Quality evaluation method, device and system for product carbon footprint data and medium

Abstract

The invention relates to the technical field of data processing, and provides a product carbon footprint data quality evaluation method, device and system and a medium. According to the implementation scheme, multi-source heterogeneous carbon footprint data of a product is analyzed to obtain all target entities and the relation between all the target entities; on the basis of all the target entities and the relation between all the target entities, a knowledge graph is constructed by constructing an intelligent agent, and a carbon footprint knowledge graph is obtained; evaluating each target path in the carbon footprint knowledge graph based on an evaluation agent to obtain abnormal carbon footprint data; diagnosing the abnormal carbon footprint data based on a diagnosis agent to obtain a diagnosis report; and determining a quality evaluation result of the product based on the abnormal carbon footprint data and the diagnosis report. According to the embodiment of the invention, the knowledge graph is constructed and the multiple agents are driven to carry out anomaly evaluation and autonomous diagnosis, so that comprehensive and explainable automatic evaluation of the product carbon footprint data quality is realized.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, system and medium for quality evaluation of product carbon footprint data. Background Technology

[0002] With the increasing demands for carbon emission accounting and green supply chain management, product carbon footprint accounting has gradually become a key step for enterprises to assess their environmental impact and meet the audit requirements of their supply chain partners. However, existing methods for evaluating the quality of carbon footprint data still have significant limitations.

[0003] On the one hand, traditional data quality verification typically relies on manual sampling or pre-set rules for superficial checks, such as determining whether data is missing or whether values ​​are within a reasonable range. Such surface-level indicator-based verification methods fail to address the complex business semantics and physical constraints inherent in product carbon footprint accounting, resulting in insufficient comprehensiveness of the evaluation results.

[0004] On the other hand, current in-depth assessments of carbon footprint data quality heavily rely on the experience and judgment of industry experts. Experts need to manually analyze supply chain structures, production processes, and accounting methods, checking the logical relationships and business consistency of each data point. This manual assessment process, which relies on experience and knowledge, is not only time-consuming, labor-intensive, and inefficient, but also difficult to scale and reuse, further limiting the development of carbon footprint data quality assessment towards automation and intelligence.

[0005] Therefore, how to break through the shallow capabilities of traditional rule-based validation and reduce over-reliance on expert knowledge to achieve a comprehensive, interpretable, and automated assessment of carbon footprint data quality is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This invention provides a method, apparatus, system, and medium for quality evaluation of product carbon footprint data, which can solve at least one of the above-mentioned technical problems.

[0007] In a first aspect, embodiments of the present invention provide a method for quality evaluation of product carbon footprint data, including: The multi-source heterogeneous carbon footprint data of the product is analyzed to obtain the various target entities and the relationships between them. Based on the target entities and the relationships between them, a knowledge graph is constructed by building intelligent agents to obtain a carbon footprint knowledge graph; The evaluation agent assesses each target path in the carbon footprint knowledge graph to obtain abnormal carbon footprint data. A diagnostic report is obtained by diagnosing the abnormal carbon footprint data based on a diagnostic agent; Based on the abnormal carbon footprint data and the diagnostic report, the quality evaluation result of the product is determined.

[0008] Secondly, embodiments of the present invention provide a quality evaluation device for product carbon footprint data, comprising: The data parsing module is used to parse the multi-source heterogeneous carbon footprint data of the product to obtain the various target entities and the relationships between them. A smart agent module is used to construct a knowledge graph based on each of the target entities and the relationships between them, thereby obtaining a carbon footprint knowledge graph. The evaluation agent module is used to evaluate each target path in the carbon footprint knowledge graph based on the evaluation agent, and obtain abnormal carbon footprint data; The diagnostic agent module is used to diagnose the abnormal carbon footprint data based on the diagnostic agent and obtain a diagnostic report; The quality evaluation result determination module is used to determine the quality evaluation result of the product based on the abnormal carbon footprint data and the diagnostic report.

[0009] Thirdly, embodiments of the present invention also provide a quality evaluation system for product carbon footprint data, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described in any one of the embodiments of the present invention.

[0010] Fourthly, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the method described in any one of the embodiments of the present invention.

[0011] This invention employs a technical solution that analyzes multi-source heterogeneous carbon footprint data of a product to obtain various target entities and the relationships between them. Based on this, and using these relationships, a knowledge graph is constructed using intelligent agents to generate a carbon footprint knowledge graph. This allows the product's carbon footprint data to be fully expressed in a graph structure, providing a semantic basis for subsequent evaluation. Subsequently, an evaluation agent performs path-by-path reasoning along the target paths in the carbon footprint knowledge graph to automatically identify potential anomalies. A diagnostic agent then diagnoses the identified anomalous carbon footprint data, outputting a structured diagnostic report to accurately pinpoint the causes of the anomalies and provide explanatory conclusions. Finally, the anomaly detection and diagnosis results are combined for product quality evaluation. Thus, this invention significantly reduces reliance on human experience. By constructing a knowledge graph and driving multiple agents to perform anomaly assessment and autonomous diagnosis, it achieves a comprehensive and interpretable automated evaluation of the product's carbon footprint data quality.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0013] The accompanying drawings are provided for a better understanding of this solution and do not constitute a limitation of the invention. Wherein: Figure 1 This is a flowchart of a product carbon footprint data quality evaluation method according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a product carbon footprint data quality evaluation device according to an embodiment of the present invention; Figure 3 This is a schematic block diagram of an electronic device used to implement the methods of embodiments of the present invention. Detailed Implementation

[0014] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0015] Figure 1 This is a flowchart of a product carbon footprint data quality evaluation method according to an embodiment of the present invention.

[0016] like Figure 1 As shown, the quality assessment method for the product's carbon footprint data may include: S110, analyze the multi-source heterogeneous carbon footprint data of the product to obtain the various target entities and the relationships between them; S120, based on each target entity and the relationships between them, constructs a knowledge graph by building intelligent agents to obtain a carbon footprint knowledge graph; S130: Based on the evaluation agent, each target path in the carbon footprint knowledge graph is evaluated to obtain abnormal carbon footprint data. S140, based on the diagnostic agent, diagnose abnormal carbon footprint data and obtain a diagnostic report; S150 determines the product's quality evaluation results based on abnormal carbon footprint data and diagnostic reports.

[0017] For example, multi-source heterogeneous carbon footprint data refers to a collection of carbon footprint-related data from different data sources, with different structural forms and semantic content. These data sources may include enterprise ERP (Enterprise Resource Planning), PLM (Product Lifecycle Management), supplier ledgers, energy consumption metering systems, lifecycle databases (LCI), bills of materials (BOM), emission factor databases, etc.

[0018] For example, Excel / CSV: bill of materials, process energy consumption; JSON: emission parameters reported by supply chain nodes; PDF: third-party certification reports; database tables: emission factors, energy consumption records; For example, a target entity refers to a key object with independent significance extracted from carbon footprint data, which is the basic node for constructing a knowledge graph.

[0019] For example, products, components, materials, energy types, production processes, emission factors, suppliers, metering equipment, etc.

[0020] For example, the relationship between target entities refers to the directed or undirected semantic connection formed between various target entities based on business logic or physical causality.

[0021] For example, a carbon footprint knowledge graph refers to a graph-structured knowledge base built on target entities and their relationships, used to express the physical links, energy flows, material flows and logical dependencies in the carbon emission accounting process.

[0022] In this example, the nodes and relationships in the carbon footprint knowledge graph may include the following:

[0023] First, nodes: Product: Product, with attributes such as Product ID (identifier), Product Name, and Specifications. Component: Parts / Raw Materials, with attributes such as Material ID, Material Name, and Unit. Process: Production Process, with attributes such as Process ID, Process Name, and Energy Consumption Type. Equipment: Equipment, with attributes such as Equipment ID, Equipment Model, and Rated Power. Supplier: Supplier, with attributes such as Supplier ID and Supplier Name. EmissionSource: Emission Source, with attributes such as Source ID, Emission Type (Direct, Indirect), and GHG Type. DataValue: Data Value, with attributes such as Data ID, Value, Unit, Timestamp, and Data Source. This is a key design feature used to decouple and associate specific numerical values ​​with entities.

[0024] Second, relationships: (Component)-[CONTAINS]→(Component): Indicates the composition relationship of the bill of materials.

[0025] (Product)-[MANUFACTURED_BY]→(Process): The product is manufactured by a certain process.

[0026] (Process)-[CONSUMES]→(Component): Raw materials consumed by the process.

[0027] (Process)-[USES]→(Equipment): Equipment used in a process.

[0028] (Equipment)-[GENERATES]→(EmissionSource): The equipment generates emissions.

[0029] (Supplier)-[PROVIDES]→(Component): Suppliers provide raw materials.

[0030] (DataValue)-[BELONGS_TO]→(Any entity node): The data value belongs to a certain entity.

[0031] (EmissionSource)-[CALCULATED_FROM]→(DataValue): The emission amount is calculated from a certain data value.

[0032] For example, the target path refers to the semantic link in the carbon footprint knowledge graph that extends outward from the product target node and constitutes the carbon emission calculation, consisting of multiple entities and relationships.

[0033] For example, (Model A laptop) - [Assembly process] → (Final assembly workshop) - [Energy consumption] → (Industrial electricity) - [Geographical location] → (Suzhou, China).

[0034] For example, abnormal carbon footprint data refers to data that does not conform to business logic, physical laws, or accounting methods after the evaluation agent infers the target path of the map.

[0035] For example, the energy consumption of the process does not match the maximum power of the equipment; the material emission factor is missing or is an incorrect version; the emission value of the supply chain node deviates significantly from the industry reference range; the path link is broken (missing node / missing factor).

[0036] For example, a diagnostic report refers to a structured explanatory document generated by a diagnostic agent based on abnormal carbon footprint data, which includes an analysis of the causes of the anomaly, the scope of impact, and possible remediation suggestions.

[0037] For example, the anomaly type is: Emission factor version error; the affected path is: Process C → Power → Emission factor; the reason is that the supplier reported the 2024 version, but the system requires the 2022 version; the suggested solution is to replace it with the 2022 version.

[0038] For example, the quality evaluation result refers to the product carbon footprint data quality evaluation conclusion obtained by combining abnormal carbon footprint data and diagnostic reports.

[0039] In this example, the product quality evaluation results are obtained by encapsulating the abnormal carbon footprint data and diagnostic report and outputting them in JSON data format (JavaScript object notation).

[0040] Understandably, for each product, steps S110-S150 in the aforementioned example can be followed to obtain quality evaluation results for multiple products.

[0041] For example, constructing an intelligent agent refers to an intelligent processing unit that performs operations such as entity extraction, relationship recognition, entity attribute completion, and graph structure generation based on various target entities and their relationships in multi-source heterogeneous carbon footprint data, in order to automatically construct a carbon footprint knowledge graph.

[0042] For example, the evaluation agent refers to an intelligent processing unit that, based on each target path from the product node to the emission source node in the carbon footprint knowledge graph, applies integrity evaluation templates, logical consistency evaluation templates, and compliance evaluation templates to conduct data integrity, logical consistency, and accounting compliance reviews on each target path, in order to identify defective carbon footprint data and output abnormal carbon footprint data.

[0043] For example, a diagnostic agent refers to an intelligent processing unit that, based on abnormal carbon footprint data, combines credibility evaluation templates, sensitivity analysis templates, and uncertainty attribution templates to perform credibility analysis, sensitivity analysis, and uncertainty attribution analysis on the abnormal carbon footprint data, thereby generating a diagnostic report.

[0044] This can be understood as building, evaluating, and diagnosing intelligent agents, which are program entities or intelligent processing units built upon a large model. These three entities can interact and collaborate through a unified data interface and process control mechanism.

[0045] Specifically, the construction agent organizes the target entities, relationships, and attributes of multi-source heterogeneous carbon footprint data into a structured carbon footprint knowledge graph, and transmits the generated graph content to the evaluation agent in real time. Based on the analysis results of the graph's target paths, the evaluation agent identifies and outputs abnormal carbon footprint data, and then provides this abnormal data as input to the diagnostic agent. The diagnostic agent further performs credibility analysis, sensitivity analysis, and uncertainty tracing on the abnormal data, and sends the diagnostic conclusions back for subsequent quality assessment or knowledge graph updates. Through the bidirectional coupling of the above data flow and instruction flow, these three agents can achieve continuous relay-style automated work in the same processing link, thereby constructing a self-checking, self-diagnosing, and self-optimizing intelligent carbon footprint processing system.

[0046] According to the above implementation method, firstly, multi-source heterogeneous carbon footprint data is parsed to uniformly generate target entities and their relationships; then, an agent-driven carbon footprint knowledge graph is constructed to achieve a structured expression of production processes, supply chain relationships, and emission logic; based on this, the evaluation agent reviews the critical paths in the graph and identifies potentially abnormal carbon footprint data; the diagnostic agent further analyzes the causes of the anomalies and generates interpretable diagnostic reports; finally, based on the anomaly information and diagnostic conclusions, a product quality evaluation result is formed. This significantly improves the automation and accuracy of carbon footprint data quality evaluation, enabling structured understanding, anomaly identification, and interpretable diagnosis of complex carbon emission data, thereby improving accounting quality, reducing reliance on manual labor, and enhancing the transparency and credibility of the evaluation process.

[0047] In one implementation, the multi-source heterogeneous carbon footprint data of a product is parsed to obtain various target entities and the relationships between them. This includes: automatically acquiring the multi-source heterogeneous carbon footprint data of the product through a pre-defined connector in response to a trigger event for acquiring carbon footprint data, wherein the multi-source heterogeneous carbon footprint data includes structured and unstructured data; for structured data, mapping each field in the structured data to each node and each node's attribute using predefined mapping rules; determining the relationships between nodes based on foreign key relationships in the structured data; determining the relationships between second entities based on each node, each node's attribute, and the relationships between nodes; for unstructured data, performing text extraction to obtain unstructured text data; performing named entity recognition on the unstructured text data to obtain various third entities; extracting relationships between each third entity to obtain the relationships between them; determining each target entity based on each second entity and each third entity; and determining the relationships between each target entity based on the relationships between the second entities and the relationships between the third entities.

[0048] For example, structured data refers to data organized with predefined fields, table structures, or record formats, where there are fixed patterns between data items that can be processed directly in a row-column manner.

[0049] For example, a product's production log table contains fields such as batch number, energy consumption value, equipment number, start and stop time, etc. These fields are fixed and can be read directly.

[0050] For example, unstructured data refers to data that is not organized in a fixed structure and cannot be directly expressed through table fields. Information needs to be extracted through natural language processing or pattern recognition methods. For instance, a passage in a supplier's raw material transportation report stating, "Due to traffic congestion during the transportation of this batch, the low-temperature storage time was extended," is unstructured text.

[0051] For example, the triggering events include: event-driven: listening to the message queue and immediately triggering ingestion when the upstream data source publishes a data update event; timed polling: for data sources that do not support event push, actively fetching incremental data at a preset period (such as every 24 hours).

[0052] For example, a mapping rule refers to a predefined set of transformation rules used to map structured data fields to nodes or attributes in a knowledge graph stored in a graph database. For instance, a mapping rule specifies that the field "EnergyConsumption" is mapped to the node attribute "CarbonEmission"; and the field "DeviceID" is mapped to "DeviceEntity".

[0053] For example, the second entity refers to the basic entity of the knowledge graph built based on structured data, which usually comes from nodes after mapping and foreign key association. For example, entities mapped from the production record table include "Production Process Entity", "Equipment Entity", "Energy Record Entity", etc.

[0054] In this example, upon receiving a trigger event for carbon footprint data collection, the system automatically accesses the Production Execution System, Energy Metering System, and Equipment Management System via pre-defined connectors to retrieve the corresponding structured carbon footprint data in batches. Subsequently, according to predefined mapping rules, the fields in the structured data tables are converted one by one into nodes and node attributes in a pre-stored knowledge graph in the graph database. For example, the field "Energy Consumption Value" in the structured table "Energy Consumption Record Table" is mapped to a node attribute, and the field "Equipment Number" is mapped to a device node, generating an "Energy Consumption Record Node". Then, based on foreign key relationships in the structured data, such as "Equipment Number" as a foreign key pointing to the Equipment Information Table, the system automatically generates a "Energy Consumption Record Node - Usage - Equipment Node" relationship. For example, the field "Process Number" in the Production Batch Record Table serves as a foreign key pointing to the "Process Table," and the system establishes a "Production Batch Node - Contains - Process Node" relationship accordingly, thus constructing a second entity and its relational network based on the structured data.

[0055] For example, a third entity refers to an entity object identified from unstructured text through named entity recognition, typically representing semantic objects such as events, equipment, materials, and locations. For instance, the entities extracted from the text "the furnace experienced a short-term shutdown" include "furnace" and "short-term shutdown," which are third entities.

[0056] In this example, the system performs text extraction on the collected unstructured data, converting natural language descriptions, maintenance logs, supplier quality inspection reports, and other content into processable text input. Then, the system calls a pre-trained Named Entity Recognition (NER) model to perform entity recognition on the text, extracting third entities including equipment names, event types, time expressions, locations, and material names. For example, from the sentence "The smelting furnace experienced a short-term shutdown at 1 AM, which may have resulted in insufficient temperature," the system extracts entities such as "smelting furnace," "short-term shutdown," and "insufficient temperature." After recognition, the system extracts relationships between the third entities, identifying causal, temporal, attribute, or action relationships. For example, it identifies the relationship "'smelting furnace' - occurred - 'short-term shutdown', 'short-term shutdown' - resulted in - 'insufficient temperature'," thus obtaining the semantic association structure between the third entities.

[0057] For example, after obtaining a second entity based on structured data and a third entity based on unstructured data, the two are further aligned to determine each target entity. Entity alignment can be based on information such as entity name, semantic embedding vector, timestamp, or device number. For instance, the "device entity (Device_102)" in the structured data and the "furnace" extracted from the text are compared semantically and by number to determine that they belong to the same device, thus merging them into a unified target entity "furnace device entity". After entity fusion, the system uses all aligned target entities as entity nodes in the final knowledge graph.

[0058] For example, based on the relationships between second entities (e.g., "production batch - includes - process" and "process - use - equipment") and the semantic relationships between third entities (e.g., "equipment - occurs - event" and "event - leads to - anomaly"), the system models the two types of relationship structures in a unified manner according to a preset relationship fusion strategy and generates the final relationship between the target entities. For example, when the structured data records "production batch A - use - smelting furnace 102" and the unstructured data contains "smelting furnace short-term shutdown", the system generates the relationship node "production batch A - affected - short-term shutdown event" accordingly, thereby making the entity relationships of the knowledge graph more complete and forming a target knowledge graph structure that can support comprehensive carbon footprint diagnosis.

[0059] According to the above implementation method, multi-source heterogeneous carbon footprint data of products is automatically acquired by triggering events. Structured and unstructured data are then parsed separately: structured data is transformed into nodes and their attributes through predefined mapping rules, and relationships between nodes are constructed based on foreign key relationships; unstructured data, after text extraction, obtains entities and their relationships through named entity recognition and relationship extraction. Subsequently, entities from both structured and unstructured sources are uniformly integrated to form target entities and their associated relationships, ultimately providing a complete, accurate, and semantically consistent data foundation for subsequent knowledge graph construction. The advantages of this approach are that it can automatically integrate multi-source data, significantly reduce manual processing costs, and ensure comprehensive coverage, semantic consistency, and structural standardization of entities and relationships, providing a high-quality data foundation for carbon footprint knowledge graph construction and subsequent intelligent analysis.

[0060] In one implementation, a carbon footprint knowledge graph is constructed by building an intelligent agent based on each target entity and the relationships between them. This includes: pairing each target entity to obtain each entity pair to be verified; for each entity pair to be verified, filling the entity pair and the attributes of the two target entities in the entity pair into placeholders in a preset first prompt word template to generate a similarity inference instruction; using the similarity inference instruction as input to the intelligent agent to obtain the matching result of the entity pair to be verified; if the matching result indicates that the two target entities in the entity pair to be verified belong to the same category, then performing rule verification on the entity pair to be verified based on a preset domain rule base to obtain the rule verification result of the entity pair to be verified; updating each target entity based on the rule verification result of each entity pair to obtain each updated target entity; and constructing the carbon footprint knowledge graph based on each updated target entity and each relationship.

[0061] For example, each target entity is randomly paired in pairs to obtain each entity pair to be verified.

[0062] For example, the first prompt word template can be specifically designed as the following text structure: "Please determine whether the following two sets of target entities describe the same entity. Please make a comprehensive inference from the dimensions of name, specifications, attribute fields, context source and semantic logic, and give a clear judgment result (yes / no) and reasoning."

[0063] Data A: {Name: «name_A», Specification: «spec_A», Attribute: «attr_A», Source: «source_A», Context: «context_A»}; Data B: {Name: «name_B», Specification: «spec_B», Attribute: «attr_B», Source: «source_B», Context: «context_B»}; Please output your judgment (yes / no) and the reasoning. «name_A», «spec_A», «attr_A», etc., are placeholders used to automatically populate the fields of the target entity pair (i.e., the target entity and its attributes) at runtime.

[0064] For example, when the entity pairs to be verified are: E1: {Name: "Copper Material", Specification: "T2", Source: "Supplier A Quotation"}; E2: {Name: "Electrolytic Copper", Specification: "T2", Source: "ERP Inventory List"}.

[0065] The system will automatically populate its fields into the first prompt word template, thus generating a similarity inference instruction: "Please determine whether the following two sets of data describe the same entity. Please make a comprehensive inference based on dimensions such as name, specifications, attribute fields, contextual source, and semantic logic."

[0066] Entity E1: {Name: "Copper Material", Specification: "T2", Attribute: "-", Source: "Supplier A Quotation", Context: "Supply Chain Procurement Record"}; Data E2: {Name: "Electrolytic Copper", Specification: "T2", Attribute: "-", Source: "ERP Inventory Table", Context: "Inventory Material Management Record"}; Please output your judgment (yes / no) and the reasoning behind it. An intelligent agent can then perform consistent judgments and output reasoning results based on this standardized template.

[0067] For example, after obtaining the entity pair to be verified, the system extracts the name field, specification field, source field and context path from the entity pair respectively, and fills these contents into the corresponding placeholders in the preset first prompt word template to generate a similarity reasoning instruction with structured instruction semantics.

[0068] For example, when the intelligent agent provides matching results for two entities of the same category, the system will further call the preset domain rule base to verify the domain knowledge constraints of the entity pair, so as to avoid the large model making erroneous merging when it has not mastered the hard rules of the domain.

[0069] For example, the rule base contains professional knowledge such as mutually exclusive rules for material categories, emission factor constraint rules, and life cycle stage constraint rules. For instance, even if the agent determines that "primary aluminum" and "recycled aluminum" are semantically similar, the domain rule base records that the corresponding material category (material_type) attributes of the two are mutually exclusive and cannot be equivalent. Therefore, the rule verification will give the rule verification conclusion of "merging is not allowed" with the following explanation: "Primary aluminum and recycled aluminum belong to materials in different life cycle stages, have different carbon footprint boundaries, and cannot be treated as the same entity."

[0070] This can be understood as follows: if the rule validation gives a rule validation conclusion of "merging is not allowed" and an additional explanation, it means that the material type attributes of the two target entities in the entity pair to be validated are not the same; if the rule validation gives a rule validation conclusion of "merging is allowed", it means that the material type attributes of the two target entities in the entity pair to be validated are the same.

[0071] For example, after obtaining all updated target entities, the system reconstructs the structured relationships between entities, between entities and emission factors, between entities and production processes, and between entities and supply chain nodes based on the semantic definitions in the domain knowledge. These relationships are then written into triples of nodes, attributes, and relationships according to a unified graph database data model, forming a structurally consistent and semantically unified carbon footprint knowledge graph. For instance, after merging the "copper material - electrolytic copper" entity, the merged entity automatically re-establishes connections with the "smelting process node," "emission factor EF_Cu_01," and "supply chain node S01" associated in the original structured data, creating a complete material flow and carbon emission chain in the knowledge graph. This lays the foundation for subsequent integrity checks, logical consistency assessments, and carbon footprint diagnosis.

[0072] According to the above implementation method, by pairing target entities and generating similarity inference instructions using prompt word templates, the constructed agent can automatically determine whether entities refer to the same object. After determining that the categories are consistent, a rigorous rule verification is performed using a domain rule base, thereby avoiding erroneous matching caused by relying solely on text similarity. The target entities are updated based on the rule verification results, and a more accurate carbon footprint knowledge graph is finally constructed based on the updated entities and relationships. This ensures that the entity alignment process is both intelligent and professional, significantly improving the accuracy, consistency, and credibility of the knowledge graph construction, and providing a more reliable data foundation for subsequent carbon footprint analysis.

[0073] In one implementation, each target entity is updated based on the rule verification results of each entity pair to be verified to obtain each updated target entity, including: if the rule verification result shows that the material type attributes of the two target entities in the entity pair to be verified are the same, then the two target entities in the entity pair to be verified are merged to obtain a merged target entity; if the rule verification result shows that the material type attributes of the two target entities in the entity pair to be verified are different, then the two target entities in the target entity pair are respectively used as first entities; and each updated target entity is determined based on each merged target entity and each first entity.

[0074] For example, the system obtains entity pair E1 and E2 to be verified, with the following attributes: E1: Name: "Copper"; Material Type: "Metal-Copper"; Specification: "T2"; Source: "Supplier A Quotation"; E2: Name: "Electrolytic Copper"; Material Type: "Metal-Copper"; Specification: "T2"; Source: "ERP Inventory List"; the intelligent agent constructs a matching conclusion of "belonging to the same category" based on the similarity reasoning instruction; the system then enters the rule verification stage. Since the domain rule base states: "If the material type attribute of two entities is the same, they are considered to be physically identical, and entity merging can be performed." After reading the material types of the two entities, the system finds that both are "Metal-Copper", therefore the material types are identical, and the rule verification passes. In this case, the system merges E1 and E2 to generate a merged target entity E12, with the following attribute examples: Name: "Copper / Electrolytic Copper (Fused)"; Material Type: "Metal-Copper"; Specification: "T2"; Source: {"Supplier A Quotation", "ERP Inventory List"} (multiple records are allowed); Confidence: Assigned by the fusion rules, for example, 0.93; The merged entity E12 will be included in the "Updated Target Entities" set for subsequent construction of the carbon footprint knowledge graph.

[0075] For example, the system obtains another pair of entities to be verified, E3 and E4, with the following attributes: E3: Name: "Resin Board"; Material Type: "Polymer"; Specification: "R1"; Source: "Supplier B Quality Inspection Report"; E4: Name: "Composite Board"; Material Type: "Metal-Aluminum"; Specification: "R1"; Source: "Production Execution System (MES)".

[0076] The system constructs an agent reasoning system that the two entities are "of the same category" (e.g., both are "sheet material"). However, further rule verification reveals that their material type attributes are: E3 → polymer; E4 → aluminum. According to the rule in the domain rule base: "If the material type attributes of two entities are not the same, they cannot be merged and should remain independent as first entities." Therefore, the system determines that the two cannot be merged and retains E3 → first entity 1 and E4 → first entity 2 as independent target entities.

[0077] For example, based on the two cases in the aforementioned example, the final updated set of target entities is as follows: merged target entity: E12 (made of copper material and electrolytic copper); first entity 1: E3 (resin board); first entity 2: E4 (composite board). These entities will then be used together with the extracted relations to construct a carbon footprint knowledge graph.

[0078] According to the above implementation method, the system determines whether the entity material types are consistent based on the rule verification results. Entities with consistent material types are automatically merged into a more complete target entity, while entities with inconsistent material types remain independent, thus forming the final updated target entities. Through this strategy of "merging consistency and retaining differences," the system can effectively avoid redundancy caused by duplicate entities and ensure that entities of different material types are not mistakenly merged. This improves the accuracy of entity representation in the knowledge graph, enhances the clarity and usability of the structure, and provides a higher quality data foundation for subsequent carbon footprint modeling and inference.

[0079] In one implementation, after constructing the carbon footprint knowledge graph based on each updated target entity and each relation, the method further includes: determining a text summary based on each updated target entity, the attributes of each updated target entity, and each relation; generating a graph evolution decision instruction based on the text summary, newly added knowledge triples, and a preset evolution decision prompt template; and using the graph evolution decision instruction as input to construct the intelligent agent to obtain a graph update operation instruction set, so that the graph database updates the carbon footprint knowledge graph according to each instruction in the graph update operation instruction set.

[0080] For example, the updated target entities, their attributes, and their relationships are collectively used as a text summary.

[0081] For example, an evolutionary decision prompt template can be as follows: "[Existing knowledge graph text summary]: {current_graph_summary};" [New Knowledge]: {new_triples}; As a knowledge graph evolution decision expert, please complete the following tasks: 1. Determine which entities and relationships need to be newly created in the new knowledge; 2. Determine which attributes of existing entities need to be updated; 3. Determine if there is any knowledge conflicting with the existing map, and provide a solution strategy; 4. Output Cypher operation commands used for actual writing to the graph database.

[0082] Please return directly to the text block containing only Cypher instructions. Here, {current_graph_summary} represents placeholders used to populate the text summary; {new_triples} represents placeholders used to populate newly added knowledge triplets.

[0083] For example, the graph evolution decision instruction refers to the natural language reasoning input text formed after filling the evolution decision prompt template. This input text is directly input into the agent to trigger the agent to generate a graph update operation instruction set.

[0084] For example, “[Existing Knowledge Graph Text Summary]:” Component comp_001 (type:Cable, supplier: sup_abc); Supplier sup_abc(location:CN); [New Knowledge]: ("ZR-YJV-0.6 / 1kV 3*150", type, "Cable"); ("sup_xyz", provides,"ZR-YJV-0.6 / 1kV 3*150"); Please complete the reasoning as a knowledge graph evolution decision expert... (The rest of the content is the same as the evolution decision prompt template in the previous example)".

[0085] For example, the graph update operation instruction set refers to a set of Cypher update statements that can be directly executed in the graph database, output by the agent based on the graph evolution decision instructions.

[0086] For example, MERGE (c:Component {id:'comp_002'}); SET c.name='ZR-YJV-0.6 / 1kV 3*150', c.type='Cable'; MERGE (s:Supplier {id:'sup_xyz'}); SET s.location='Unknown'; MERGE (s)-[r:PROVIDES]→(c); SET r.create_time=date().

[0087] For example, the graph database receives and parses a set of graph update operation instructions to update the carbon footprint knowledge graph according to the individual instructions therein.

[0088] According to the above implementation method, a text summary of the knowledge graph is generated using the updated target entities and their attribute relationships. This summary is then combined with newly added knowledge triples and a preset evolutionary decision prompt template to construct graph evolutionary decision instructions. The constructing agent generates an executable set of graph update operation instructions, thereby driving the graph database to automatically update the knowledge graph. This allows the agent to understand the impact of new knowledge on the graph in a structured way and automatically generate correct update instructions, achieving continuous self-evolution of the knowledge graph. This reduces manual maintenance costs while ensuring that the graph content remains up-to-date, complete, and consistent.

[0089] In one implementation, an evaluation agent assesses each target path in the carbon footprint knowledge graph to obtain abnormal carbon footprint data. This includes: traversing the carbon footprint knowledge graph along preset semantic relationships using a graph traversal algorithm, starting from a product node and ending at an emission source node, to obtain each target path; wherein the graph traversal algorithm includes depth-first search and breadth-first search algorithms; encapsulating the node sequence and relationship sequence in each target path to generate a natural language document; generating an integrity evaluation instruction based on the natural language document and a preset integrity evaluation template; generating a consistency evaluation instruction based on the natural language document and a preset logical consistency evaluation template; generating a compliance evaluation instruction based on the natural language document and a preset compliance evaluation template; and using the integrity evaluation instruction, consistency evaluation instruction, and compliance evaluation instruction as input to the evaluation agent pair to obtain abnormal carbon footprint data.

[0090] For example, in a real-world carbon footprint analysis scenario, the system first uses a product node as the starting point for traversal, such as "electric vehicle battery pack," and an emission source node as the ending point, such as "carbon emissions from aluminum electrolysis." Based on predefined semantic relationships (e.g., "composed of," "originates from," "processed from," "generates emissions," etc.), the system performs path retrieval using a graph traversal algorithm within the carbon footprint knowledge graph.

[0091] When using a depth-first search algorithm, the system will explore the deepest level of the relationship chain sequentially, for example, from "battery pack" → "cell" → "cathode material" → "aluminum foil" → "aluminum ingot" → "electrolytic aluminum emission source", to obtain a complete target path. When using a breadth-first search algorithm, the system will expand all reachable nodes at the same level from the starting point, for example, from "battery pack" to "cell", "structural components", "BMS (battery management system) module", etc., and then descend layer by layer until it finds the link pointing to "electrolytic aluminum emission source", thus obtaining another target path.

[0092] After obtaining the target path, the system encapsulates the node sequence (e.g., "battery pack → cell → cathode material → aluminum foil → aluminum ingot → emission source") and its corresponding relational sequence (e.g., "composed of... → manufactured by... → produced by... → smelted by... → emission generation") into a natural language document. For example, the corresponding natural language document can be generated as: "The battery pack contains cells, and the cathode material used is made of aluminum foil, which comes from aluminum ingots. Aluminum ingots generate carbon emissions during the electrolytic smelting process. Therefore, the emission path of this product is: battery pack—cell—cathode material—aluminum foil—aluminum ingot—emission source of electrolytic process."

[0093] For example, an integrity evaluation template may look like this: Role: You are a data quality audit expert.

[0094] Task: Assess the data integrity of the provided carbon footprint knowledge graph.

[0095] Spectral information: {NLDOC}; Please focus on checking the following: 1. Is the supply chain path from the product to all raw materials and energy consumption complete, and are there any breaks? 2. Are any key entities missing necessary attributes (such as quantity, emission factor)? 3. Are the accounting boundaries complete, and are there any omissions of emission sources? Please output all integrity defects and indicate the corresponding entity or path.

[0096] Here, {NLDOC} represents a placeholder for a natural language document.

[0097] For example, an integrity evaluation instruction refers to the specific input content generated by the system after filling the placeholders in the integrity evaluation template with natural language documents. This instruction is used to drive the evaluation agent to perform integrity checks on knowledge graph paths, attributes, and emission boundaries.

[0098] For example, “Role: You are a data quality audit expert.”

[0099] Task: Assess the data integrity of the provided carbon footprint knowledge graph.

[0100] Information from the diagram: The battery pack is composed of cells, which are made from positive electrode materials. The positive electrode materials are made from aluminum foil, which is produced from aluminum ingots. However, no information on the upstream energy consumption of aluminum ingots was found in the supply chain.

[0101] Please focus on checking the following: 1. Is the supply chain path from the product to all raw materials and energy consumption complete, and are there any breaks? 2. Are any key entities missing necessary attributes? 3. Are the accounting boundaries complete, and are there any omissions of emission sources? Please output all integrity defects and indicate the corresponding entity or path.

[0102] For example, a logical consistency evaluation template can be as follows: Role: You are a logic verification expert.

[0103] Task: Discover logical inconsistencies in the carbon footprint knowledge graph.

[0104] Spectral information: {NLDOC}; Please focus on checking the following: 1. Are the input and output quantities of the same material logically reasonable? 2. Are the attribute descriptions of the same entity consistent across different data sources? 3. Is there any data that violates common sense (such as a negative value for "transportation distance")? Please list all logical contradictions and explain why.

[0105] Here, {NLDOC} represents a placeholder for a natural language document.

[0106] For example, a consistency evaluation instruction refers to the specific content generated by the system after filling the natural language document into the logical consistency evaluation template, which is used to drive the evaluation agent to identify problems such as attribute conflicts, numerical contradictions, or unreasonable unit conversions.

[0107] For example, “Role: You are a logic verification expert.”

[0108] Task: Discover logical inconsistencies in the carbon footprint knowledge graph.

[0109] Information from the graph: The input of positive electrode material is 200kg, but the output is recorded as 150kg; in addition, the copper content of "electrolytic copper" in the inventory system is 99.8%, while the copper content of the same type of copper material in the supplier's quotation is 92%, which is a conflict.

[0110] Please focus on checking the following: 1. Are the input and output figures logically sound? 2. Are there any contradictions in the attribute descriptions of the same entity? 3. Are there any numerical records that violate common sense? Please list all the contradictions and explain why.

[0111] For example, a compliance assessment template can be as follows: Role: You are a carbon footprint accounting compliance expert, familiar with ISO 14067.

[0112] Task: Evaluate whether the spectral data meets the accounting standards.

[0113] Spectral information: {NLDOC}; Please check the following against the standards: 1. Is the classification of emission sources accurate? 2. Are the sources and time ranges of emission factors compliant? 3. Are the data records timely and traceable? Please list all compliance risk points.

[0114] For example, a compliance evaluation instruction refers to the specific audit content obtained after the system injects a natural language document into the compliance evaluation template. This instruction is used to drive the evaluation agent to judge the compliance with emission factors, scope classifications, and standard requirements.

[0115] For example, the emission data shows that the electrolytic aluminum process is marked as Scope 3 emissions, but it should be classified as Scope 1; the emission factors used in the aluminum foil manufacturing process are from publicly available data from 2012, which do not match the current accounting year (2024). Mapping this emission data to the {NLDOC} (placeholder) in the compliance assessment template will generate compliance assessment instructions.

[0116] For example, abnormal carbon footprint data refers to the collection of various abnormal data results output by the evaluation agent based on the integrity evaluation command, the consistency evaluation command, and the compliance evaluation command, including issues such as path breakage, missing attributes, numerical conflicts, inconsistent units, and non-compliant emission factors. This data is used for subsequent credibility evaluation, sensitivity analysis, and uncertainty tracing.

[0117] According to the above implementation method, by performing graph traversal in the knowledge graph starting from product nodes and ending at emission source nodes, various target paths that can fully reflect the carbon footprint formation chain are obtained. The nodes and relation sequences in the paths are encapsulated into natural language documents that can be understood by large models. Subsequently, three types of evaluation instructions—completeness, logical consistency, and compliance—are generated based on different evaluation templates and input into the evaluation agent, thereby automatically identifying abnormal carbon footprint data in the paths that are missing, conflicting, or do not meet the standard requirements. In this way, the system can conduct a comprehensive, in-depth, and interpretable quality review of the carbon footprint chain in complex supply chains without manual verification, significantly improving the coverage and accuracy of anomaly detection.

[0118] In one implementation, a diagnostic agent diagnoses abnormal carbon footprint data to obtain a diagnostic report, including: generating a credibility evaluation instruction based on the abnormal carbon footprint data and a preset credibility evaluation template; generating a sensitivity analysis instruction based on a carbon footprint knowledge graph and a preset sensitivity analysis template; generating an uncertainty tracing instruction based on the abnormal carbon footprint data and a preset uncertainty tracing template; and using the credibility evaluation instruction, sensitivity analysis instruction, and uncertainty tracing instruction as inputs to the diagnostic agent to obtain the diagnostic report.

[0119] For example, a credibility evaluation template can be as follows: "As a seasoned carbon footprint verification expert, please assess the credibility of the following calculation results:" Calculation background: {diagnosis_context}; Please evaluate from four dimensions: 1. Data Quality: How complete, accurate, and timely is the key data? 2. Methodological Compliance: Does the accounting method comply with the requirements of the {accounting_standard} standard? 3. Logical soundness: Does the result align with industry benchmarks and historical trends? 4. Uncertainty: What is the extent to which the uncertainty of the main parameters affects the results? Please provide an overall credibility score (0-10 points), along with a detailed explanation of the scoring rationale, key risk points, and improvement suggestions.

[0120] Here, {diagnosis_context} represents a placeholder for anomalous carbon footprint data.

[0121] For example, the credibility evaluation instruction refers to the structured prompt text formed by filling the credibility evaluation template with abnormal carbon footprint data, which can be directly input into the diagnostic agent to trigger the large model to generate credibility diagnostic content.

[0122] In this example, abnormal carbon footprint data, such as "Abnormalities: Product A's range three emissions are 42% higher than the industry benchmark; the source of key emission factors is unknown; some energy consumption data timestamps are missing," can be used to replace the credibility evaluation template {diagnosis_context} with this abnormal carbon footprint data to obtain the credibility evaluation instruction.

[0123] For example, a sensitivity analysis template can be as follows: Sensitivity analysis is conducted based on the dependencies in the accounting process. Computation dependency graph: {computation_dependencies}; Please analyze: 1. Which input parameters have the greatest impact on the final carbon footprint? Please rank them.

[0124] If the key parameters (emission factors, activity data) change by ±10%, how will the results change? Identify the top three carbon hotspots and rank them by their impact, explaining the basis for your judgment.

[0125] Here, {computation_dependencies} represents the edges (i.e. the relationships between entities) in the carbon footprint knowledge graph.

[0126] In this example, replacing each relation corresponding to each edge in the carbon footprint knowledge graph with {computation_dependencies} will yield the uncertainty tracing instruction.

[0127] For example, an uncertainty tracing template can be as follows: "Uncertainty source information:" Data source credibility: {data_sources_credibility}; Parameter selection rationality: {parameter_selection_rationale}; Model assumptions applicability: {model_assumptions_applicability}; Please identify: 1. From which stages does the greatest uncertainty originate? 2. What is the possible range of uncertainty? 3. How can uncertainty be reduced by improving data quality? 4. Which uncertainties are systemically unavoidable?

[0128] Among them, {data_sources_credibility}, {parameter_selection_rationale}, and {model_assumptions_applicability} are all placeholders for anomalous carbon footprint data in the uncertainty tracing template.

[0129] For example, the uncertainty tracing instruction refers to the prompt text formed after filling the uncertainty-related information corresponding to the abnormal carbon footprint data into the uncertainty tracing template, which is used to trigger the large model to output uncertainty attribution analysis.

[0130] In this example, uncertainty-related information, such as "Data source credibility: Some activity data comes from supplier self-reporting, which is difficult to verify; Parameter selection rationality: The copper emission factor is cited from a 2018 study, which is not timely enough; Model assumption applicability: It assumes that all transportation is fully loaded, which may deviate from the actual situation," can be filled into the placeholders in the uncertainty tracing template to obtain the uncertainty tracing instructions.

[0131] For example, by using the credibility evaluation command, sensitivity analysis command, and uncertainty tracing command as inputs to the diagnostic agent, a diagnostic report can be obtained that includes credibility score, risk point description, sensitivity ranking, carbon hotspots, main sources of uncertainty, and actionable improvement suggestions.

[0132] According to the above implementation method, by filling abnormal carbon footprint data into preset templates for credibility evaluation, sensitivity analysis, and uncertainty tracing, three types of instructions targeting different diagnostic dimensions are generated. These instructions are then input into the diagnostic agent to automatically generate diagnostic reports. The diagnostic agent can perform in-depth analysis of abnormal data from multiple perspectives, including overall credibility, sensitivity of key parameters, and sources of uncertainty in the data and model, thereby outputting structured and interpretable diagnostic results. This achieves multi-dimensional intelligent diagnosis of abnormal carbon footprint data, not only improving the comprehensiveness and accuracy of the diagnosis but also significantly reducing reliance on human expert experience, making the quality assessment process more efficient and transparent.

[0133] Figure 2 This is a structural block diagram of a product carbon footprint data quality evaluation device according to an embodiment of the present invention.

[0134] like Figure 2 As shown, the quality assessment device for the product's carbon footprint data may include: The data parsing module 510 is used to parse the multi-source heterogeneous carbon footprint data of the product to obtain the various target entities and the relationships between the various target entities; The intelligent agent module 520 is used to construct a knowledge graph based on each of the target entities and the relationships between the target entities, thereby obtaining a carbon footprint knowledge graph. Evaluation agent module 530 is used to evaluate each target path in the carbon footprint knowledge graph based on the evaluation agent to obtain abnormal carbon footprint data; The diagnostic agent module 540 is used to diagnose the abnormal carbon footprint data based on the diagnostic agent and obtain a diagnostic report; The quality evaluation result determination module 550 is used to determine the quality evaluation result of the product based on the abnormal carbon footprint data and the diagnostic report.

[0135] In one implementation, the intelligent agent construction module includes: The pairing unit is used to pair each of the target entities to obtain each pair of entities to be verified. The similarity reasoning instruction generation unit is used to fill the attributes of the entity pair to be verified and the two target entities in the entity pair to be verified into the placeholders in the preset first prompt word template for each of the entity pairs to be verified, thereby generating a similarity reasoning instruction. A similarity reasoning unit is used to take the similarity reasoning instruction as input to the constructed agent and obtain the matching result of the entity pair to be verified; The rule verification unit is used to perform rule verification on the entity pair to be verified based on a preset domain rule library if the matching result is that the two target entities in the entity pair to be verified have the same category, and obtain the rule verification result of the entity pair to be verified. The entity update unit is used to update each target entity based on the rule verification results of each of the entity pairs to be verified, so as to obtain each updated target entity. The graph construction unit is used to construct the carbon footprint knowledge graph based on each of the updated target entities and each of the relationships.

[0136] In one implementation, the entity update unit includes: The merging subunit is used to merge the two target entities in the entity pair to be verified if the rule verification result shows that the material type attributes of the two target entities in the entity pair to be verified are the same, so as to obtain the merged target entity. As a sub-unit, if the rule verification result is that the material type attributes of the two target entities in the entity pair to be verified are different, then the two target entities in the target entity pair are respectively regarded as the first entity; An update subunit is used to determine each of the updated target entities based on each of the merged target entities and each of the first entities.

[0137] In one implementation, after the atlas construction unit, the system further includes: The text summary determination subunit is used to determine a text summary based on each of the updated target entities, the attributes of each of the updated target entities, and each of the relationships. The graph evolution decision instruction generation subunit is used to generate graph evolution decision instructions based on the text summary, the newly added knowledge triplet and the preset evolution decision prompt template. The graph update subunit is used to take the graph evolution decision instructions as input to the constructing agent to obtain a graph update operation instruction set, so that the graph database updates the carbon footprint knowledge graph according to each instruction in the graph update operation instruction set.

[0138] In one implementation, the evaluation agent module includes: The traversal unit is used to traverse the carbon footprint knowledge graph along preset semantic relationships, starting from the product node and ending at the emission source node, to obtain each of the target paths; wherein, the graph traversal algorithm includes a depth-first search algorithm and a breadth-first search algorithm; The encapsulation unit is used to encapsulate the node sequence and relation sequence in each of the target paths to generate a natural language document; The integrity evaluation instruction generation unit is used to generate integrity evaluation instructions based on the natural language document and the preset integrity evaluation template. A consistency evaluation instruction generation unit is used to generate a consistency evaluation instruction based on the natural language document and a preset logical consistency evaluation template. The compliance evaluation instruction generation unit is used to generate compliance evaluation instructions based on the natural language document and the preset compliance evaluation template. An evaluation unit is used to take the integrity evaluation instruction, the consistency evaluation instruction, and the regularity evaluation instruction as inputs to the evaluation agent pair, respectively, to obtain the abnormal carbon footprint data.

[0139] In one embodiment, the diagnostic agent module includes: A credibility evaluation instruction generation unit is used to generate a credibility evaluation instruction based on the abnormal carbon footprint data and a preset credibility evaluation template. The sensitivity analysis instruction generation unit is used to generate sensitivity analysis instructions based on the carbon footprint knowledge graph and a preset sensitivity analysis template. An uncertainty tracing instruction generation unit is used to generate uncertainty tracing instructions based on the abnormal carbon footprint data and a preset uncertainty tracing template. The diagnostic unit is used to take the credibility evaluation instruction, the sensitivity analysis instruction, and the uncertainty tracing instruction as inputs to the diagnostic agent to obtain the diagnostic report.

[0140] In one embodiment, the data parsing module includes: The data acquisition unit is used to automatically acquire multi-source heterogeneous carbon footprint data of the product through a preset connector in response to a trigger event for acquiring carbon footprint data. The multi-source heterogeneous carbon footprint data includes structured data and unstructured data. The structured data unit is used to map each field in the structured data to each node and each node's attribute using predefined mapping rules; determine the relationship between each node based on each foreign key relationship in the structured data; and determine the relationship between each second entity based on each node, each node's attribute, and the relationship between each node. An unstructured data unit is used to extract text from the unstructured data to obtain unstructured text data; perform named entity recognition on the unstructured text data to obtain various third entities; and extract relationships from the various third entities to obtain the relationships between the various third entities. The target entity determination unit is configured to determine each target entity based on each of the second entities and each of the third entities; The relationship determination unit is used to determine the relationship between the target entities based on the relationship between the second entities and the relationship between the third entities.

[0141] The specific functions and examples of each module and submodule of the system in this embodiment of the invention can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0142] The acquisition, storage, and application of user personal information involved in the technical solution of this invention all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0143] This invention also provides a quality evaluation system for product carbon footprint data, comprising: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method described in any one of the embodiments of the present invention.

[0144] The beneficial effects of the product carbon footprint data quality evaluation system in this embodiment of the invention are equivalent to the beneficial effects of the product carbon footprint data quality evaluation method described above, and will not be repeated here.

[0145] This invention also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the method described in any one of the embodiments of this invention.

[0146] The beneficial effects of the storage medium of the present invention are equivalent to the beneficial effects of the above-mentioned method for quality evaluation of product carbon footprint data, and will not be elaborated here.

[0147] Figure 3 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present invention is shown. Electronic device 800 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 800 may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0148] like Figure 3 As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. The RAM 803 may also store various programs and data required for the operation of the electronic device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0149] Multiple components in electronic device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows electronic device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0150] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the quality assessment method for product carbon footprint data. For example, in some embodiments, the quality assessment method for product carbon footprint data can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the quality assessment method for product carbon footprint data described above can be performed. Alternatively, in other embodiments, the computing unit 801 may be configured, by any other suitable means (e.g., by means of firmware), to perform a quality assessment method for product carbon footprint data.

[0151] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0152] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0153] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0154] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0155] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0156] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0157] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0158] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method of quality assessment of product carbon footprint data, characterized in that, The method comprises the following steps: analyzing multi-source heterogeneous carbon footprint data of a product to obtain target entities and relationships between the target entities; constructing a knowledge graph based on the target entities and the relationships between the target entities to obtain a carbon footprint knowledge graph; evaluating each target path in the carbon footprint knowledge graph based on an evaluation agent to obtain abnormal carbon footprint data; diagnosing the abnormal carbon footprint data based on a diagnosis agent to obtain a diagnosis report; determining a quality evaluation result of the product based on the abnormal carbon footprint data and the diagnosis report.

2. The method of claim 1, wherein, The method of constructing a knowledge graph based on the target entities and the relationships between the target entities to obtain a carbon footprint knowledge graph comprises the following steps: pairing each of the target entities to obtain a pair of to-be-verified entities; for each pair of to-be-verified entities, filling the pair of to-be-verified entities and attributes of the two target entities in the pair of to-be-verified entities into placeholders in a preset first prompt word template to generate a similarity reasoning instruction; inputting the similarity reasoning instruction into the constructed agent to obtain a matching result of the pair of to-be-verified entities; if the matching result is that the two target entities in the pair of to-be-verified entities are of the same category, performing rule checking on the pair of to-be-verified entities based on a preset domain rule library to obtain a rule checking result of the pair of to-be-verified entities; updating each target entity based on the rule checking result of each pair of to-be-verified entities to obtain an updated target entity; constructing the carbon footprint knowledge graph based on each updated target entity and the relationships.

3. The method of claim 2, wherein, The method of updating each target entity based on the rule checking result of each pair of to-be-verified entities to obtain an updated target entity comprises the following steps: if the rule checking result is that the material type attributes of the two target entities in the pair of to-be-verified entities are the same, merging the two target entities in the pair of to-be-verified entities to obtain a merged target entity; if the rule checking result is that the material type attributes of the two target entities in the pair of to-be-verified entities are different, taking the two target entities in the pair of to-be-verified entities as first entities respectively; determining each updated target entity based on each merged target entity and each first entity.

4. The method of claim 2, wherein, After constructing the carbon footprint knowledge graph based on each updated target entity and the relationships, the method further comprises the following steps: determining a text summary based on each updated target entity, attributes of each updated target entity, and the relationships; generating a graph evolution decision instruction based on the text summary, new knowledge triples, and a preset evolution decision prompt template; inputting the graph evolution decision instruction into the constructed agent to obtain a graph update operation instruction set, so that a graph database updates the carbon footprint knowledge graph according to each instruction in the graph update operation instruction set.

5. The method of claim 1, wherein, The evaluation agent evaluates each target path in the carbon footprint knowledge graph based on the evaluation agent to obtain abnormal carbon footprint data, including: Taking the product node as the starting point and the emission source node as the end point, the graph traversal algorithm is used to traverse along the preset semantic relationship in the carbon footprint knowledge graph to obtain each target path; wherein the graph traversal algorithm includes depth-first search algorithm and breadth-first search algorithm; The node sequence and the relationship sequence in each target path are encapsulated to generate a natural language document; Based on the natural language document and the preset integrity evaluation template, an integrity evaluation instruction is generated; Based on the natural language document and the preset logical consistency evaluation template, a consistency evaluation instruction is generated; Based on the natural language document and the preset compliance evaluation template, a compliance evaluation instruction is generated; The integrity evaluation instruction, the consistency evaluation instruction and the compliance evaluation instruction are respectively taken as the input of the evaluation agent pair to obtain the abnormal carbon footprint data.

6. The method of claim 1, wherein, The diagnosis agent diagnoses the abnormal carbon footprint data based on the diagnosis agent to obtain a diagnosis report, including: Based on the abnormal carbon footprint data and the preset credibility evaluation template, a credibility evaluation instruction is generated; Based on the carbon footprint knowledge graph and the preset sensitivity analysis template, a sensitivity analysis instruction is generated; Based on the abnormal carbon footprint data and the preset uncertainty traceability template, an uncertainty traceability instruction is generated; The credibility evaluation instruction, the sensitivity analysis instruction and the uncertainty traceability instruction are respectively taken as the input of the diagnosis agent to obtain the diagnosis report.

7. The method of claim 1, wherein, The multi-source heterogeneous carbon footprint data of the product is parsed to obtain each target entity and the relationship between each target entity, including: In response to a trigger event of obtaining carbon footprint data, the multi-source heterogeneous carbon footprint data of the product is automatically obtained through a preset connector, wherein the multi-source heterogeneous carbon footprint data includes structured data and unstructured data; For the structured data, each field in the structured data is mapped to each node and the attribute of each node through a pre-defined mapping rule; based on each foreign key relationship in the structured data, the relationship between each node is determined; based on each node, the attribute of each node and the relationship between each node, each second entity and the relationship between each second entity are determined; For the unstructured data, the unstructured data is text extracted to obtain unstructured text data; the unstructured text data is named entity recognized to obtain each third entity; and each third entity is relation extracted to obtain the relationship between each third entity; Based on each second entity and each third entity, each target entity is determined; Based on the relationship between each second entity and the relationship between each third entity, the relationship between each target entity is determined.

8. An apparatus for quality evaluation of product carbon footprint data, characterized by, including: The data analysis module is used to analyze the multi-source heterogeneous carbon footprint data of the product to obtain each target entity and the relationship between each target entity; The agent module is constructed to construct a knowledge graph based on each target entity and the relationship between each target entity, and obtain a carbon footprint knowledge graph. The evaluation agent module is configured to evaluate each target path in the carbon footprint knowledge graph based on an evaluation agent, and obtain abnormal carbon footprint data. The diagnosis agent module is configured to diagnose the abnormal carbon footprint data based on a diagnosis agent, and obtain a diagnosis report. The quality evaluation result determination module is configured to determine the quality evaluation result of the product based on the abnormal carbon footprint data and the diagnosis report.

9. A system for quality evaluation of product carbon footprint data, characterized by It comprises: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to make the computer execute the method according to any one of claims 1-7.

Images