Enterprise full-service carbon fusion accounting method and system

Abstract

The invention relates to the technical field of carbon emission management, and discloses an enterprise full-service carbon fusion accounting method and system, and the method comprises the steps: generating a material association mapping table through logistics document records and bill of material records, and carrying out the field dependence analysis of supplier carbon activity records and enterprise internal carbon activity records according to the material association mapping table; the method comprises the following steps: forming a cross-link constrained field dependency relationship table, carrying out structure consistency check in combination with accounting document records, outputting stability-graded structure alignment mapping, carrying out multi-dimensional aperture difference screening based on enterprise internal carbon activity records and a preset accounting boundary table, generating an enterprise aperture rule table, and maintaining accounting aperture consistency. And finally, node adjacency topology reconstruction is carried out through structure alignment mapping and an enterprise aperture rule table, a graph attention encoder embedded with a multi-scale local enhancement module is constructed, aggregation coding is carried out according to a unified activity record stream, a full-link fusion path index is obtained, identification fractures and link gaps are cut through, and the tracking integrity of a product-level carbon footprint path is ensured.

Description

Technical Field

[0001] This invention relates to the field of carbon emission management technology, and more specifically, to a method and system for carbon integration accounting across all business operations of an enterprise. Background Technology

[0002] The application scenario is a carbon data fusion and accounting platform for enterprise-level multi-business systems and multi-level suppliers. It runs on a data middleware or cloud-edge collaborative architecture. It is limited by cross-regional link bandwidth and latency, heterogeneous system interface capabilities and data update frequency, storage consistency and computing resource quotas, etc. Existing technologies mostly adopt abstract methods such as modular data collection, templated report filling, field mapping and caliber comparison, rule verification and anomaly screening, hierarchical summary and lifecycle list aggregation, time smoothing and missing data compensation, identity identification association and link marking evidence storage. It is usually based on the premise that the data field set is relatively stable, the supply chain structure changes in a low frequency, and the accounting boundaries and calibers are preset and unified.

[0003] In actual operation, there are two objective unstable factors: first, the data structure and field semantics of upstream suppliers change frequently with different regions and versions; second, the accounting boundaries and standards of different business lines within the enterprise are adjusted according to operational strategies and compliance requirements. These factors together weaken the effectiveness of field mapping and standard comparison and introduce cross-link identification breaks and gaps in the chain, further making it difficult to maintain the consistency of inventory aggregation and path marking. Ultimately, the integrity of the path tracking of product-level carbon footprint cannot be guaranteed. Therefore, the technical problem that needs to be solved is how to ensure the integrity of the path tracking of product-level carbon footprint while reducing the dispersion of upstream carbon data structure and maintaining the consistency of accounting standards across scenarios within the enterprise.

[0004] In view of this, the present invention proposes a carbon fusion accounting method and system for all business operations of an enterprise to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, this invention provides a method and system for enterprise-wide carbon fusion accounting.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Firstly, a comprehensive carbon accounting method for enterprises is provided, including: Acquire logistics document records, bill of materials records, supplier carbon activity records, and internal enterprise carbon activity records. Generate a material association mapping table based on the logistics document records and bill of materials records. Perform field dependency analysis on the supplier carbon activity records and internal enterprise carbon activity records based on the material association mapping table to generate a field dependency relationship table with cross-process constraints. Obtain accounting voucher records, perform structural consistency checks based on the field dependency table and accounting voucher records, and generate a structural alignment mapping including stability grading; Based on the enterprise's internal carbon activity records and preset accounting boundary table, perform multi-dimensional caliber difference screening to generate an enterprise caliber rule table with priority order; Perform node adjacency topology reconstruction on the structure alignment mapping and enterprise caliber rule table, construct a graph attention encoder with embedded multi-scale local enhancement modules, generate initial node features based on the unified activity record stream, aggregate and encode the initial node features based on the graph attention encoder, generate a full-link fusion path index, and generate carbon fusion accounting results based on the full-link fusion path index.

[0007] In some embodiments, a material association mapping table is generated based on logistics document records and bill of materials records, including: Spatiotemporal trajectory data is obtained from logistics document records, and material hierarchy data is obtained from bill of materials records. The node freight sequence representing the temporal relationship of transportation nodes is extracted from the spatiotemporal trajectory data, and the hierarchical relationship sequence of material supply is extracted from the material hierarchy data. The material transport chain sequence corresponding to each transport activity is determined based on the node freight sequence, and the supply chain sequence of the material supply process is determined based on the hierarchical relationship sequence. The material transport chain sequence is screened for effectiveness based on the supply chain sequence, and material transport chains that do not match the supply chain sequence are eliminated; Based on the selected material transportation chain and supply chain sequence, establish the association mapping relationship between nodes and materials to obtain the material association mapping table.

[0008] In some embodiments, field dependency analysis is performed on supplier carbon activity records and internal enterprise carbon activity records based on the material association mapping table to generate a field dependency relationship table with cross-stage constraints, including: Based on the material association mapping table, extract the migration order of materials between different supply links and construct the supply chain flow sequence. The set of data fields that can participate in dependency analysis is defined based on the carbon activity data of supply nodes and the internal carbon activity data of enterprises in the process flow sequence. Construct a cross-stage dependency chain for fields based on the defined set of data fields, and generate a field dependency table based on the cross-stage dependency chain.

[0009] In some embodiments, a confidence mask matrix is ​​generated by constructing a dual-tower verification model based on a field dependency table and accounting voucher data, including: Extract the set of field combinations to be validated from the field dependency table, and extract the corresponding set of business field combinations from the accounting voucher data; A dual-tower structure is constructed by combining the set of fields to be verified with the set of business fields to form multiple field combination correspondence pairs; Calculate the structural similarity distance between the corresponding pairs of field combinations, and determine the credibility of the field combination structure based on the similarity distance; Generate a confidence mask matrix based on the confidence level of each field combination to represent the confidence level of the field combination structure in the field dependency table.

[0010] In some embodiments, a method for generating a confidence mask matrix includes: Extract the structural difference data of various field combinations in historical business scenarios and generate a set of structural difference data. Based on the historical frequency of structural changes in the structural difference data set, the structural reliability level of each field combination is determined. The confidence mask matrix is ​​modified according to the structural confidence level to generate the modified confidence mask matrix.

[0011] In some embodiments, a method for constructing a graph attention encoder with embedded multi-scale local enhancement modules includes: Based on the corrected confidence mask matrix, low-confidence node connections are removed from the initial node topology formed by the structure alignment mapping to generate a pruned topology. The priority order of topology reconnection for different nodes in various business scenarios is determined based on the enterprise's rules table. Based on the topology reconnection priority, the pruned topology is reconnected to generate an enhanced topology. A graph attention encoder with embedded multi-scale local enhancement modules is constructed based on the enhanced topology.

[0012] In some embodiments, the multi-scale local enhancement module in the graph attention encoder includes: Neighborhood node sampling at different scales is performed on each node in the enhanced topology to form a neighborhood structure at multiple scales; Node feature aggregation is performed on the neighborhood structure at multiple scales to generate local enhancement features for nodes at each scale. By fusing node local enhancement features at different scales, multi-scale local enhancement features are generated for use by the image attention encoder.

[0013] In some embodiments, a method for aggregating and encoding initial node features based on a graph attention encoder to generate a full-link fused path index includes: Based on carbon activity data from supply nodes and internal enterprise carbon activity data, a set of original node characteristics is constructed. By utilizing the enhanced topology, cross-link feature alignment is performed on the original feature set of nodes to generate a node-aligned feature set. Initial node features are generated based on the node alignment feature set, and node feature aggregation encoding is performed on the initial node features using the multi-scale local enhancement module in the graph attention encoder to obtain node aggregate features; Generate a full-link fusion path index based on node aggregation characteristics.

[0014] In some embodiments, a method for generating a full-link fusion path index based on node aggregation features includes: Calculate the feature transfer weights between nodes based on the node aggregation characteristics, and generate node feature transfer paths based on the feature transfer weights. Based on the node characteristics and transmission path, the data transmission direction of each link in the supply chain is determined, and a data flow path chain is generated. Based on the data flow path chain, the carbon activity data of supply nodes and internal enterprise carbon activity data are used to perform full-link traceability and cumulative calculation to generate a full-link fusion path index.

[0015] Secondly, a comprehensive enterprise carbon accounting system is provided, which is used to implement the aforementioned comprehensive enterprise carbon accounting method, including: Dependency generation module: used to obtain logistics document records, bill of materials records, supplier carbon activity records and internal enterprise carbon activity records, generate a material association mapping table based on logistics document records and bill of materials records, and perform field dependency analysis on supplier carbon activity records and internal enterprise carbon activity records based on the material association mapping table to generate a field dependency relationship table with cross-process constraints. Structure alignment generation module: used to obtain accounting voucher records, perform structure consistency checks based on field dependency relationship table and accounting voucher records, and generate structure alignment mapping including stability level; Calibration Generation Module: Used to perform multi-dimensional caliber difference screening based on the company's internal carbon activity records and preset accounting boundary tables, and generate a corporate caliber rule table with priority order annotation; The graph accounting encoding module is used to perform node adjacency topology reconstruction on the structure alignment mapping and enterprise caliber rule table, construct a graph attention encoder with embedded multi-scale local enhancement modules, generate initial node features based on the unified activity record stream, aggregate and encode the initial node features based on the graph attention encoder, generate the full-link fusion path index, and generate carbon fusion accounting results based on the full-link fusion path index.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention generates a material association mapping table using logistics document records and bill of materials records. Based on this, it performs field dependency analysis on supplier carbon activity records and internal enterprise carbon activity records to form a field dependency relationship table with cross-stage constraints. First, it constrains structural dispersion and identifier breakage at the data entry point. Then, it obtains accounting voucher records and performs structural consistency checks based on the field dependency relationship table to generate a structural alignment mapping with stability levels. This ensures that records from different sources and versions enter the same caliber under a measurable alignment level. Subsequently, it conducts multi-dimensional caliber difference screening based on internal enterprise carbon activity records and a preset accounting boundary table to output an enterprise caliber rule table with priority order marking. This ensures stable transmission of accounting boundaries in cross-business line scenarios. Finally, the node adjacency topology is reconstructed by performing structural alignment mapping and enterprise caliber rule table, and a graph attention encoder with embedded multi-scale local enhancement module is constructed. Initial node features are generated and aggregated based on the unified activity record stream to obtain the full-link fusion path index. In this way, the gaps across links are filled and the link continuity is maintained during the list aggregation and path marking process. This reduces the dispersion of upstream carbon data structure and maintains the consistency of internal accounting caliber under fluctuating scenarios, ensuring the integrity of product-level carbon footprint path tracking. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a carbon fusion accounting method for all business operations of an enterprise according to the present invention; Figure 2 This is a schematic diagram of the structure of an enterprise-wide carbon fusion accounting system according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. In the following detailed description, many specific details are set forth to provide a thorough understanding of the exemplary embodiments described. However, it will be apparent to those skilled in the art that the described embodiments may be practiced without some or all of these specific details. In other exemplary embodiments, well-known structures have not been described in detail to avoid unnecessarily obscuring the concepts of this disclosure. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention. Furthermore, the various aspects described in the embodiments may be combined arbitrarily without conflict.

[0019] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0020] Example 1 Please see Figure 1 As shown in the figure, this embodiment discloses a method for enterprise-wide carbon fusion accounting, including: S10: Obtain spatiotemporal trajectory data from logistics document records, and obtain material-level data, supply node carbon activity data, and internal enterprise carbon activity data from bill of materials records. Generate a material association mapping table based on logistics document records and bill of materials records, and perform field dependency analysis on supply node carbon activity data and internal enterprise carbon activity data based on the material association mapping table to generate a field dependency relationship table with cross-link constraints. In this embodiment, spatiotemporal trajectory data is used to record the temporal sequence and spatial flow relationship of materials between various transportation and processing nodes. Material hierarchy data is used to record the material hierarchy relationship between raw materials, semi-finished products, and finished products. Supply node carbon activity data is used to characterize the carbon activity records of upstream supply nodes in various material supply links, such as carbon emissions and energy consumption. Enterprise internal carbon activity data is used to characterize the enterprise's internal carbon activity records in warehousing, production, processing, transportation, and other links. By generating a material association mapping table based on logistics document records and bill of materials records, each transportation node and supply node can be bound to specific materials and their hierarchical relationships. Then, based on the material association mapping table, field dependency analysis is performed on the supply node carbon activity data and the enterprise internal carbon activity data to generate a field dependency relationship table with cross-link constraints, which is used to characterize the dependency relationship of data fields related to the same material in different links throughout the entire supply chain path.

[0021] For example, spatiotemporal trajectory data corresponding to a certain product batch can be obtained, where the node sequence consists of upstream supply node S1, processing node S2, and internal enterprise warehousing node E1. The corresponding material hierarchy data records the hierarchical relationship between raw material code M101, semi-finished product code M201, and finished product code M301. When generating a material association mapping table based on logistics document records and bill of materials records, the structure shown in Table 1 can be obtained. In this table, the node identifier field is mapped one-to-one with the material code field to establish the association relationship between the node and the material. Based on this, field dependency analysis is performed on the carbon emission fields and energy consumption fields related to nodes S1, S2, E1, and material codes M101, M201, and M301 in the carbon activity data of the supply nodes and the internal enterprise carbon activity data. This generates a set of field dependency table elements to represent cross-stage constraint relationships such as "upstream carbon emission fields depend on the activity records of raw material code M101" and "internal enterprise carbon emission fields depend on the aggregation results of finished product code M301". This lays the foundation for unifying the accounting standards of carbon data from different sources, as shown in Table 1. Table 1 Furthermore, it is understandable that by constructing a material association mapping table using spatiotemporal trajectory data and material-level data, and then performing field dependency analysis on the carbon activity data of supply nodes and the internal carbon activity data of the enterprise based on the material association mapping table, a field dependency relationship table with cross-link constraints can be generated. Compared with the existing technology of simply mapping at the field name or report caliber level, this not only reduces the dispersion of upstream supply node carbon data caused by differences in field meaning and hierarchy, but also forms a unified skeleton in the data transmission link to constrain the correspondence between the internal carbon activity data of the enterprise and the carbon activity data of supply nodes. In the mixed scenario of multiple business systems and multi-level suppliers, this provides a stable field dependency foundation for subsequent accounting voucher verification, enterprise caliber rule table construction, and topology pruning and reconnection in graph attention encoder, which is conducive to improving the integrity and consistency of product-level carbon footprint path tracking.

[0022] A material association mapping table is generated based on logistics document records and bill of materials records, including: Spatiotemporal trajectory data is obtained from logistics document records, and material hierarchy data is obtained from bill of materials records. The node freight sequence representing the temporal relationship of transportation nodes is extracted from the spatiotemporal trajectory data, and the hierarchical relationship sequence of material supply is extracted from the material hierarchy data. The material transport chain sequence corresponding to each transport activity is determined based on the node freight sequence, and the supply chain sequence of the material supply process is determined based on the hierarchical relationship sequence. The material transport chain sequence is screened for effectiveness based on the supply chain sequence, and material transport chains that do not match the supply chain sequence are eliminated; Based on the selected material transportation chain and supply chain sequence, establish the association mapping relationship between nodes and materials to obtain the material association mapping table.

[0023] In this embodiment, spatiotemporal trajectory data is used to record the flow path of materials between transportation nodes over time. Material hierarchy data is used to record the hierarchical relationship between raw materials, semi-finished products, and finished products. The node freight sequence is a sequence of nodes arranged chronologically extracted from the spatiotemporal trajectory data. The material supply hierarchy sequence is a sequence of materials extracted from the material hierarchy data and unfolded according to the process sequence. The material transportation chain sequence is a node link corresponding to each transportation activity divided based on the node freight sequence. The supply chain sequence is a material supply chain path unfolded based on the hierarchy sequence. The material association mapping table is used to establish a one-to-one correspondence between the filtered material transportation chain and the node identifier field and material code field in the supply chain.

[0024] It should be noted that when generating the material association mapping table based on logistics document records and bill of materials records, the process first involves extracting the node freight sequence representing the temporal relationship of transportation nodes from the spatiotemporal trajectory data, and extracting the hierarchical relationship sequence of material supply from the material hierarchy data. Then, based on the node freight sequence, the flow between adjacent nodes is divided into material transportation chain sequences corresponding to each transportation activity. Next, a supply chain sequence of the material supply process is generated based on the hierarchical relationship sequence. Furthermore, the supply chain sequence is used to effectively screen the material transportation chain sequences, eliminating material transportation chains that do not match the supply chain sequence in terms of node order or material hierarchy. Finally, based on the screened material transportation chains and supply chain sequences, an association mapping relationship is established between each transportation node and the corresponding material, thereby obtaining the material association mapping table.

[0025] For example, in this instance, the node freight sequence S1, S2, E1 can be extracted from the spatiotemporal trajectory data, corresponding to a batch of materials passing through the upstream supply node S1, the processing node S2, and the internal enterprise node E1 in sequence. At the same time, the material supply hierarchy sequence M101, M201, M301 can be extracted from the material hierarchy data, corresponding to the raw material code M101 being processed into the semi-finished product code M201 and finally assembled into the finished product code M301. When determining the material transportation chain sequence based on the node freight sequence, S1→S2 and S2→E1 can be regarded as two transportation links. Then, the supply chain sequence M101→M201→M301 can be obtained based on the hierarchy sequence. By comparison, it can be seen that S1 corresponds to M101, S2 corresponds to M201, and E1 corresponds to M301. Thus, the node-material association relationship of S1–M101, S2–M201, and E1–M301 is formed in the material association mapping table, enabling those skilled in the art to clearly understand the cross-link flow of materials between different nodes.

[0026] Based on the material association mapping table, perform field dependency analysis on supplier carbon activity records and internal enterprise carbon activity records to generate a field dependency relationship table with cross-stage constraints, including: Based on the material association mapping table, extract the migration order of materials between different supply links and construct the supply chain flow sequence. The set of data fields that can participate in dependency analysis is defined based on the carbon activity data of supply nodes and the internal carbon activity data of enterprises in the process flow sequence. Construct a cross-stage dependency chain for fields based on the defined set of data fields, and generate a field dependency table based on the cross-stage dependency chain.

[0027] In this embodiment, extracting the migration order of materials between different supply links based on the material association mapping table and constructing the supply chain link flow sequence means that, based on the aforementioned material association mapping table which has already established a one-to-one correspondence between node identifier fields and material code fields, following the flow order of raw material codes, semi-finished product codes, and finished product codes between each node, extracting an ordered sequence at the link level such as "raw material entering the factory → intermediate processing → final assembly". This supply chain link flow sequence is used as a constraint condition to limit which data field sets in the carbon activity data of supply nodes and the enterprise's internal carbon activity data can participate in dependency analysis, thereby avoiding the erroneous inclusion of fields unrelated to the current material or the current supply chain link into the scope of the field dependency relationship table construction.

[0028] For example, based on the material association mapping table, if node S1 corresponds to raw material code M101, node S2 corresponds to semi-finished product code M201, and node E1 corresponds to finished product code M301, then the migration order of materials between different supply links, S1→S2→E1, can be extracted from the material association mapping table, and a supply chain link flow sequence of "upstream supply link, processing link, and internal enterprise link" can be constructed. When limiting the set of data fields that can participate in dependency analysis in the carbon activity data of supply nodes and the carbon activity data of internal enterprises, only the carbon emission, energy consumption, and output fields related to M101, M201, and M301 can be retained. These fields can be organized according to the link flow order into a data field set such as {(upstream supply link, raw material carbon emission field), (processing link, semi-finished product carbon emission field), (internal enterprise link, finished product carbon emission field)}. This allows those skilled in the art to directly use the associated fields of the same material in different links as nodes to establish a dependency link of "upstream field → processing field → internal enterprise field" along the supply chain link flow sequence when constructing the cross-link dependency relationship chain of fields.

[0029] S20: Obtain accounting voucher data, calculate the structural similarity distance between the accounting voucher data and the field dependency table, and generate a confidence mask matrix based on the structural similarity distance to characterize the credibility of the structural combination of each field in the field dependency table; Based on the field dependency table and accounting voucher data, a dual-tower validation model is constructed to generate a confidence mask matrix, including: Extract the set of field combinations to be validated from the field dependency table, and extract the corresponding set of business field combinations from the accounting voucher data; A dual-tower structure is constructed by combining the set of fields to be verified with the set of business fields to form multiple field combination correspondence pairs; Calculate the structural similarity distance between the corresponding pairs of field combinations, and determine the credibility of the field combination structure based on the similarity distance; Generate a confidence mask matrix based on the confidence level of each field combination to represent the confidence level of the field combination structure in the field dependency table.

[0030] In this embodiment, accounting voucher data is used to characterize the actual business structure that the enterprise has confirmed in the financial accounting system. The field dependency table is used to characterize the theoretical field dependency relationship between carbon activity data of supply nodes and carbon activity data of the enterprise in different supply chain links. By calculating the structural similarity distance between accounting voucher data and field dependency table, the consistency between the "theoretical dependency chain" and the "actual accounting chain" can be compared from the structural level. On this basis, a confidence mask matrix is ​​generated to characterize the credibility of the structure of each field combination in the field dependency table. This enables the automatic weakening or shielding of the corresponding links for field combinations with low credibility during subsequent topology pruning and reconnection, thereby reducing carbon data transmission deviations caused by adjustments in the enterprise's internal accounting standards or inconsistencies in the meaning of fields across systems.

[0031] It should be noted that extracting the set of field combinations to be verified from the field dependency table and the corresponding set of business field combinations from the accounting voucher data is to construct a "field combination structure from a carbon accounting perspective" and a "field combination structure from a financial accounting perspective" on both sides of the dual-tower verification model. When constructing the dual-tower structure, the set of field combinations to be verified and the set of business field combinations are paired to form multiple field combination correspondence pairs. For each pair of field combinations, the structural similarity distance is calculated. The structural similarity distance can comprehensively consider factors such as the supply chain link to which the field belongs, the semantics of the field name, the range of field values, and the dependency relationship with other fields. When the structural similarity distance is small, it means that the field combination in the field dependency table is highly consistent with the business field combination in the accounting voucher data in terms of structure, and its credibility is high. When the structural similarity distance is large, it means that there is a large difference between the two in terms of structure, and its credibility is low. Based on this, a credibility label can be attached to each field combination relationship, providing a quantitative basis for the subsequent generation of the confidence mask matrix.

[0032] For example, the field dependency relationship table corresponding to the aforementioned nodes S1, S2, E1 and material codes M101, M201, M301 can be used to extract three types of field combinations: "carbon emission field of raw materials in the upstream supply chain", "carbon emission field of semi-finished products in the processing chain", and "carbon emission field of finished products in the internal chain". Then, the corresponding three types of business field combinations can be found in the accounting voucher data: "amount field of raw material purchase account", "amount field of production cost account", and "amount field of inventory account". To facilitate a clear and intuitive display of the correspondence between these three pairs of field combinations in the instruction manual, they can be summarized as shown in Table 2 below: Table 2 Based on Table 2 above, the dual-tower verification model can calculate the structural similarity distance for the corresponding relationship of each row of field combinations and map the distance value to the credibility of the field combination structure. Then, according to the credibility of each field combination, the corresponding confidence coefficient is filled in at the position of the same dimension as the field dependency relationship table to form a confidence mask matrix for representing the credibility of each field combination structure in the field dependency relationship table. Compared with the existing technology that relies solely on human experience or the frequency of field occurrence to judge the reliability of field mapping, this embodiment introduces accounting voucher data to construct a dual-tower verification model, realizing automatic consistency verification of carbon accounting field dependencies. This is beneficial for maintaining the stability and credibility of the field dependency structure when the internal accounting boundary changes across business scenarios.

[0033] Methods for generating confidence mask matrices include: Extract the structural difference data of various field combinations in historical business scenarios and generate a set of structural difference data. Based on the historical frequency of structural changes in the structural difference data set, the structural reliability level of each field combination is determined. The confidence mask matrix is ​​modified according to the structural confidence level to generate the modified confidence mask matrix.

[0034] Understandably, the structural difference dataset is used to record the differences in structural changes of each field combination in the field dependency table under different historical business scenarios. The structural confidence level is used to characterize the degree to which the structure of a certain field combination remains stable in historical scenarios. The confidence mask matrix is ​​used to represent the confidence of the structure of each field combination in the field dependency table. By further modifying the confidence mask matrix based on the aforementioned structural similarity distance calculation and combining it with historical structural difference data, short-term structural deviations caused by single business data fluctuations can be effectively avoided. This allows the confidence mask matrix to more accurately reflect the long-term stability of each field combination, thus providing a more robust confidence basis for subsequent topology pruning and reconnection.

[0035] When extracting structural difference data for various field combinations under historical business scenarios to generate a structural difference data set, the structural changes of the field combinations "upstream supply chain raw material carbon emission field," "processing chain semi-finished product carbon emission field," and "internal enterprise chain finished product carbon emission field" involved in the field dependency table in this example during the three historical accounting periods can be used as the basic data to form the structural difference data set shown in Table 3 below. Then, based on the number of structural changes of each field combination in Table 3, its historical frequency is calculated. When the number of structural changes is low, it indicates that the field combination has maintained structural stability in historical business scenarios and its structural confidence level is high. When the number of structural changes is high, it indicates that the field combination has significant structural fluctuations in historical scenarios and its structural confidence level is low. Finally, the aforementioned confidence mask matrix is ​​corrected based on the structural confidence level, so that the confidence mask matrix retains the structural similarity distance results while further integrating long-term structural stability factors, thereby generating a corrected confidence mask matrix, as shown in Table 3 below: Table 3 Furthermore, by modifying the confidence mask matrix based on structural trust level, compared to the existing technology that relies solely on single-period business data to determine the reliability of field mapping, this embodiment can simultaneously consider the structural similarity distance of field combinations in the current accounting period and the frequency of structural changes in multiple historical accounting periods. This allows the generated modified confidence mask matrix to more accurately reflect the true stability of field combinations under cross-supply chain links, cross-accounting period business fluctuations, and changes in accounting standards. This provides a more robust and trustworthy basis for subsequent pruning and reconnection of the initial node topology, thereby improving the robustness of the carbon data transmission path and enhancing the consistency of product-level carbon footprint end-to-end tracking.

[0036] S30: Based on the enterprise's internal carbon activity data and preset accounting boundary constraint operators, perform multi-dimensional space cutting to generate an enterprise-specific rule table; In this embodiment, internal carbon activity data is used to characterize the carbon emissions, energy consumption, and output of an enterprise in various internal processes such as warehousing, production, processing, and internal transportation. The accounting boundary constraint operator is used to formally describe the accounting boundary under different business lines and management perspectives within the enterprise, abstracting rules such as "whether it is included in the accounting of a certain product, whether it is included in the carbon footprint of a certain process segment, and whether it belongs to a certain organizational boundary" into computable constraints. Multidimensional space segmentation is used to segment the carbon activity data space containing multiple dimensions such as time dimension, organizational dimension, business dimension, and material dimension according to the accounting boundary constraint operator, dividing data belonging to different accounting scenarios into non-overlapping or partially overlapping subspaces. The enterprise caliber rule table is used to encode the selection results of the above accounting subspaces into a set of accounting caliber parameters that can be referenced by the subsequent graph attention encoder and field dependency table.

[0037] It should be noted that in the specific process of performing multi-dimensional spatial segmentation based on internal enterprise carbon activity data and preset accounting boundary constraint operators, the basic coordinate axes such as time dimension, organizational dimension, business dimension, and material dimension can be clearly defined in the internal enterprise carbon activity data first. For example, the time dimension can be divided by payment period or day, the organizational dimension by factory, workshop, or production line, the business dimension by procurement, production, warehousing, and sales activities, and the material dimension by material code fields such as M101, M201, and M301. Then, the accounting boundary constraint operators are applied to the above multi-dimensional space, such as... For the "product-level carbon footprint accounting caliber", carbon activity records are selected only within a specific time interval, a specific organizational scope, and a specific set of materials. Through multi-dimensional space cutting operations, records that meet the constraints are assigned to the subspace corresponding to the accounting caliber, while records that do not meet the constraints are excluded. Finally, for each accounting scenario, the value range and selection rules in each dimension are encoded into an enterprise caliber rule table, which can be referenced later when pruning and reconnecting topology structures and aligning node features. This enables those skilled in the art to call the enterprise's internal accounting boundaries in a unified vector form in different scenarios.

[0038] For example, two different accounting boundary constraint operators can be defined for the production line corresponding to the region identifier field A1. One is to "only account for carbon activity records related to the finished product output corresponding to material code M301", and the other is to "account for carbon activity records throughout the entire process from material code M101 to M301". In the enterprise's internal carbon activity data, if the time dimension is the current payment period, the organizational dimension is factory F1 and workshop L1, the business dimension includes production and warehousing activities, and the material dimension includes M101, M201, and M301, then when applying the first accounting boundary constraint operator for multi-dimensional spatial segmentation, only the material code field M301 and the business... When applying the second type of accounting boundary constraint operator to perform multi-dimensional space cutting for carbon activity records with material code fields M101, M201, and M301 and business dimensions of raw material warehousing, semi-finished product processing, and finished product warehousing, all carbon activity records with these material code fields are retained. These two different selection rules are encoded into two enterprise-specific rule tables. For example, vector 1,0,0 indicates that only the finished product stage is counted, while vector 1,1,1 indicates that the entire process stage is counted. This enterprise-specific rule table can be used as a parameter input to control the priority of topology reconnection and the range of feature aggregation under different business scenarios when constructing the enhanced topology structure and graph attention encoder.

[0039] S40: The initial node topology is pruned and reconnected using a confidence mask matrix. A graph attention encoder with embedded multi-scale local enhancement modules is constructed by combining the enterprise caliber rule table. Initial node features are generated based on carbon activity data of supply nodes and internal carbon activity data of enterprises. The initial node features are aggregated and encoded based on the graph attention encoder to generate a full-link fusion path index. The carbon fusion accounting result is generated based on the full-link fusion path index.

[0040] Methods for constructing graph attention encoders with embedded multi-scale local enhancement modules include: Based on the corrected confidence mask matrix, low-confidence node connections are removed from the initial node topology formed by the structure alignment mapping to generate a pruned topology. The priority order of topology reconnection for different nodes in various business scenarios is determined based on the enterprise's rules table. Based on the topology reconnection priority, the pruned topology is reconnected to generate an enhanced topology. A graph attention encoder with embedded multi-scale local enhancement modules is constructed based on the enhanced topology.

[0041] In this embodiment, the initial node topology is a supply chain graph structure built based on the material association mapping table. Nodes represent supply nodes and internal enterprise nodes, and the connections between nodes represent the flow of materials in different stages. The confidence mask matrix is ​​used to characterize the confidence of the combination structure of each field in the field dependency relationship table, and reflects the reliability of each edge in real business by mapping it to the node connection relationship in the initial node topology. The pruned topology structure refers to the graph structure retained after deleting low-confidence node connection relationships using the confidence mask matrix. The enhanced topology structure is the graph structure obtained by topologically reconnecting the connection relationships between nodes based on the pruned topology structure and the enterprise caliber rule table. The graph attention encoder is an encoding network that aggregates the initial features of nodes on the enhanced topology structure. The multi-scale local enhancement module is used to extract multi-level features of the local structure of nodes at different neighborhood scales. The full-link fusion path index is the full-link carbon accounting output obtained by aggregating and calculating the carbon activity data of supply nodes and internal enterprise carbon activity data under the constraints of the above graph structure and encoder.

[0042] It should be noted that in the specific process of pruning and reconnecting the initial node topology using the confidence mask matrix, we can first locate the corresponding low-confidence node connections in the initial node topology based on the low-confidence field combination structure in the confidence mask matrix. These connections can then be deleted from the graph to generate the pruned topology. For example, in this instance, if there are two connections between the original node flows S1 and S2 from different sources, one of which originates from a field combination with low structural similarity and historical structural differences, then during pruning, the high-confidence connection can be retained while the low-confidence connection can be deleted. Subsequently, the priority order of topology reconnection for different nodes under various business scenarios is determined according to the enterprise caliber rule table. For example, if the enterprise caliber rule table indicates that the current accounting scenario only focuses on the carbon accounting of the entire process from raw materials to finished products, then the connection relationship that ensures the path connectivity between S1, S2, and E1 is restored first in the pruned topology structure. Meanwhile, the connection relationship of bypass nodes that is not related to the current accounting caliber is set with a lower reconnection priority. Finally, the topology relationship reconnection is performed on the pruned topology structure according to the topology reconnection priority, resulting in an enhanced topology structure that removes low-reliability connections and ensures the connectivity of the main carbon data path under the constraints of the current enterprise caliber rule table.

[0043] In this embodiment, after obtaining the enhanced topology, a graph attention encoder embedding a multi-scale local enhancement module can be constructed on the graph structure. Initial node features are generated based on the carbon activity data of the supply nodes and the enterprise's internal carbon activity data. Then, the graph attention encoder aggregates and encodes the initial node features under the constraints of the enhanced topology, ultimately obtaining the aggregated node features used for subsequent carbon fusion accounting. In this example, the initial node features of nodes S1, S2, and E1 can respectively include their corresponding material codes M101, M201, and M301, carbon emissions, energy consumption, and financial correlation fields aligned with accounting voucher data. The graph attention encoder, through the multi-scale local enhancement module, in one hop, These features are weighted and aggregated over two- or even multi-hop neighborhoods, and the carbon activity information of upstream node S1 and intermediate node S2 is gradually transmitted to the enterprise internal node E1. This allows the final generated end-to-end fused path index to fully reflect the carbon footprint accumulation process from upstream supply nodes to enterprise internal nodes on the enhanced topology after pruning and reconnection. Compared to the method of simply summing on the unverified topology, this embodiment uses a confidence mask matrix to constrain pruning and reconnection and combines it with the enterprise caliber rule table to construct a graph attention encoder. This can effectively suppress the interference of low-confidence paths on the accounting results, while maintaining the path continuity and caliber consistency of the end-to-end fused path index when the supply chain structure and accounting boundary change.

[0044] The multi-scale local enhancement module in the graph attention encoder includes: Neighborhood node sampling at different scales is performed on each node in the enhanced topology to form a neighborhood structure at multiple scales; Node feature aggregation is performed on the neighborhood structure at multiple scales to generate local enhancement features for nodes at each scale. By fusing node local enhancement features at different scales, multi-scale local enhancement features are generated for use by the image attention encoder.

[0045] In this embodiment, the multi-scale neighborhood structure is used to characterize the structural information of nodes in different neighborhood ranges within the enhanced topology. The node feature aggregation processing at different scales is used to extract local features from one-hop neighborhoods, two-hop neighborhoods, or neighborhoods with more than one hop, respectively. The multi-scale local enhancement features are used to supplement the structural expressive ability of nodes in different local ranges in the graph attention encoder. By sampling, aggregating, and fusing neighborhood nodes at multiple scales, the node features can reflect both the local context information of the node and the larger-scale structural associations in the cross-link carbon activity data transmission process, enabling the graph attention encoder to have stronger multi-scale expressive ability when performing node feature aggregation encoding.

[0046] It should be noted that when sampling neighbor nodes at different scales for each node in the enhanced topology, neighborhood sets of different scales, such as one-hop neighborhood, two-hop neighborhood, and three-hop neighborhood, can be obtained for each node according to the connectivity of the node in the enhanced topology. In this example, there is a direct connection between nodes S1, S2, and E1 in the enhanced topology. Therefore, the one-hop neighborhood of S1 is {S2}, and the two-hop neighborhood is {E1}. The one-hop neighborhood of S2 is {S1, E1}, and the two-hop neighborhood is an empty set. The one-hop neighborhood of E1 is {S2}, and the two-hop neighborhood is {S1}. These neighborhoods of different ranges are used as neighborhood structures of multiple scales for subsequent node feature aggregation processing, which can be intuitively represented as the structured data shown in Table 4 below: Table 4 Understandably, performing node feature aggregation on neighborhood structures at multiple scales means performing weighted aggregation on the features of neighboring nodes at each scale. This involves superimposing, averaging, or attention-weighting the node feature vectors corresponding to neighboring nodes in the carbon activity data of the supply node and the internal carbon activity data of the enterprise. This results in the node's local enhancement features obtained at the one-hop scale emphasizing the influence from directly adjacent nodes, while the node's local enhancement features obtained at the two-hop or three-hop scale emphasize the cross-link influence from indirectly related nodes. For example, in this instance, node feature aggregation can be performed on the one-hop neighborhood structure {S1, E1} of S2. After aggregating the feature vectors of nodes S1 and E1, the local enhancement features of node S2 at the one-hop scale are generated. Furthermore, feature superposition is not performed when the two-hop neighborhood structure is an empty set, so that the aggregation results at each scale have a clear structural source.

[0047] Furthermore, when fusing node local enhancement features of different scales to generate multi-scale local enhancement features for use by the graph attention encoder, local enhancement features of one-hop scale, local enhancement features of two-hop scale, and local enhancement features of other scales can be fused in a weighted manner. This allows the fused features to contain both local information of the nodes and cross-link propagation information. Compared to the method of relying solely on a single-scale neighborhood for feature aggregation, this embodiment introduces a multi-scale local enhancement module, which enables the graph attention encoder to have stronger robustness in handling the diversity of supply chain structures and differences in accounting standards within enterprises, thereby improving the ability of the full-link fusion path index to capture full-link features.

[0048] Methods for generating end-to-end fused path indexes by aggregating and encoding initial node features based on graph attention encoders include: Based on carbon activity data from supply nodes and internal enterprise carbon activity data, a set of original node characteristics is constructed. By utilizing the enhanced topology, cross-link feature alignment is performed on the original feature set of nodes to generate a node-aligned feature set. Initial node features are generated based on the node alignment feature set, and node feature aggregation encoding is performed on the initial node features using the multi-scale local enhancement module in the graph attention encoder to obtain node aggregate features; Generate a full-link fusion path index based on node aggregation characteristics.

[0049] Understandably, the original node feature set is used to carry the original numerical representation of carbon activity data of supply nodes and internal enterprise carbon activity data on the graph structure, including but not limited to the carbon emissions, energy consumption, output, and amount and material code fields of each node after alignment with accounting voucher data. The node alignment feature set is a feature set obtained by combining the original node feature set with the enhanced topology structure to perform cross-link feature alignment, which is used to ensure that nodes in the same supply chain link or the same material path are comparable in feature dimensions. The node initial feature is the input feature vector obtained by performing dimensional compression or linear transformation on the node alignment feature set. The node aggregated feature is the output feature vector obtained by performing feature aggregation encoding on the node initial feature in the graph attention encoder through the multi-scale local enhancement module. The full-link fusion path index is the full-link carbon accounting output obtained by the transmission and accumulation calculation of all node aggregated features on the supply chain path.

[0050] It should be noted that when constructing the original feature set of nodes based on carbon activity data of supply nodes and carbon activity data within the enterprise, the carbon emissions, energy consumption, and related accounting amount fields of the material codes M101, M201, and M301 corresponding to nodes S1, S2, and E1 in the aforementioned example can be concatenated in a fixed order to form feature vectors. All node feature vectors are then used to form the original feature set of nodes. Subsequently, the original feature set of nodes is aligned across links using an enhanced topology structure. For example, when multiple upstream supply nodes converge to the same processing node, the carbon emissions and amount fields can be normalized or summed on the same material path, so that the processing node and the internal nodes of the enterprise can be consistent with the actual material flow in terms of feature dimensions, thereby obtaining the node aligned feature set. Furthermore, based on the node aligned feature set, a node initial feature with unified dimensions is generated through linear mapping or embedding transformation, providing a standardized feature representation for the input of the graph attention encoder.

[0051] For example, we can construct original feature sets for nodes S1, S2, and E1, setting the feature vector of S1 as (raw material carbon emissions 10, raw material energy consumption 5, raw material procurement amount 100), the feature vector of S2 as (semi-finished product carbon emissions 6, semi-finished product energy consumption 4, production cost amount 80), and the feature vector of E1 as (finished product carbon emissions 3, finished product energy consumption 2, inventory value 120). When using the enhanced topology structure for cross-stage feature alignment, we can rely on the path M101→M201→M301 in the material association mapping table and the enterprise's definition. The rule table uses a full-process accounting approach, accumulating the carbon emission features of S1 and S2 along the path into the feature dimension corresponding to E1. This ensures that the node alignment features of E1 contain both carbon activity information from this stage and information transmitted from upstream stages. The aligned features are then mapped to node initial features of uniform length through a linear transformation and input into a graph attention encoder embedded with a multi-scale local enhancement module. Under the constraint of enhanced topology, the graph attention encoder performs multi-scale neighborhood feature aggregation on the node initial features. The resulting node aggregation features can be used to perform carbon fusion accounting along the data flow path chain in subsequent steps.

[0052] The above-described process of aggregating and encoding initial node features based on a graph attention encoder to generate a full-link fused path index is superior to the existing method of simply overlaying supply node carbon activity data and internal enterprise carbon activity data at the table or report level. On the one hand, it utilizes an enhanced topology structure and an enterprise-specific rule table to ensure that the feature alignment and feature aggregation process strictly follows the supply chain path and unified accounting boundary depicted by the material association mapping table, reducing the impact of upstream carbon data structure dispersion and internal caliber differences on the accounting results. On the other hand, the multi-scale local enhancement module in the graph attention encoder integrates the local structure of nodes and cross-link dependencies at different neighborhood scales, enabling the full-link fused path index to maintain the integrity of path tracking and the consistency of accounting caliber in complex supply chains and multi-business scenarios, thereby effectively supporting the accurate calculation of product-level carbon footprint and full-link traceability.

[0053] Methods for generating a full-link fusion path index based on node aggregation features include: Calculate the feature transfer weights between nodes based on the node aggregation characteristics, and generate node feature transfer paths based on the feature transfer weights. Based on the node characteristics and transmission path, the data transmission direction of each link in the supply chain is determined, and a data flow path chain is generated. Based on the data flow path chain, the carbon activity data of supply nodes and internal enterprise carbon activity data are used to perform full-link traceability and cumulative calculation to generate a full-link fusion path index.

[0054] Understandably, node aggregation features are node feature vectors obtained by the graph attention encoder after performing multi-scale feature aggregation on the initial features of nodes in the enhanced topology. They are used to comprehensively represent the carbon activity information and structural context information gathered by each node in the supply chain path. Feature propagation weights are used to quantify the influence of node aggregation features when they propagate along the enhanced topology between adjacent nodes. Node feature propagation paths are directed path sequences formed from upstream supply nodes through intermediate processing nodes to internal enterprise nodes under the constraint of feature propagation weights. Data flow path chains are data transmission direction links of each link in the supply chain abstracted based on node feature propagation paths. By calculating the feature propagation weights between nodes based on node aggregation features and generating node feature propagation paths, and then determining data flow path chains based on node feature propagation paths, and finally performing full-link traceability and accumulation calculations on carbon activity data of supply nodes and internal enterprise carbon activity data along the data flow path chains, a full-link fusion path index that conforms to the constraints of the enhanced topology and enterprise caliber rule table can be generated.

[0055] For example, in the enhanced topology, nodes S1, S2, and E1 correspond to raw material code M101, semi-finished product code M201, and finished product code M301, respectively. The node aggregation features output by the graph attention encoder can be set as vectors: S1:(15,7), S2:(12,6), E1:(20,9), where each component represents the carbon emission feature and energy consumption feature after multi-scale aggregation, respectively. When calculating the feature transfer weights between nodes based on the node aggregation features, the feature transfer weight from S1 to S2 is 0.8, from S2 to E1 is 0.9, and from S1 to E1 is 0.2. Based on this, The primary path is the node feature transmission path S1→S2→E1, while the secondary path is S1→E1. The data transmission direction of each link in the supply chain is determined based on the node feature transmission path. That is, the data of the upstream supply link S1 is preferentially transmitted to the internal link E1 of the enterprise via the processing link S2, forming a data flow path chain "upstream supply link → processing link → internal link of the enterprise". On this basis, the carbon activity data of the supply nodes corresponding to S1, S2 and E1 and the carbon activity data of the enterprise are traced and accumulated along the data flow path chain. For example, the carbon emissions of S1 and S2 are accumulated to the carbon emissions of E1 according to the weight, so as to obtain the full-process full-link fusion path index of finished product M301.

[0056] Specifically, the process of generating carbon fusion accounting results based on the end-to-end fusion path index is actually to use the index as a dynamic topology navigation map to drive the calculation. By using the end-to-end fusion path index, the physical carbon emission values ​​corresponding to the supplier's carbon activity records and the enterprise's internal carbon activity records can be accurately traced back and locked. The feature transfer weights obtained from the aforementioned calculation are used as dynamic carbon emission allocation coefficients to perform weighted calculations on the original emission data of each node, thereby determining the actual carbon load of each business link in the current accounting scenario.

[0057] In this embodiment, by introducing feature transfer weights, node feature transfer paths, and data flow path chains based on node aggregation features, compared to the existing technology that simply adds supply node carbon activity data and internal enterprise carbon activity data at the report level based on static field correspondence, this approach ensures that carbon data is transferred and accumulated along the real supply chain path, avoiding erroneous merging or omission of cross-link data. Furthermore, it explicitly reflects the node importance and path contribution learned in the graph attention encoder into the feature transfer weights, enabling the end-to-end fusion path index to maintain consistency and integrity of end-to-end traceability even under changes in supply chain structure and adjustments to internal enterprise accounting boundaries. This more accurately supports the path tracking and compliant disclosure of product-level carbon footprints.

[0058] Example 2 Please see Figure 2 As shown, based on the same inventive concept, this embodiment discloses a corporate full-business carbon fusion accounting system. For details not covered in this embodiment, please refer to the relevant parts of Embodiment 1. The system includes: Dependency generation module: used to obtain logistics document records, bill of materials records, supplier carbon activity records and internal enterprise carbon activity records, generate a material association mapping table based on logistics document records and bill of materials records, and perform field dependency analysis on supplier carbon activity records and internal enterprise carbon activity records based on the material association mapping table to generate a field dependency relationship table with cross-process constraints. Structure alignment generation module: used to obtain accounting voucher records, perform structure consistency checks based on field dependency relationship table and accounting voucher records, and generate structure alignment mapping including stability level; Calibration Generation Module: Used to perform multi-dimensional caliber difference screening based on the company's internal carbon activity records and preset accounting boundary tables, and generate a corporate caliber rule table with priority order annotation; The graph accounting encoding module is used to perform node adjacency topology reconstruction on the structure alignment mapping and enterprise caliber rule table, construct a graph attention encoder with embedded multi-scale local enhancement modules, generate initial node features based on the unified activity record stream, aggregate and encode the initial node features based on the graph attention encoder, generate the full-link fusion path index, and generate carbon fusion accounting results based on the full-link fusion path index.

[0059] The detailed description above, in conjunction with the accompanying drawings, describes examples but does not represent all examples that can be implemented or fall within the scope of the claims. The terms “example” and “exemplary” are used in this specification to mean “serving as an example, instance or illustration” and do not mean “superior to or better than other examples”.

[0060] Throughout this specification, the phrase "an embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. Therefore, the use of these phrases may refer to more than one embodiment. Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0061] It should also be noted that these embodiments may be described as processes depicted as flowcharts, structural diagrams, or block diagrams. Although a flowchart may describe the operations as sequential processes, many of these operations can be performed in parallel or concurrently, and the order of these operations may be rearranged.

Claims

1. A method for carbon fusion accounting across all business operations of an enterprise, characterized in that, include: Acquire logistics document records, bill of materials records, supplier carbon activity records, and internal enterprise carbon activity records. Generate a material association mapping table based on the logistics document records and bill of materials records. Perform field dependency analysis on the supplier carbon activity records and internal enterprise carbon activity records based on the material association mapping table to generate a field dependency relationship table with cross-process constraints. Obtain accounting voucher records, perform structural consistency checks based on the field dependency table and accounting voucher records, and generate a structural alignment mapping including stability grading; Based on the enterprise's internal carbon activity records and preset accounting boundary table, perform multi-dimensional caliber difference screening to generate an enterprise caliber rule table with priority order; Perform node adjacency topology reconstruction on the structure alignment mapping and enterprise caliber rule table, construct a graph attention encoder with embedded multi-scale local enhancement modules, generate initial node features based on the unified activity record stream, aggregate and encode the initial node features based on the graph attention encoder, generate a full-link fusion path index, and generate carbon fusion accounting results based on the full-link fusion path index.

2. The enterprise-wide carbon integration accounting method according to claim 1, characterized in that, A material association mapping table is generated based on logistics document records and bill of materials records, including: Spatiotemporal trajectory data is obtained from logistics document records, and material hierarchy data is obtained from bill of materials records. The node freight sequence representing the temporal relationship of transportation nodes is extracted from the spatiotemporal trajectory data, and the hierarchical relationship sequence of material supply is extracted from the material hierarchy data. The material transport chain sequence corresponding to each transport activity is determined based on the node freight sequence, and the supply chain sequence of the material supply process is determined based on the hierarchical relationship sequence. The material transport chain sequence is screened for effectiveness based on the supply chain sequence, and material transport chains that do not match the supply chain sequence are eliminated; Based on the selected material transportation chain and supply chain sequence, establish the association mapping relationship between nodes and materials to obtain the material association mapping table.

3. A method for enterprise-wide carbon integration accounting according to claim 2, characterized in that, Based on the material association mapping table, perform field dependency analysis on supplier carbon activity records and internal enterprise carbon activity records to generate a field dependency relationship table with cross-stage constraints, including: Based on the material association mapping table, extract the migration order of materials between different supply links and construct the supply chain flow sequence. The set of data fields that can participate in dependency analysis is defined based on the carbon activity data of supply nodes and the internal carbon activity data of enterprises in the process flow sequence. Construct a cross-stage dependency chain for fields based on the defined set of data fields, and generate a field dependency table based on the cross-stage dependency chain.

4. A method for enterprise-wide carbon integration accounting according to claim 3, characterized in that, Based on the field dependency table and accounting voucher data, a dual-tower validation model is constructed to generate a confidence mask matrix, including: Extract the set of field combinations to be validated from the field dependency table, and extract the corresponding set of business field combinations from the accounting voucher data; A dual-tower structure is constructed by combining the set of fields to be verified with the set of business fields to form multiple field combination correspondence pairs; Calculate the structural similarity distance between the corresponding pairs of field combinations, and determine the credibility of the field combination structure based on the similarity distance; Generate a confidence mask matrix based on the confidence level of each field combination to represent the confidence level of the field combination structure in the field dependency table.

5. A method for enterprise-wide carbon integration accounting according to claim 4, characterized in that, Methods for generating confidence mask matrices include: Extract the structural difference data of various field combinations in historical business scenarios and generate a set of structural difference data. Based on the historical frequency of structural changes in the structural difference data set, the structural reliability level of each field combination is determined. The confidence mask matrix is ​​modified according to the structural confidence level to generate the modified confidence mask matrix.

6. A method for enterprise-wide carbon integration accounting according to claim 5, characterized in that, Methods for constructing graph attention encoders with embedded multi-scale local enhancement modules include: Based on the corrected confidence mask matrix, low-confidence node connections are removed from the initial node topology formed by the structure alignment mapping to generate a pruned topology. The priority order of topology reconnection for different nodes in various business scenarios is determined based on the enterprise's rules table. Based on the topology reconnection priority, the pruned topology is reconnected to generate an enhanced topology. A graph attention encoder with embedded multi-scale local enhancement modules is constructed based on the enhanced topology.

7. A method for enterprise-wide carbon integration accounting according to claim 6, characterized in that, The multi-scale local enhancement module in the graph attention encoder includes: Neighborhood node sampling at different scales is performed on each node in the enhanced topology to form a neighborhood structure at multiple scales; Node feature aggregation is performed on the neighborhood structure at multiple scales to generate local enhancement features for nodes at each scale. By fusing node local enhancement features at different scales, multi-scale local enhancement features are generated for use by the image attention encoder.

8. A method for enterprise-wide carbon integration accounting according to claim 7, characterized in that, Methods for generating end-to-end fused path indexes by aggregating and encoding initial node features based on graph attention encoders include: Based on carbon activity data from supply nodes and internal enterprise carbon activity data, a set of original node characteristics is constructed. By utilizing the enhanced topology, cross-link feature alignment is performed on the original feature set of nodes to generate a node-aligned feature set. Initial node features are generated based on the node alignment feature set, and node feature aggregation encoding is performed on the initial node features using the multi-scale local enhancement module in the graph attention encoder to obtain node aggregate features; Generate a full-link fusion path index based on node aggregation characteristics.

9. A method for enterprise-wide carbon integration accounting according to claim 8, characterized in that, Methods for generating a full-link fusion path index based on node aggregation features include: Calculate the feature transfer weights between nodes based on the node aggregation characteristics, and generate node feature transfer paths based on the feature transfer weights. Based on the node characteristics and transmission path, the data transmission direction of each link in the supply chain is determined, and a data flow path chain is generated. Based on the data flow path chain, the carbon activity data of supply nodes and internal enterprise carbon activity data are used to perform full-link traceability and cumulative calculation to generate a full-link fusion path index.

10. A carbon fusion accounting system for all enterprise businesses, used to implement the carbon fusion accounting method for all enterprise businesses according to any one of claims 1-9, characterized in that, include: Dependency generation module: used to obtain logistics document records, bill of materials records, supplier carbon activity records and internal enterprise carbon activity records, generate a material association mapping table based on logistics document records and bill of materials records, and perform field dependency analysis on supplier carbon activity records and internal enterprise carbon activity records based on the material association mapping table to generate a field dependency relationship table with cross-process constraints. Structure alignment generation module: used to obtain accounting voucher records, perform structure consistency checks based on field dependency relationship table and accounting voucher records, and generate structure alignment mapping including stability level; Calibration Generation Module: Used to perform multi-dimensional caliber difference screening based on the company's internal carbon activity records and preset accounting boundary tables, and generate a corporate caliber rule table with priority order annotation; The graph accounting encoding module is used to perform node adjacency topology reconstruction on the structure alignment mapping and enterprise caliber rule table, construct a graph attention encoder with embedded multi-scale local enhancement modules, generate initial node features based on the unified activity record stream, aggregate and encode the initial node features based on the graph attention encoder, generate the full-link fusion path index, and generate carbon fusion accounting results based on the full-link fusion path index.

Images