Enterprise carbon information disclosure standard adaptation method and system

Abstract

The invention provides an enterprise carbon information disclosure standard adaptation method and system, and the method comprises the steps: carrying out the structural analysis and semantic extraction of a carbon information disclosure standard list, obtaining a clause standard semantic set, constructing a clause knowledge graph corresponding to the clause standard semantic set, carrying out the entity analysis and feature extraction of enterprise carbon monitoring data, and obtaining a clause knowledge graph corresponding to the clause standard semantic set; a carbon monitoring feature set is obtained, the carbon monitoring feature set is mapped to a clause knowledge graph based on a double-tower alignment algorithm, an enterprise carbon information graph is obtained, missing field prediction and context complementation are carried out on the enterprise carbon information graph, and a complemented carbon information graph is obtained; performing time sequence prediction on the complemented carbon information atlas based on a conditional diffusion model, and performing soft and hard constraint optimization on the complemented carbon information atlas after time sequence conditional diffusion based on a real-time disclosure standard to obtain an optimized carbon information atlas; and performing standard field screening and carbon information disclosure on the optimized carbon information atlas according to a real-time disclosure standard to obtain enterprise disclosure carbon information.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for adapting corporate carbon information disclosure standards. Background Technology

[0002] Carbon information disclosure is a core measure for enterprises to address climate risks, fulfill environmental responsibilities, and meet regulatory requirements. The accuracy, compliance, and forward-looking nature of the information directly affect a company's sustainable reputation and market competitiveness. With the development of knowledge graph and deep learning technologies, building intelligent carbon information management platforms has become an important technological direction for improving the quality and efficiency of disclosure.

[0003] However, current mainstream carbon information processing and disclosure methods still face significant technical bottlenecks in addressing the rapidly evolving compliance environment and the complex data realities of enterprises. First, existing methods are mostly designed for single or static disclosure standards, making it difficult to dynamically adapt to multiple coexisting and frequently updated international and domestic standard systems such as TCFD, ISSB, and GRI. This forces companies to repeatedly prepare data and compile reports when facing different regulators or investors, resulting in inefficiency and inconsistencies. Second, most technical frameworks focus on the calculation and reporting of historical emissions data, lacking the ability to deeply integrate real-time monitoring data, production logic, and future scenario predictions. This prevents the generation of forward-looking compliance disclosure information with risk warning and strategic insight value, causing disclosure reports to lag behind management decisions. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a corporate carbon information disclosure standard adaptation system, which has the advantages of dynamic standard adaptation and forward-looking disclosure. It solves the problem of low disclosure efficiency caused by the complexity and variability of standard systems, the heterogeneity of corporate data from multiple sources, and the lack of cross-standard adaptation and future scenario modeling capabilities in existing methods in carbon information disclosure scenarios.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for adapting to corporate carbon information disclosure standards, comprising the following steps: The pre-acquired list of carbon information disclosure standards is structured and semantically extracted to obtain a semantic set of clause standards. A clause knowledge graph corresponding to the semantic set of clause standards is then constructed based on a semantic clustering algorithm. Entity parsing and feature extraction are performed on the pre-acquired corporate carbon monitoring data to obtain a carbon monitoring feature set. The carbon monitoring feature set is then mapped to the clause knowledge graph based on a bidirectional attention matching algorithm to obtain a corporate carbon information graph. Based on the graph message passing mechanism, the missing field prediction and context completion of the enterprise carbon information map are performed to obtain the complete carbon information map. Obtain the real-time disclosure standard, perform graph structure-guided temporal conditional diffusion on the completed carbon information map based on the structured constraints of the real-time disclosure standard, and perform soft and hard constraint optimization on the completed carbon information map after temporal conditional diffusion to obtain the optimized carbon information map; Carbon information is disclosed by performing carbon information disclosure on the optimized carbon information map in accordance with the real-time disclosure standards, thereby obtaining the carbon information disclosed by the enterprise.

[0006] According to a preferred embodiment of the present invention, a pre-acquired list of carbon information disclosure standards is subjected to structured parsing and semantic extraction to obtain a clause standard semantic set, and a clause knowledge graph corresponding to the clause standard semantic set is constructed based on a semantic clustering algorithm. Entity parsing and feature extraction are performed on the pre-acquired corporate carbon monitoring data to obtain a carbon monitoring feature set. The carbon monitoring feature set is then mapped to the clause knowledge graph based on a bidirectional attention matching algorithm to obtain a corporate carbon information graph. Based on the graph message passing mechanism, the missing field prediction and context completion of the enterprise carbon information map are performed to obtain the complete carbon information map. Obtain the real-time disclosure standard, perform graph structure-guided temporal conditional diffusion on the completed carbon information map based on the structured constraints of the real-time disclosure standard, and perform soft and hard constraint optimization on the completed carbon information map after temporal conditional diffusion to obtain the optimized carbon information map; Carbon information is disclosed by performing carbon information disclosure on the optimized carbon information map in accordance with the real-time disclosure standards, thereby obtaining the carbon information disclosed by the enterprise.

[0007] According to another preferred embodiment of the present invention, the step of performing structured parsing and semantic extraction on the pre-acquired list of carbon information disclosure standards to obtain a semantic set of clause standards includes: The text format of the pre-obtained list of carbon information disclosure standards is standardized to obtain a unified disclosure standard list; The hierarchical structure of each disclosure standard in the unified disclosure standard list is parsed to obtain a standard structured text set; The standard structured text set is segmented into sentences and semantic units to obtain a set of clause semantic units; Entity recognition is performed on the set of semantic units of the clauses to obtain a standard entity set, and relation extraction is performed on the standard entity set to obtain an entity relation set; The standard entity set is semantically aligned and vectorized using the entity relationship set to obtain the standard semantic set of terms.

[0008] According to another preferred embodiment of the present invention, the step of constructing the clause knowledge graph corresponding to the clause standard semantic set based on the semantic clustering algorithm includes: Semantic similarity is calculated for each clause standard semantic in the aforementioned clause standard semantic set to obtain a semantic similarity matrix; Based on the semantic similarity matrix, density clustering is performed on the standard semantics of each clause to obtain a clause topic cluster set; Cluster center calculation is performed on each clause topic cluster in the clause topic cluster set to obtain the entity node set; Semantic edges are constructed for the entity node set based on the entity relationships corresponding to the standard semantics of each clause in each clause theme cluster, resulting in semantic association edges. The semantic association edges are then weighted according to the number of entity relationships across entity nodes corresponding to each semantic association edge, resulting in weighted association edges. Based on the weighted association edges and the entity node set, a topic clause graph structure is generated, and the standard semantics of each clause are mapped to the topic clause graph structure to obtain a clause knowledge graph.

[0009] According to another preferred embodiment of the present invention, the step of performing entity parsing and feature extraction on the pre-acquired enterprise carbon monitoring data to obtain a carbon monitoring feature set includes: Data structure identification and structured parsing are performed on the pre-acquired enterprise carbon monitoring data to obtain parsed monitoring data; The units of the analyzed monitoring data are standardized and the data is cleaned. The field formats of the cleaned analyzed monitoring data are then standardized to obtain standard monitoring data. The standard monitoring data is subjected to attribute field identification and carbon emission entity identification to obtain a primary carbon emission entity set. The primary carbon emission entity set is then subjected to semantic alignment and semantic disambiguation to obtain a standard carbon emission entity set. Based on the standard monitoring data, the standard carbon emission entity set is subjected to relation extraction to obtain a primary entity relation set, and the primary entity relation set is subjected to time-series dependency parsing to obtain an entity link relation set. A monitoring data sequence set is constructed based on the standard monitoring data and the entity link relationship set, and time-series features are extracted from the monitoring data sequence set to obtain a monitoring time-series feature set. The standard carbon emission entity set is embedded into the monitoring time series feature set to obtain the carbon monitoring feature set.

[0010] According to another preferred embodiment of the present invention, the step of mapping the carbon monitoring feature set to the clause knowledge graph based on the bidirectional attention matching algorithm to obtain the enterprise carbon information graph includes: Dual-tower encoding is performed on the carbon monitoring feature set and the entity node set of the clause knowledge graph to obtain the encoded monitoring feature set and the encoded node feature set, respectively. Calculate the first attention weight matrix of the encoding monitoring feature set relative to the encoding node feature set and the second attention weight matrix of the encoding node feature set relative to the encoding monitoring feature set; Based on the first attention weight, context weight enhancement and feature pooling operations are performed on the encoded monitoring feature set to obtain an enhanced monitoring feature set; and based on the second attention weight, context weight enhancement and feature pooling operations are performed on the encoded node feature set to obtain an enhanced node feature set. Calculate the feature similarity matrix between the enhanced monitoring feature set and the enhanced node feature set, and generate a similarity mapping relationship set based on the feature similarity matrix; Based on the similarity mapping relationship set, the carbon monitoring feature set is mapped to the corresponding entity nodes of the clause knowledge graph to obtain the enterprise carbon information graph.

[0011] According to another preferred embodiment of the present invention, the step of predicting missing fields and completing context based on the graph messaging mechanism to obtain a completed carbon information map includes: The missing field set is obtained by scanning the enterprise carbon information map through traversal. Select the target missing fields one by one from the missing field set, perform semantic relationship filtering and neighborhood sampling on the target missing fields, and obtain a neighborhood information graph; Each entity node in the neighborhood information graph is encoded with node features to obtain a neighborhood node feature set. Then, based on an attention mechanism, the neighborhood node feature set is subjected to cross-type message passing and multi-semantic feature aggregation to obtain aggregated node features. The target missing field is predicted using the features of the aggregated node to obtain the field prediction result. The field prediction result is then weighted by confidence and conflict resolution is performed to obtain the target filling field. The target missing fields in the enterprise carbon information map are filled in using the target filling fields until the target missing field is the last missing field in the missing field set. The filled enterprise carbon information map is then used as the complete carbon information map.

[0012] According to another preferred embodiment of the present invention, the time-series conditional diffusion of the graph structure-guided carbon information map based on the structured constraints of the real-time disclosure standard includes: Temporal features are extracted from each entity node in the completed carbon information map to obtain an entity temporal feature sequence. The real-time disclosure standard is structured and parsed to obtain the parsed real-time standard, and disclosure constraints are extracted from the parsed real-time standard. The completed carbon information map is graph structure encoded to obtain the map structure conditions; The graph structure conditions and the disclosure constraints are used as diffusion conditions to perform forward diffusion on the entity time-series feature sequence to obtain a noisy entity feature sequence. Based on the diffusion conditions, the noisy entity feature sequence is back-diffused to obtain the predicted carbon emission time series sequence, and the predicted carbon emission time series sequence is mapped to each entity node of the completed carbon information map.

[0013] According to another preferred embodiment of the present invention, the step of forward diffusing the entity temporal feature sequence with the graph structure conditions and the disclosure constraints as diffusion conditions to obtain a temporal noise sequence includes: The entity time series feature sequence is subjected to feature standardization, and the standardized entity time series feature sequence set is subjected to trend filtering to obtain the initial time series state; A conditional noise level schedule is generated based on the disclosure constraints, and a disclosure-guided noise schedule is extracted from the conditional noise level schedule. Based on the aforementioned spectrum structure conditions, a diffusion covariance structure is constructed to obtain a structural diffusion kernel. Then, based on the cross-attention mechanism, a conditional diffusion distribution is constructed according to the disclosure-guided noise scheduling and the structural diffusion kernel. Based on the conditional diffusion distribution, the initial time series state is forward diffused according to a preset diffusion step size to obtain a time series noise-added sequence, and the time series noise sequence corresponding to the time series noise-added sequence is calculated based on the reparameterization algorithm.

[0014] According to another preferred embodiment of the present invention, the step of performing soft and hard constraint optimization on the completed carbon information map after time-series conditional diffusion to obtain an optimized carbon information map includes: The predicted carbon emission time series is extracted from the completed carbon information map after the time series condition diffusion. The predicted carbon emission time series is then compared for rule consistency and the difference is quantified according to the disclosure constraints to obtain the difference marker sequence. A sliding window aggregation is performed on the differential marker sequence to obtain a trend offset vector; The trend offset vector is inverted and its amplitude is limited to obtain a correction guide sequence. The predicted carbon emission time series is then conditionally resampled based on the correction guide sequence to obtain a corrected carbon emission time series. The corrected carbon emission time series is mapped to each entity node of the completed carbon information map to obtain the optimized carbon information map.

[0015] To achieve at least one of the above-mentioned objectives, the present invention further provides an enterprise carbon information disclosure standard adaptation system, the system comprising a standard extraction module, a spectrum mapping module, a field completion module, a time-series diffusion module, and an information disclosure module, wherein: The standard extraction module performs structured parsing and semantic extraction on the pre-acquired list of carbon information disclosure standards to obtain a clause standard semantic set, and constructs a clause knowledge graph corresponding to the clause standard semantic set based on a semantic clustering algorithm. The graph mapping module performs entity parsing and feature extraction on the pre-acquired corporate carbon monitoring data to obtain a carbon monitoring feature set, and maps the carbon monitoring feature set to the clause knowledge graph based on a bidirectional attention matching algorithm to obtain a corporate carbon information graph. The field completion module performs missing field prediction and context completion on the enterprise carbon information map based on the graph message passing mechanism to obtain a completed carbon information map. The time-series diffusion module obtains the real-time disclosure standard, performs graph-structure-guided time-series conditional diffusion on the completed carbon information map based on the structured constraints of the real-time disclosure standard, and optimizes the completed carbon information map after time-series conditional diffusion by soft and hard constraints to obtain an optimized carbon information map. The information disclosure module discloses carbon information from the optimized carbon information map according to the real-time disclosure standard, thereby obtaining the carbon information disclosed by the enterprise.

[0016] The present invention further provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement the above-described method for adapting corporate carbon information disclosure standards.

[0017] (III) Beneficial Effects Compared with existing technologies, the present invention provides a method and system for adapting corporate carbon information disclosure standards, which has the following beneficial effects: This corporate carbon information disclosure standard adaptation method performs structured parsing, semantic extraction, and clustering of carbon information disclosure standards from multiple sources and in multiple formats. This transforms the originally loose and heterogeneous cross-standard clause information into a unified semantic expression system. By unifying text format, hierarchical parsing, and semantic unit decomposition, the structural consistency of standard clauses is ensured. Entity recognition, relation extraction, and semantic alignment further unify the standard semantics at the conceptual level, laying the foundation for cross-standard logical alignment. By constructing entity nodes based on density clustering and generating weighted semantic edges using entity relationships, a clause knowledge graph with semantic aggregation and structural expression capabilities can be formed. This reveals the semantic relationships between different disclosure standards and provides a standardized and reasonable knowledge framework for subsequent mapping of corporate data to standard clauses and completion of missing fields.

[0018] This corporate carbon information disclosure standard adaptation method effectively captures semantic dependencies between various types of nodes, including enterprises, emission items, and clauses, through semantic link filtering, neighborhood sampling, and heterogeneous attention aggregation. This ensures contextual consistency in the prediction results. By introducing confidence weighting and soft update conflict resolution, it avoids noise fields from polluting the graph, improving the credibility and stability of the supplementary data. By explicitly transforming real-time disclosure standards into quantifiable disclosure constraints and embedding them into a conditional diffusion framework, the predicted future carbon emission time series not only maintains the physical consistency of historical evolution but also automatically aligns with disclosure requirements during the generation stage. By using noise scheduling to adjust the confidence of non-compliant intervals during the forward diffusion stage, it enhances the model's sensitivity to key disclosure items. Soft and hard constraint optimization achieves rule-based rationality correction through difference quantification, trend offset extraction, and conditional resampling, rather than simply modifying numerical values ​​to meet standards. This ensures the authenticity and interpretability of the disclosed information. By integrating predictions with standards, it can not only provide early warnings of carbon emission risks but also proactively plan emission reduction paths, thereby improving the completeness, compliance, and usability of carbon information disclosure.

[0019] This corporate carbon information disclosure standard adaptation method achieves machine readability of disclosure requirements, field logic, and format specifications through structured parsing. Based on the parsed disclosure entity list and logical rules, corresponding fields can be accurately extracted from the optimized carbon information map, achieving highly matched entity-level carbon emission data reporting. Through logical compliance conversion, it ensures that the extracted fields fully comply with the disclosure rules in terms of value range, total amount relationship, and proportion relationship, avoiding logical contradictions that may occur in manual reporting. This improves the company's response efficiency to different regulatory and market requirements, and multiple compliance reports can be generated from a single data processing, greatly reducing compliance costs. This improves the flexibility and comprehensiveness of carbon information disclosure adaptation and enhances the efficiency of corporate carbon information disclosure. Attached Figure Description

[0020] Figure 1 The diagram shown is a flowchart of a method for adapting corporate carbon information disclosure standards according to the present invention. Detailed Implementation

[0021] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0022] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0023] Example 1: Please combine Figure 1 This invention discloses a method for adapting to corporate carbon information disclosure standards, the method comprising the following steps: The pre-acquired list of carbon information disclosure standards is structured and semantically extracted to obtain a semantic set of clause standards. Based on a semantic clustering algorithm, a clause knowledge graph corresponding to the semantic set of clause standards is constructed.

[0024] The carbon information disclosure standards in the aforementioned list refer to unified rules and standards used to calculate and disclose corporate greenhouse gas emissions and climate-related risks. These include corporate greenhouse gas emission accounting and reporting requirements, Task Force on Climate-related Financial Disclosures (TCFD), International Financial Reporting Standards S2 Climate-related Disclosures (IFRS S2), and Global Reporting Initiative Standards (GRI). Because there are different standards for carbon information disclosure, companies need to adapt to different disclosure standards when making carbon information disclosures.

[0025] Specifically, the process of performing structured parsing and semantic extraction on the pre-acquired list of carbon information disclosure standards to obtain a semantic set of clause standards includes: The text format of the pre-obtained list of carbon information disclosure standards is standardized to obtain a unified disclosure standard list; The hierarchical structure of each disclosure standard in the unified disclosure standard list is parsed to obtain a standard structured text set; The standard structured text set is segmented into sentences and semantic units to obtain a set of clause semantic units; Entity recognition is performed on the set of semantic units of the clauses to obtain a standard entity set, and relation extraction is performed on the standard entity set to obtain an entity relation set; The standard entity set is semantically aligned and vectorized using the entity relationship set to obtain the standard semantic set of terms.

[0026] The text format unification refers to converting carbon information disclosure standards in different formats into a unified text format that facilitates subsequent processing. Carbon information disclosure standards exist in various formats such as PDF, HTML, and Word. For example, tools such as Marker, Mammoth, and Pandoc can be used to uniformly convert the carbon information disclosure standards into Markdown text format. Regular expressions can be used for hierarchical structure parsing to generate a hierarchical tree structured text. The text segmentation refers to using syntactic analysis tools such as SpaCy or Stanza to split sentences in each standard structured text. The semantic unit partitioning refers to dividing the segmented standard structured text into the smallest semantic units, each containing sentences, conditions, requirements, and indicators. The entity recognition refers to extracting carbon disclosure-specific entities from the clause semantic units, such as emission range, emission type, unit of measurement, and behavioral actions. The relation extraction refers to identifying the structural relationships between entities, such as "company," "disclosure," and "scope." The sequential relationship between "1 emission" can be achieved by using semantic analysis models such as Transformer or BERT for semantic unit segmentation, entity recognition, and relation extraction. Semantic alignment refers to unifying semantic entities with different expressions but the same concept. Similarity algorithms can be used for semantic alignment, and the encoding layer of the BERT model can be used for vectorized encoding.

[0027] In detail, the construction of the clause knowledge graph corresponding to the standard semantic set of clauses based on the semantic clustering algorithm includes: Semantic similarity is calculated for each clause standard semantic in the aforementioned clause standard semantic set to obtain a semantic similarity matrix; Based on the semantic similarity matrix, density clustering is performed on the standard semantics of each clause to obtain a clause topic cluster set; Cluster center calculation is performed on each clause topic cluster in the clause topic cluster set to obtain the entity node set; Semantic edges are constructed for the entity node set based on the entity relationships corresponding to the standard semantics of each clause in each clause theme cluster, resulting in semantic association edges. The semantic association edges are then weighted according to the number of entity relationships across entity nodes corresponding to each semantic association edge, resulting in weighted association edges. Based on the weighted association edges and the entity node set, a topic clause graph structure is generated, and the standard semantics of each clause are mapped to the topic clause graph structure to obtain a clause knowledge graph.

[0028] Semantic similarity can be calculated using a cosine similarity algorithm. The density clustering can be DBSCAN density clustering. The cluster center calculation refers to taking the clause standard semantic closest to the cluster center of each clause topic cluster as the corresponding entity node. The entity relationship serves as the basis for cross-topic semantic connection between clauses. Therefore, semantic association edges can be constructed between two related entity nodes. The edge weighting of the semantic association edges based on the number of entity relationships across entity nodes corresponding to each semantic association edge refers to counting the number of all entity relationship pairs connecting two entity nodes and normalizing this number as the edge weight.

[0029] By performing structured analysis, semantic extraction, and clustering of carbon information disclosure standards from multiple sources and in multiple formats, the originally loose and heterogeneous clause information across standards is transformed into a unified semantic expression system. The consistency of standard clauses in structure is ensured through text format unification, hierarchical parsing, and semantic unit decomposition. Entity identification, relation extraction, and semantic alignment further unify standard semantics at the conceptual level, laying the foundation for cross-standard logical alignment. By constructing entity nodes based on density clustering and generating weighted semantic edges using entity relationships, a clause knowledge graph with semantic aggregation and structural expression capabilities can be formed. This reveals the semantic connections between different disclosure standards and provides a standardized and reasonable knowledge framework for subsequent mapping of enterprise data to standard clauses and completion of missing fields.

[0030] Entity parsing and feature extraction are performed on the pre-acquired corporate carbon monitoring data to obtain a carbon monitoring feature set. The carbon monitoring feature set is then mapped to the clause knowledge graph based on a bidirectional attention matching algorithm to obtain a corporate carbon information graph.

[0031] The enterprise carbon monitoring data refers to data obtained from multi-faceted carbon monitoring of enterprise production activities, including carbon monitoring data from various equipment sensors, energy consumption systems, metering instruments, reports, logs, and IoT platforms. Due to the diverse sources of enterprise carbon monitoring data, entity parsing and feature extraction are required to achieve unified processing.

[0032] In this embodiment of the invention, the step of performing entity parsing and feature extraction on the pre-acquired enterprise carbon monitoring data to obtain a carbon monitoring feature set includes: Data structure identification and structured parsing are performed on the pre-acquired enterprise carbon monitoring data to obtain parsed monitoring data; The units of the analyzed monitoring data are standardized and the data is cleaned. The field formats of the cleaned analyzed monitoring data are then standardized to obtain standard monitoring data. The standard monitoring data is subjected to attribute field identification and carbon emission entity identification to obtain a primary carbon emission entity set. The primary carbon emission entity set is then subjected to semantic alignment and semantic disambiguation to obtain a standard carbon emission entity set. Based on the standard monitoring data, the standard carbon emission entity set is subjected to relation extraction to obtain a primary entity relation set, and the primary entity relation set is subjected to time-series dependency parsing to obtain an entity link relation set. A monitoring data sequence set is constructed based on the standard monitoring data and the entity link relationship set, and time-series features are extracted from the monitoring data sequence set to obtain a monitoring time-series feature set. The standard carbon emission entity set is embedded into the monitoring time series feature set to obtain the carbon monitoring feature set.

[0033] The data structure identification refers to identifying the structured data patterns of data formats from different monitoring sources. For example, data structure can be identified based on file extensions. The structured parsing refers to extracting the structural information and internal data of enterprise carbon monitoring data and converting it into a unified structured representation. For example, a data warehouse can be used to read and parse JSON or CSV data. The unit unification refers to unifying the units of power, energy consumption, fuel consumption, and temperature. The data cleaning includes outlier detection and missing value imputation. For example, outlier detection can be performed using methods such as IQR or Z-score, and missing value imputation can be performed using methods such as mean interpolation. The field format unification refers to standardizing the format of fields such as date and enumeration values. The attribute field identification can be performed using regular expressions, that is, extracting attribute fields from the standard monitoring data. Carbon emission entity identification can be performed using a bidirectional long-short-term network conditional random field.

[0034] Specifically, the semantic alignment refers to unifying entities with different expressions such as "CO2" and "carbon dioxide" into the same expression; the semantic disambiguation refers to distinguishing the semantics of different meanings of the same expression through contextual semantic analysis; the temporal dependency resolution includes temporal alignment of each primary entity relationship based on the timestamps in the standard monitoring data, and identifying the temporal dependency links between entities corresponding to the entity relationships, thereby obtaining an entity link relationship set; the monitoring data sequence set is a data sequence constructed for each emission source in the standard monitoring data according to the temporal order of each entity link relationship; the temporal feature extraction refers to calculating the temporal change trend features such as mean and skewness of each monitoring data sequence, as well as features such as frequency domain features; and embedding the standard carbon emission entity set into the monitoring temporal feature set refers to vectorizing the standard carbon emission entity set into an entity feature set, and embedding each entity feature into each monitoring temporal feature set through feature concatenation or attention weighted fusion.

[0035] In detail, the step of mapping the carbon monitoring feature set to the clause knowledge graph based on the bidirectional attention matching algorithm to obtain the enterprise carbon information graph includes: Dual-tower encoding is performed on the carbon monitoring feature set and the entity node set of the clause knowledge graph to obtain the encoded monitoring feature set and the encoded node feature set, respectively. Calculate the first attention weight matrix of the encoding monitoring feature set relative to the encoding node feature set and the second attention weight matrix of the encoding node feature set relative to the encoding monitoring feature set; Based on the first attention weight, context weight enhancement and feature pooling operations are performed on the encoded monitoring feature set to obtain an enhanced monitoring feature set; and based on the second attention weight, context weight enhancement and feature pooling operations are performed on the encoded node feature set to obtain an enhanced node feature set. Calculate the feature similarity matrix between the enhanced monitoring feature set and the enhanced node feature set, and generate a similarity mapping relationship set based on the feature similarity matrix; Based on the similarity mapping relationship set, the carbon monitoring feature set is mapped to the corresponding entity nodes of the clause knowledge graph to obtain the enterprise carbon information graph.

[0036] The dual-tower encoding refers to using the left tower of a pre-trained dual-tower model to encode the carbon monitoring feature set and the right tower to encode the entity node set. The dual-tower model consists of two encoder models trained using positive and negative samples labeled with the correspondence between entity nodes and carbon monitoring features. After dual-tower encoding, the encoded monitoring feature set and the encoded node feature set have the same feature dimension, facilitating subsequent attention weight calculation. The first attention weight matrix is ​​calculated by generating a query vector from the encoded monitoring feature set and a key-value vector from the encoded node feature set, using an attention mechanism. The second... The attention weight matrix is ​​a weight matrix calculated by generating a query vector from the encoding node feature set and a key-value vector from the encoding monitoring feature set, and then performing an attention mechanism. The context weight enhancement refers to performing attention-weighted summation on the encoding monitoring feature set using the attention distribution corresponding to the first attention weight, and performing attention-weighted summation on the encoding node feature set using the attention distribution corresponding to the second attention weight. The feature pooling refers to compressing the feature dimensions of the encoding monitoring feature set and the encoding node feature set after context weight enhancement using a pooling layer, thereby reducing computational complexity and improving mapping efficiency. The cosine similarity algorithm can be used to calculate the feature similarity matrix.

[0037] By performing structured parsing, format standardization, entity recognition, and semantic normalization on enterprise carbon monitoring data, consistency in semantics and structure can be achieved among monitoring data from different monitoring devices and systems. This avoids processing difficulties caused by data redundancy, field conflicts, and dimensional inconsistencies. Entity links established through relation extraction and temporal dependency parsing can accurately depict the evolution of emission behavior over time, providing a complete context for subsequent temporal feature extraction. The entity feature embedding mechanism further enhances the ability to express the correlation between emission entities and temporal changes. Precise alignment between monitoring features and entity nodes in the regulatory knowledge graph is achieved through dual-tower coding and bidirectional attention mechanisms. While preserving the semantic structure of both, cross-domain associations are strengthened, enabling accurate mapping of enterprise monitoring behavior to the regulatory framework. This improves the standardization of data fusion, the accuracy of semantic alignment, and the accuracy of graph construction.

[0038] Based on the graph message passing mechanism, the missing fields of the enterprise carbon information map are predicted and context is completed to obtain the completed carbon information map.

[0039] In particular, since there may be missing or uncollected carbon data in the corporate carbon monitoring data, there may be missing fields in the corporate carbon information map, which will cause errors in subsequent time series prediction analysis. In order to further improve the quality of carbon disclosure data, it is necessary to perform missing field prediction and context completion operations.

[0040] In detail, the method of predicting missing fields and completing context in the enterprise carbon information map based on the graph messaging mechanism to obtain a complete carbon information map includes: The missing field set is obtained by scanning the enterprise carbon information map through traversal. Select the target missing fields one by one from the missing field set, perform semantic relationship filtering and neighborhood sampling on the target missing fields, and obtain a neighborhood information graph; Each entity node in the neighborhood information graph is encoded with node features to obtain a neighborhood node feature set. Then, based on an attention mechanism, the neighborhood node feature set is subjected to cross-type message passing and multi-semantic feature aggregation to obtain aggregated node features. The target missing field is predicted using the features of the aggregated node to obtain the field prediction result. The field prediction result is then weighted by confidence and conflict resolution is performed to obtain the target filling field. The target missing fields in the enterprise carbon information map are filled in using the target filling fields until the target missing field is the last missing field in the missing field set. The filled enterprise carbon information map is then used as the complete carbon information map.

[0041] The semantic relationship filtering refers to filtering out the entity link relationships where the target missing field is located. The neighborhood sampling refers to using the subgraph structure corresponding to all entity nodes and weighted associated edges in the entity link relationship as a neighborhood information graph. The node feature encoding refers to generating corresponding neighborhood node features based on the semantic features of each entity node and the mapped carbon monitoring features. The cross-type message passing refers to using a heterogeneous attention message passing method to obtain the attention message vectors between node features in different domains, and then aggregating semantic features through graph convolution. A pre-trained lightweight multilayer perceptron model can be used for field prediction. The confidence weighting refers to weighted fusion based on the confidence of all fields in the field prediction results. Conflict resolution can be achieved using a soft update method, i.e., weighted averaging of conflicting fields to make the final result smoothly converge with the predicted value.

[0042] By performing semantic link filtering, neighborhood sampling, and heterogeneous attention aggregation, the semantic dependencies between various types of nodes, such as enterprises, emission items, and clauses, can be effectively captured, making the prediction results contextually consistent. By introducing confidence weighting and soft update conflict resolution, the graph can be prevented from being polluted by noisy fields, improving the credibility and stability of the supplementary data, thereby enhancing the integrity of carbon data. It can also significantly enhance the accuracy of subsequent time series prediction, anomaly detection, and clause compliance analysis.

[0043] The real-time disclosure standard is obtained, and the structured constraints of the real-time disclosure standard are used to perform graph structure-guided temporal conditional diffusion on the completed carbon information map. The completed carbon information map after temporal conditional diffusion is then optimized by soft and hard constraints to obtain an optimized carbon information map.

[0044] In this embodiment of the invention, the real-time disclosure standard refers to the carbon emission disclosure standard that enterprises need to follow in accordance with real-time project requirements or existing disclosure regulations. The real-time disclosure standard can be one of the standards such as TCFD, IFRS S2, or GRI, or an updated version of these standards. For example, the classification standard 32 for the power industry in IFRS S2. In actual disclosure, the carbon emission data of enterprises often has a certain time delay. For example, the power industry needs to design the carbon information disclosure for several years to come, and there is also the requirement for future information prediction that "the proportion of coal power will drop to a certain percentage by 2030". Therefore, it is necessary to perform time-series condition diffusion to generate more comprehensive carbon information disclosure.

[0045] In detail, the temporal conditional diffusion of the graph structure-guided carbon information map based on the structured constraints of the real-time disclosure standard includes: Temporal features are extracted from each entity node in the completed carbon information map to obtain an entity temporal feature sequence. The real-time disclosure standard is structured and parsed to obtain the parsed real-time standard, and disclosure constraints are extracted from the parsed real-time standard. The completed carbon information map is graph structure encoded to obtain the map structure conditions; The graph structure conditions and the disclosure constraints are used as diffusion conditions to perform forward diffusion on the entity time-series feature sequence to obtain a noisy entity feature sequence. Based on the diffusion conditions, the noisy entity feature sequence is back-diffused to obtain the predicted carbon emission time series sequence, and the predicted carbon emission time series sequence is mapped to each entity node of the completed carbon information map.

[0046] The temporal feature extraction refers to extracting temporal features based on the entity link relationships corresponding to each entity node using temporal network models such as long short-term memory neural networks. The structured parsing method is the same as the steps described above and will not be repeated here. Extracting disclosure constraints from the real-time parsing standard refers to extracting standard entities from the real-time parsing standard, converting the clauses in each real-time parsing standard into constraints such as numerical ranges, equation relationships, and classification labels based on the standard entities, and then using a unified vectorized format encoding to aggregate all constraints into disclosure constraints. The graph structure encoding refers to using a graph neural network to encode the structure of the completed carbon information graph and the information of each entity node into a feature vector, and using this feature vector as the graph structure condition.

[0047] Specifically, the step of forward-diffusion of the entity temporal feature sequence with the graph structure conditions and the disclosure constraints as diffusion conditions to obtain a temporal noise sequence includes: The entity time series feature sequence is subjected to feature standardization, and the standardized entity time series feature sequence set is subjected to trend filtering to obtain the initial time series state; A conditional noise level schedule is generated based on the disclosure constraints, and a disclosure-guided noise schedule is extracted from the conditional noise level schedule. Based on the aforementioned spectrum structure conditions, a diffusion covariance structure is constructed to obtain a structural diffusion kernel. Then, based on the cross-attention mechanism, a conditional diffusion distribution is constructed according to the disclosure-guided noise scheduling and the structural diffusion kernel. Based on the conditional diffusion distribution, the initial time series state is forward diffused according to a preset diffusion step size to obtain a time series noise-added sequence, and the time series noise sequence corresponding to the time series noise-added sequence is calculated based on the reparameterization algorithm.

[0048] The feature standardization method can be Z-score feature standardization, or trend filtering can be performed using first-order trend filtering, STL decomposition filtering, or Hodrick-Prescott filtering. First-order trend filtering can better preserve the inflection points in the sequence. The initial time series state is the starting state of diffusion. The conditional noise level scheduling refers to... Noise scheduling, wherein disclosure-guided noise scheduling refers to constraint-guided noise scheduling. Noise scheduling, specifically the noise level scheduling based on the disclosure constraints, refers to setting high noise for intervals that do not meet the confidence level of the constraints and low noise for intervals that do meet the confidence level. High noise weakens the model's confidence level on future values ​​that do not meet the criteria, guiding it to rely more strongly on the conditions during backpropagation; low noise maintains the prediction confidence level for stable segments. Extracting disclosure-guided noise scheduling refers to scheduling from the overall noise level. The noise factor sub-sequences that influence the gradient of the constrained dimension are extracted and used as disclosure-guided noise scheduling. The diffusion covariance structure can be constructed according to the graph Laplacian diffusion kernel method. The conditional diffusion distribution constructed based on the disclosure-guided noise scheduling and the structure diffusion kernel using the cross-attention mechanism refers to using the disclosure-guided noise scheduling as the query vector and the structure diffusion kernel as the key vector, and fusing them through the cross-attention mechanism to obtain the conditional diffusion distribution parameters used to generate, for example, the mean correction term and the covariance correction term of the conditional Gaussian noise distribution. Forward diffusion and backward diffusion can be performed according to the diffusion transition probability of the Markov chain.

[0049] In detail, the step of performing soft and hard constraint optimization on the completed carbon information map after time-series conditional diffusion to obtain an optimized carbon information map includes: The predicted carbon emission time series is extracted from the completed carbon information map after the time series condition diffusion. The predicted carbon emission time series is then compared for rule consistency and the difference is quantified according to the disclosure constraints to obtain the difference marker sequence. A sliding window aggregation is performed on the differential marker sequence to obtain a trend offset vector; The trend offset vector is inverted and its amplitude is limited to obtain a correction guide sequence. The predicted carbon emission time series is then conditionally resampled based on the correction guide sequence to obtain a corrected carbon emission time series. The corrected carbon emission time series is mapped to each entity node of the completed carbon information map to obtain the optimized carbon information map.

[0050] The rule consistency comparison refers to comparing whether the carbon emission data in each predicted carbon emission time series meets the corresponding disclosure constraints. The difference quantification refers to marking the carbon emission data that does not meet the constraints and calculating the corresponding difference value. The normalized difference value and the corresponding carbon emission data are used as difference markers. The sliding window aggregation refers to averaging the short-term fluctuations of the difference marker sequence through a sliding window to extract the systematic trend of the constraint violation and obtain the trend offset vector. The inversion refers to reversing the sign of the trend offset vector to generate a reverse guidance signal. The amplitude can be limited by the Clipping function, and conditional resampling can be performed by back diffusion or local least squares projection.

[0051] By explicitly transforming real-time disclosure standards into quantifiable disclosure constraints and embedding them into a conditional diffusion framework, the predicted future carbon emission time series not only maintains the physical consistency of historical evolution but also automatically aligns with disclosure requirements during the generation stage. By using noise scheduling to adjust the confidence level of non-compliant intervals during the forward diffusion stage, the model's sensitivity to key disclosure items is enhanced. Soft and hard constraint optimization achieves rule-based rationality correction through difference quantification, trend offset extraction, and conditional resampling, rather than simply modifying numerical values ​​to meet standards. This ensures the authenticity and interpretability of the disclosed information. By integrating predictions with standards, not only can carbon emission risks be warned in advance, but emission reduction paths can also be proactively planned, thereby improving the completeness, compliance, and usability of carbon information disclosure.

[0052] Carbon information is disclosed by performing carbon information disclosure on the optimized carbon information map in accordance with the real-time disclosure standards, thereby obtaining the carbon information disclosed by the enterprise.

[0053] In detail, the process of disclosing carbon information in accordance with the optimized carbon information map according to the real-time disclosure standard to obtain the enterprise's disclosed carbon information includes: The real-time disclosure standard is structured and parsed to obtain the parsed real-time standard, and the list of disclosure entities, disclosure format requirements and field logic rules are extracted from the parsed real-time standard respectively. Initialize the disclosure format template based on the disclosure entity list, the disclosure format requirements, and the field logic rules; Based on the disclosed entity list, field matching and extraction are performed on each entity node in the optimized carbon information map to obtain the disclosed field list; The disclosure field list is logically compliantly converted according to the field logic rules, and the format of the logically compliant disclosure field list is normalized according to the disclosure format requirements to obtain a standard field list. The standard field list is mapped to the disclosure format template to obtain the carbon information disclosed by the enterprise.

[0054] The disclosure entity list refers to the list of entities that need to be disclosed in the real-time parsing standard, such as "Company Name", "Plant Power Consumption Rate", "Line Loss Rate", and "Fossil Fuel Combustion - Total Emissions". The disclosure format requirements refer to the field format corresponding to each disclosure entity. For example, the unit of the field format corresponding to the disclosure entity "Total Emissions from Fossil Fuel Combustion per Kilowatt-hour of Electricity" in the power industry is "gCO2e / kWh". The field logic rules refer to the field value range and field logic corresponding to each disclosure entity. For example, the field logic rule corresponding to the disclosure entity "Unit Nature" is any one of "Listed Company", "Central Enterprise / State-owned Enterprise", "Private Enterprise", and "Public Institution", and the total emissions should be equal to the sum of all sub-items. The disclosure format template can be initialized through an XBRL or JSON Schema generator. The logical compliance conversion refers to performing logical value range conversion and logical relationship correction. For example, the arithmetic logic check automatically verifies the sum relationship and the proportional relationship to ensure that the numerical relationships of Scope1, Scope2, and Scope3 are not contradictory. The format normalization refers to the unit normalization and structural format normalization.

[0055] By performing structured parsing, the disclosure requirements, field logic, and format specifications are made machine-readable. Based on the parsed list of disclosure entities and logical rules, corresponding fields can be accurately extracted from the optimized carbon information map, achieving highly matched entity-level carbon emission data reporting. Through logical compliance conversion, it is ensured that the extracted fields fully comply with the disclosure rules in terms of value range, total quantity relationship, and proportion relationship, avoiding logical contradictions that may occur in manual reporting. This improves the response efficiency of enterprises to different regulatory and market requirements. Multiple compliance reports can be generated from a single data processing, greatly reducing compliance costs. This improves the flexibility and comprehensiveness of carbon information disclosure adaptation and enhances the efficiency of enterprise carbon information disclosure.

[0056] Example 2: This invention discloses a corporate carbon information disclosure standard adaptation system, which includes a standard extraction module, a spectrum mapping module, a field completion module, a time-series diffusion module, and an information disclosure module, wherein: The standard extraction module performs structured parsing and semantic extraction on the pre-acquired list of carbon information disclosure standards to obtain a clause standard semantic set, and constructs a clause knowledge graph corresponding to the clause standard semantic set based on a semantic clustering algorithm. The graph mapping module performs entity parsing and feature extraction on the pre-acquired corporate carbon monitoring data to obtain a carbon monitoring feature set, and maps the carbon monitoring feature set to the clause knowledge graph based on a bidirectional attention matching algorithm to obtain a corporate carbon information graph. The field completion module performs missing field prediction and context completion on the enterprise carbon information map based on the graph message passing mechanism to obtain a completed carbon information map. The time-series diffusion module obtains the real-time disclosure standard, performs graph-structure-guided time-series conditional diffusion on the completed carbon information map based on the structured constraints of the real-time disclosure standard, and optimizes the completed carbon information map after time-series conditional diffusion by soft and hard constraints to obtain an optimized carbon information map. The information disclosure module discloses carbon information from the optimized carbon information map according to the real-time disclosure standard, thereby obtaining the carbon information disclosed by the enterprise.

[0057] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0058] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0059] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A method for adapting to enterprise carbon disclosure standards, characterized by, The method comprises: performing structured analysis and semantic extraction on a pre-acquired carbon information disclosure standard list to obtain a clause standard semantic set, and constructing a clause knowledge graph corresponding to the clause standard semantic set based on a semantic clustering algorithm; performing entity analysis and feature extraction on pre-acquired enterprise carbon monitoring data to obtain a carbon monitoring feature set, and mapping the carbon monitoring feature set into the clause knowledge graph based on a bidirectional attention matching algorithm to obtain an enterprise carbon information graph; performing missing field prediction and context completion on the enterprise carbon information graph based on a graph message passing mechanism to obtain a completed carbon information graph; obtaining real-time disclosure standards, performing graph structure guided time sequence condition diffusion on the completed carbon information graph based on the structured constraint conditions of the real-time disclosure standards, and performing soft and hard constraint optimization on the completed carbon information graph after time sequence condition diffusion to obtain an optimized carbon information graph; performing carbon information disclosure on the optimized carbon information graph according to the real-time disclosure standards to obtain enterprise disclosed carbon information.

2. The enterprise carbon information disclosure standard adaptation method according to claim 1, characterized in that, The structured analysis and semantic extraction on the pre-acquired carbon information disclosure standard list to obtain a clause standard semantic set comprises: performing text format unification on the pre-acquired carbon information disclosure standard list to obtain a unified disclosure standard list; performing hierarchical structure analysis on each disclosure standard in the unified disclosure standard list to obtain a standard structured text set; performing text division and semantic unit division on the standard structured text set to obtain a clause semantic unit set; performing entity recognition on the clause semantic unit set to obtain a standard entity set, and performing relation extraction on the standard entity set to obtain an entity relation set; performing semantic alignment and vectorization coding on the standard entity set using the entity relation set to obtain a clause standard semantic set.

3. The enterprise carbon disclosure standard adaptation method of claim 1, wherein, The construction of the clause knowledge graph corresponding to the clause standard semantic set based on the semantic clustering algorithm comprises: performing semantic similarity calculation on each clause standard semantic in the clause standard semantic set to obtain a semantic similarity matrix; performing density clustering on each clause standard semantic based on the semantic similarity matrix to obtain a clause topic cluster set; performing cluster center calculation on each clause topic cluster in the clause topic cluster set to obtain an entity node set; constructing semantic edges for the entity node set according to the entity relations corresponding to each clause standard semantic in each clause topic cluster to obtain semantic association edges, and performing edge weighting on the semantic association edges according to the number of entity relations across entity nodes corresponding to each semantic association edge to obtain weighted association edges; generating a topic clause graph structure based on the weighted association edges and the entity node set, and mapping each clause standard semantic into the topic clause graph structure to obtain a clause knowledge graph.

4. The enterprise carbon disclosure standard adaptation method of claim 1, wherein, The entity analysis and feature extraction on the pre-acquired enterprise carbon monitoring data to obtain a carbon monitoring feature set comprises: performing data structure identification and structured analysis on the pre-acquired enterprise carbon monitoring data to obtain analyzed monitoring data; The parsed monitoring data is unified in units and cleaned, and the parsed monitoring data after data cleaning is unified in field format to obtain standard monitoring data; The standard monitoring data is subjected to attribute field identification and carbon emission entity identification to obtain a primary carbon emission entity set, and the primary carbon emission entity set is subjected to semantic alignment and semantic disambiguation to obtain a standard carbon emission entity set; The standard carbon emission entity set is subjected to relationship extraction based on the standard monitoring data to obtain a primary entity relationship set, and the primary entity relationship set is subjected to time sequence dependency analysis to obtain an entity link relationship set; The standard monitoring data and the entity link relationship set are used to construct a monitoring data sequence set, and time sequence feature extraction is performed on the monitoring data sequence set to obtain a monitoring time sequence feature set; The standard carbon emission entity set is embedded in the monitoring time sequence feature set to obtain a carbon monitoring feature set.

5. The enterprise carbon disclosure standard adaptation method of claim 1, wherein, The carbon monitoring feature set is mapped into the clause knowledge graph based on the bidirectional attention matching algorithm to obtain an enterprise carbon information graph, including: The carbon monitoring feature set and the entity node set of the clause knowledge graph are subjected to double-tower coding respectively to obtain a coded monitoring feature set and a coded node feature set; A first attention weight matrix of the coded monitoring feature set relative to the coded node feature set and a second attention weight matrix of the coded node feature set relative to the coded monitoring feature set are calculated; The coded monitoring feature set is subjected to context weight enhancement and feature pooling operation based on the first attention weight to obtain an enhanced monitoring feature set, and the coded node feature set is subjected to context weight enhancement and feature pooling operation based on the second attention weight to obtain an enhanced node feature set; A feature similarity matrix between the enhanced monitoring feature set and the enhanced node feature set is calculated, and a similarity mapping relationship set is generated based on the feature similarity matrix; The carbon monitoring feature set is mapped into the corresponding entity node of the clause knowledge graph based on the similarity mapping relationship set to obtain an enterprise carbon information graph.

6. The enterprise carbon information disclosure standard adaptation method of claim 1, wherein, The enterprise carbon information graph is subjected to missing field prediction and context completion based on the graph message passing mechanism to obtain a completed carbon information graph, including: The enterprise carbon information graph is scanned for missing fields in a traversal manner to obtain a missing field set; A target missing field in the missing field set is selected one by one, and the target missing field is subjected to semantic relationship screening and neighborhood sampling to obtain a neighborhood information graph; Each entity node in the neighborhood information graph is subjected to node feature coding to obtain a neighborhood node feature set, and the neighborhood node feature set is subjected to cross-type message passing and multi-semantic feature aggregation based on an attention mechanism to obtain an aggregated node feature; The target missing field is predicted based on the aggregated node feature to obtain a field prediction result, and the field prediction result is subjected to confidence weighting and conflict resolution to obtain a target filled field; The target missing field in the enterprise carbon information graph is filled in by using the target filling field, until the target missing field is the last missing field in the missing field set, and the filled enterprise carbon information graph is taken as a completed carbon information graph.

7. The enterprise carbon disclosure standards adaptation method of claim 1, wherein, The graph structure of the completed carbon information graph is guided by the structured constraint condition based on the real-time disclosure standard, and the time sequence condition diffusion is performed. The time sequence feature sequence of each entity node in the completed carbon information graph is extracted, and the entity time sequence feature sequence is obtained. The real-time disclosure standard is structurally analyzed to obtain an analyzed real-time standard, and a disclosure constraint condition is extracted from the analyzed real-time standard. The graph structure condition is obtained by graph structure coding of the completed carbon information graph. The graph structure condition and the disclosure constraint condition are taken as diffusion conditions to perform forward diffusion on the entity time sequence feature sequence to obtain a noisy entity feature sequence. Based on the diffusion condition, the noisy entity feature sequence is reversely diffused to obtain a predicted carbon emission time sequence, and the predicted carbon emission time sequence is mapped into each entity node of the completed carbon information graph.

8. The enterprise carbon information disclosure standard adaptation method of claim 7, wherein, The graph structure condition and the disclosure constraint condition are taken as diffusion conditions to perform forward diffusion on the entity time sequence feature sequence to obtain a time sequence noise sequence, including: The feature standardization operation is performed on the entity time sequence feature sequence, and the trend filtering is performed on the feature-standardized entity time sequence feature sequence set to obtain an initial time sequence state. A conditional noise level schedule is generated based on the disclosure constraint condition, and a disclosure guide noise schedule is extracted from the conditional noise level schedule. The diffusion covariance structure is constructed according to the graph structure condition to obtain a structure diffusion kernel, and the conditional diffusion distribution is constructed based on the cross-attention mechanism according to the disclosure guide noise schedule and the structure diffusion kernel. The initial time sequence state is forward diffused according to the preset diffusion step based on the conditional diffusion distribution to obtain a time sequence noise sequence, and the time sequence noise sequence corresponding to the time sequence noise sequence is calculated based on the reparameterization algorithm.

9. The enterprise carbon information disclosure standard adaptation method of claim 8, wherein, The completed carbon information graph after the time sequence condition diffusion is optimized to obtain an optimized carbon information graph, including: The predicted carbon emission time sequence is extracted from the completed carbon information graph after the time sequence condition diffusion, the predicted carbon emission time sequence is compared for rule consistency and quantified for difference according to the disclosure constraint condition to obtain a difference mark sequence. The difference mark sequence is aggregated by a sliding window to obtain a trend offset vector. The trend offset vector is inverted and amplitude-limited to obtain a correction guide sequence, and the predicted carbon emission time sequence is conditionally resampled according to the correction guide sequence to obtain a corrected carbon emission time sequence. The corrected carbon emission time sequence is mapped into each entity node of the completed carbon information graph to obtain an optimized carbon information graph.

10. A system for adapting standards for corporate carbon disclosure, characterized by, The system includes a standard extraction module, a graph mapping module, a field completion module, a time sequence diffusion module, and an information disclosure module, wherein: The standard extraction module is configured to perform structural analysis and semantic extraction on a pre-acquired carbon information disclosure standard list to obtain a clause standard semantic set, and construct a clause knowledge graph corresponding to the clause standard semantic set based on a semantic clustering algorithm. The graph mapping module is configured to perform entity analysis and feature extraction on pre-acquired enterprise carbon monitoring data to obtain a carbon monitoring feature set, and map the carbon monitoring feature set to the clause knowledge graph based on a bidirectional attention matching algorithm to obtain an enterprise carbon information graph. The field completion module is configured to perform missing field prediction and context completion on the enterprise carbon information graph based on a graph message passing mechanism to obtain a completed carbon information graph. The time series diffusion module is configured to acquire real-time disclosure standards, perform graph structure guided time series condition diffusion on the completed carbon information graph based on structural constraint conditions of the real-time disclosure standards, and perform soft and hard constraint optimization on the completed carbon information graph after the time series condition diffusion to obtain an optimized carbon information graph. The information disclosure module is configured to perform carbon information disclosure on the optimized carbon information graph according to the real-time disclosure standards to obtain enterprise disclosed carbon information.

Images