0183] The transmission path is the chain of network nodes that a file passes through when it is transferred from a third-party storage location to a third-party sending interface on the cloud platform. This includes intermediate nodes such as routers, switches, and servers. For example, the path is: Enterprise local server → Regional router → Cloud platform edge node → Cloud platform sending interface (node ​​chain: S1 → R1 → E1 → I1).

[0184] Input / output trajectory logs are data stream records of each node (such as a router or server) in the transmission path as it receives (input) and sends (output) data to the keep-alive file. These records include data volume, timestamps, and node status (such as whether there was packet loss). For example, node R1's log might read "Input data volume 2MB (10:05:00), Output data volume 2MB (10:05:02), No packet loss"; node E1's log might read "Input data volume 1.9MB (10:05:03), Output data volume 1.9MB (10:05:05), Packet loss rate 1%".

[0185] Analysis metrics consistent with input and output are used to evaluate the quantitative indicators of the consistency of input and output data at nodes along the transmission path, reflecting the reliability of node data transmission.

[0186] For example, data integrity (difference between input and output data volume ≤ 5%), time synchronization (difference between input and output time ≤ 10s), and packet loss rate (≤ 2%).

[0187] The trajectory matrix is ​​a matrix that structures the input and output trajectory logs of each node in the transmission path according to the analysis indicators. The rows represent nodes, the columns represent indicators, and the elements are the quantitative values ​​of the indicators, as shown in Table 3.

[0188] Table 3

[0189] node Data integrity (%) Time synchronization (s) Packet loss rate (%) S1 100 2 0 R1 98 3 0 E1 95 5 1

[0190] The eigenvectors of the trajectory matrix are key features extracted from the trajectory matrix that reflect the overall security of the transmission path, such as the average data integrity of nodes, the maximum packet loss rate, and the standard deviation of time synchronization. The resulting eigenvectors are, for example, [average data integrity 97.7%, maximum packet loss rate 1%, standard deviation of time synchronization 1.2s].

[0191] The vector analysis model is a machine learning model used to analyze the feature vectors of the trajectory matrix and output the safety level of the path nodes. The model input is the feature vector and the output is a safety coefficient of 0-100 (the higher the value, the stronger the safety). It is pre-trained. For example, the input feature vector is [97.7%, 1%, 1.2s] and the output node E1 has a safety coefficient of 82.

[0192] The reference coefficient is the minimum value selected from the security coefficients of all nodes in the transmission path. It represents the security of the weakest node in the transmission path and is used as a benchmark for subsequent encryption strength adjustment. For example, if the security coefficients of the nodes in the transmission path are [S1:95, R1:90, E1:75, I1:85], the reference coefficient is 75 (the security coefficient of E1).

[0193] A unique location is a storage location where a file exists only once on the longest transmission path to which it belongs (i.e., the file transmission path is unique and there are no other file sharing nodes). For example, if the longest transmission path of a test report is "S1→R1→I1", and only this report is transmitted on this path (no other files), then its storage location is a unique location.

[0194] Multiple locations mean that there are multiple storage locations for the file on the longest transmission path to which the file belongs (i.e., multiple files share some transmission nodes). For example, two files, "Energy Consumption Record Table" and "Emission Detection Report", are transmitted on the longest transmission path "S1→R1→I1" at the same time, and both of them are stored in multiple locations.

[0195] In a multi-location scenario, the coefficient array is formed by arranging the reference coefficients corresponding to each file in the transmission path in the transmission order (from front to back). For example, if the reference coefficients of two files in multiple locations are 75 (front) and 80 (back), the coefficient array is [75, 80].

[0196] Third-party encryption refers to the encryption processing performed on files by a third party before uploading them (such as internal enterprise encryption), including encryption algorithms (such as RSA) and key types (such as symmetric / asymmetric).

[0197] The structure mapping relationship is the correspondence between the encryption operations performed by the third party (such as algorithm and key length) and the encryption structure created by the cloud platform, ensuring compatibility between the two encryptions. For example, if the third party uses "AES-128" encryption and the cloud platform's encryption structure is "AES-256", the structure mapping relationship is "algorithm compatible (AES), key length upgraded (128→256)".

[0198] The attribute mapping relationship between basic attributes and structural attributes maintains the correspondence between the inherent attributes of a file (such as file size, format, and creation time) and the attributes of the cloud platform's encryption structure (such as key length, encryption block size, and validity period). This is used to match encryption parameters. For example, the mapping between the basic file attributes "size 50MB, PDF format" and the encryption structure attributes "key length 256 bits (suitable for large files), encryption block size 4KB (suitable for PDF segmentation)".

[0199] The re-encryption operation is performed when a third party has already encrypted the file. The cloud platform performs a second encryption on the file based on the structural mapping relationship and attribute mapping relationship, combined with the reference coefficient (to enhance security). For example, if the third party encrypts the file with AES-128 and the reference coefficient is 75 (medium security level), the platform will encrypt it again with AES-256 based on the mapping relationship (the key length increases with the reference coefficient) and bind the reference coefficient as the encryption salt value.

[0200] The new encryption operation is the first encryption performed by the cloud platform on the file based on the preset encryption structure and the reference coefficient when the third party has not encrypted the file. For example, for a plain image file that has not been encrypted by the third party, with a reference coefficient of 60 (low security level), the platform uses the "hash + AES-256" encryption structure (increasing the key strength when the reference coefficient is low) to perform the new encryption.

[0201] The beneficial effects of the above technical solution are as follows: by dynamically creating encrypted structures for different data types, and combining environmental difference detection and transmission path security analysis to adapt encryption strategies, the end-to-end security control of data from upload to storage is achieved. This ensures the integrity and compliance of third-party data, and improves the adaptability and security of encryption through layered encryption (combined with the security coefficient of the weakest node), thereby enhancing the anti-tampering capability of cloud platform data storage and providing a reliable encrypted data foundation for subsequent audits.

[0202] This invention provides an intelligent green product review and verification method based on a cloud platform, which, after obtaining the output result of the product to be reviewed, further includes:

[0203] The output results of all products to be reviewed are classified according to the green attribute, and abnormal and normal features are extracted from the process set of each review specification in the classification results to determine the first occurrence state of different abnormal features and the second occurrence state of different normal features.

[0204] Based on the first occurrence state, the second occurrence state and the corresponding review specifications, hidden associations are mined, and content not covered by the current review specifications is listed as initial items;

[0205] Determine the expansion probability of each initial item, and regard initial items with expansion probabilities greater than preset probabilities as expansion items, so as to update the corresponding audit specifications.

[0206] In this embodiment, the classification is based on the similarity of the green attributes of the products, dividing the output results into several groups to ensure that the green attribute characteristics of products within the same group are highly consistent, which facilitates subsequent feature analysis. For example, plastic products with "degradability ≥ 80% and energy consumption ≤ 3kWh / piece" are classified into the "high environmental protection plastic group"; electronic devices with "heavy metal content ≤ 0.01% and standby power consumption ≤ 0.5W" are classified into the "low pollution electronic group". Specifically, the K-means clustering algorithm is used, with the green attribute feature vector as input, and the optimal number of clusters (such as 5-8 groups) is determined according to the principle of maximizing the silhouette coefficient, and classification labels are generated.

[0207] The process set of an audit specification is a collection of all relevant data generated during the audit process for a specific audit specification. This includes audit steps, sub-data annotation results, production process verification records, and anomaly handling trajectories. For example, the process set of a raw material environmental compliance audit specification includes raw material test report annotation results (compliant / non-compliant), supplier qualification verification records, heavy metal content detection deviation values, and auditor review opinions.

[0208] In this embodiment, the abnormal features are characteristics extracted from the audit specification process set that reflect the product's non-compliance with audit requirements. These include explicit non-compliance items, abnormal data fluctuations, and pseudo-label association items. For example, the abnormal features in the "Production Energy Consumption Audit Specification" process set are "Unit energy consumption exceeds the standard for 3 consecutive batches (4.2kWh / piece > 4kWh / piece)" and "Energy consumption record does not match the production log timestamp (pseudo-label association)".

[0209] In this embodiment, normal characteristics are stable characteristics extracted from the audit specification process that reflect the product's compliance with audit requirements. These include compliance indicators, process compliance records, and trend data of continuous compliance. For example, the normal characteristics in the "pollutant emission audit specification" process set are "the exhaust gas emission concentration is stable at 50mg / m³ for 10 consecutive batches (standard ≤60mg / m³)" and "each batch is accompanied by a third-party testing report (process compliance)".

[0210] In this embodiment, the first occurrence state is the pattern of occurrence of abnormal features in the audit specification process, including the frequency of occurrence (e.g., a certain abnormal feature appears 20 times in 100 audits), the stage of occurrence (e.g., concentrated in the "raw material inspection" stage), and the associated product type (e.g., only appears in plastic products). For example, the first occurrence state of the abnormal feature of packaging material recyclability is "occurring 15 times (accounting for 10% of the total number of audits), concentrated in the 'finished packaging' stage, and mainly associated with paper packaging products".

[0211] In this embodiment, the second occurrence state is the regularity of the occurrence of normal features in the audit specification process, including the stable frequency of occurrence (such as a normal feature appearing in 90% of the audits), the fluctuation range of indicators (such as energy consumption values ​​concentrated in 2.5-3kWh / piece), and the universality across product types (such as the "low standby power consumption" feature that exists in both electronic and mechanical products). For example, the second occurrence state of the normal feature of meeting the production water consumption standard is "occurring 85 times (accounting for 70% of the total audits), water consumption value fluctuation ≤ ±0.3L / piece, and stable occurrence in both textile and chemical products".

[0212] In this embodiment, hidden association mining involves analyzing the potential connections between the first occurrence state, the second occurrence state, and the corresponding audit specifications to discover patterns or dependencies that are not explicitly recorded. Specifically, association rule algorithms (such as Apriori) are used to calculate the support (≥15%) and confidence (≥60%) between feature states and audit specification entries, and to filter hidden relationships with high correlation. For example, mining reveals a hidden association between "electronic device standby power consumption exceeds the standard (abnormal feature)" and "charger materials are not environmentally friendly (not covered by the current specification)" (positive correlation in frequency of occurrence, correlation coefficient 0.7); "degradable materials meet the standard (normal feature)" and "humidity control in the production workshop is between 40% and 60% (not covered by the specification)" have a strong correlation.

[0213] Initial items are features or indicators discovered during hidden correlation mining that are not covered by current audit standards. In other words, they are potential audit points that are not included in existing standards but are closely related to the green attributes of products. For example, the current "Electronic Equipment Energy Consumption Audit Standard" does not include "Environmental friendliness of charger materials", but mining has found that it is strongly correlated with excessive standby power consumption, so "Environmental friendliness of charger materials" is listed as an initial item; the current "Paper Packaging Audit Standard" does not include "VOCs content of printing inks", but it is closely related to the compliance of biodegradability, so it is listed as an initial item.

[0214] The expansion probability is the likelihood that an initial item will be included in the corresponding review specifications. It is calculated based on its relevance to the green attribute, industry universality, and technical detectability (value ranges from 0 to 100%). Specifically, the Analytic Hierarchy Process (AHP) is used to assign scores for relevance (weight 0.4), universality (weight 0.3), and detectability (weight 0.3), and the expansion probability is calculated by weighting these scores.

[0215] The preset probability is the threshold value for determining whether an initial item can be included in the review specifications as an extended item. It is dynamically adjusted according to the industry type (e.g., the preset probability for high-pollution industries is reduced to 60%, and environmental protection-related items are given priority).

[0216] Extended items are initial items that satisfy the condition that the extended probability is greater than or equal to the preset probability. They have the necessity and feasibility to be included in the review specifications and can supplement the coverage of existing specifications. For example, "environmental friendliness of charger materials" and "VOC content of printing inks" with an extended probability of 85% are listed as extended items.

[0217] The update of the audit specifications involves integrating the extended items into the corresponding audit specifications, supplementing or revising existing clauses to make the specifications more aligned with the actual green attributes of products and the latest environmental protection requirements of the industry. For example, the "Electronic Equipment Energy Consumption Audit Specification" has added the clause "Charger materials must comply with RoHS 2.0 environmental requirements"; the "Paper Packaging Audit Specification" has added the indicator "Printing ink VOCs content ≤ 50g / L". Specifically, the extended items are reviewed by the Audit Specification Management Committee, and after approval, the specification text is revised and updated synchronously to the Attribute-Specification Database, indicating the update time and basis.

[0218] The beneficial effects of the above technical solution are: by classifying product output results according to green attributes, we can explore the hidden feature correlations in the review process, identify and dynamically update the extended items not covered by existing standards, so that the review standards can continuously adapt to the new features of the green attributes of products and the environmental protection requirements of the industry, improve the comprehensiveness and timeliness of the standards, reduce the review omissions caused by the lag in standards, and provide more accurate standard basis for green product review.

[0219] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A cloud-based intelligent green product auditing and verification method, characterized in that, include: Step 1: Based on the green attributes of the product to be reviewed and in conjunction with the attribute-specification database, obtain the green review specification process, and retrieve the uploaded materials based on each review specification in the green review specification process from the cloud platform; Before retrieving the uploaded materials based on each review standard in the green review process from the cloud platform, the process also includes: When a third party uploads initial data to the cloud platform, the data types involved in the initial data are traversed and recorded, and an encrypted structure is created for each data type based on the green attribute. The data types are pure image type, pure text type, and mixed image and text type. The system obtains the environmental differences between the location environment of the data file based on the third-party storage location and the preset environment. When the environmental difference is greater than the preset difference, a dual detection mechanism is activated to perform content detection and integrity detection on the data file, and a new file is constructed and saved. Otherwise, retain the aforementioned data file; The transmission path to the third-party sending interface is determined based on the storage location of each file. The input and output trajectory logs of each path node and its adjacent nodes in the transmission path are retrieved from the historical database, and the input and output trajectory logs are analyzed and a trajectory matrix is ​​constructed based on the analysis indicators consistent with the input and output. The feature vectors of the trajectory matrix are obtained and input into the vector analysis model to obtain the safety coefficients of the corresponding path nodes; Select the smallest security factor from all security factors under the transmission path as the reference factor; When the storage location of the corresponding retained file is a separate location, the corresponding reference coefficient is assigned to the corresponding retained file. Here, the separate location means that there is only one retained file on the longest path to which the corresponding retained file belongs. When the storage location of the corresponding retain file is multiple locations, the reference coefficients are sorted sequentially from front to back to obtain a coefficient array. If the first coefficient in the coefficient array is the minimum coefficient, the minimum coefficient is assigned to the retain file at each of the multiple locations. If the last coefficient in the coefficient array is the minimum coefficient, the original reference coefficients at the multiple locations are kept unchanged and assigned to the corresponding retain file. Otherwise, the minimum coefficient in the coefficient array is locked, and the reference coefficients before the minimum coefficient are kept unchanged. The reference coefficients after the minimum coefficient are replaced with the corresponding minimum coefficient and assigned to the corresponding retain file. Determine whether each stored file has undergone third-party encryption. If so, establish a structural mapping relationship between the encryption operation and the encryption structure, as well as an attribute mapping relationship between the basic attributes of the stored file and the structural attributes of the encryption structure. Combined with the assigned coefficients, perform a second encryption operation on the data file. If it does not exist, perform a new encryption operation on the data file according to the encryption structure and the corresponding assigned coefficient; Store the encrypted files on the cloud platform; Step 2: Match the uploaded materials with each review condition in the review specifications, and perform standard conversion on the corresponding sub-materials according to the material requirements of each review condition. Then, mark the converted sub-materials according to the execution logic of the review standards of the review conditions to obtain the review set. Step 3: Perform green verification and anti-counterfeiting verification on each production stage of the product to be audited to obtain a verification set; Step 4: Perform a comprehensive analysis on the audit set and the verification set to obtain the output results of the product to be audited.

2. The intelligent green product review and verification method based on a cloud platform according to claim 1, characterized in that, Based on the green attributes of the products to be reviewed and in conjunction with the attribute-specification database, the green review process is derived, including: Construct a feature vector for the green attribute of the product to be reviewed, wherein the feature vector includes a first-level dimension parameter and a second-level dimension parameter; Based on the feature vector, a hierarchical retrieval is performed on the attribute-norm database, and the association relationship between norms at each layer is constructed through knowledge graph technology. The hierarchical retrieval includes: basic norm layer retrieval, industry norm layer retrieval, and dynamic norm layer retrieval. Based on the full lifecycle timeline logic of the product to be reviewed, the initial conditions for review under each timeline logic are extracted, and an initial review standard process is generated. Based on the aforementioned relationships and the product scale of the products to be reviewed, the initial review process is adjusted to obtain a green review process.

3. The intelligent green product review and verification method based on a cloud platform according to claim 1, characterized in that, Based on the document requirements for each audit condition, the corresponding sub-documents will be standardized and converted, including: Based on the review criteria, a dynamic conversion rule base is generated for the required materials. At the same time, the metadata and content features of the sub-materials are analyzed using natural language processing technology to extract material features. The data features are matched with a dynamic conversion rule base to generate converted sub-data.

4. The intelligent green product review and verification method based on a cloud platform according to claim 1, characterized in that, The converted sub-data is labeled according to the execution logic of the audit criteria described above, resulting in an audit set, including: The execution logic of the audit standard is broken down into logical nodes and node relationships, wherein the logical nodes include: judgment nodes, jump nodes and / or condition nodes; The transformed sub-data is feature-mapped, and matching features with logical nodes are extracted to generate a feature matching matrix; A labeling decision tree is established based on node relationships, and decisions are made based on the feature matching matrix. All labeling results are integrated to form an audit set. The root node of the labeling decision tree is the overall audit standard target, the leaf nodes are specific labeling results, and the non-leaf nodes are logical judgment conditions.

5. The intelligent green product review and verification method based on a cloud platform according to claim 1, characterized in that, Anti-counterfeiting verification includes: Based on the permissions of each production stage, determine the production cycle of the corresponding production stage and obtain the production status under the most recent production cycle. When the production status matches the verification indicators of the green verification results, a true label is assigned to the corresponding production process. Otherwise, a false label is assigned to the corresponding production stage.

6. The intelligent green product review and verification method based on a cloud platform according to claim 1, characterized in that, A comprehensive analysis of the audit set and the verification set yields the output results of the product to be audited, including: The audit set is divided into static features and dynamic features, and the verification set is divided into process features and time sequence features. Extract the first feature containing non-conformities from the static and dynamic features, and extract the second feature belonging to pseudo-labels from the link features and time-series features to determine the co-occurrence risk vector; Based on the remaining features of the static and dynamic features, and the remaining features based on the link features and time sequence features, a defect risk vector is determined. The co-occurrence risk vector and the defect risk vector are input into the two-vector analytical model to obtain the output results of the product to be audited.

7. The intelligent green product auditing and verification method based on a cloud platform according to claim 1, characterized in that, After obtaining the output results of the product to be reviewed, the following are also included: The output results of all products to be reviewed are classified according to the green attribute, and abnormal and normal features are extracted from the process set of each review specification in the classification results to determine the first occurrence state of different abnormal features and the second occurrence state of different normal features. Based on the first occurrence state, the second occurrence state and the corresponding review specifications, hidden associations are mined, and content not covered by the current review specifications is listed as initial items; Determine the expansion probability of each initial item, and regard initial items with expansion probabilities greater than preset probabilities as expansion items, so as to update the corresponding audit specifications.

Images