A method and system for real-time accounting of carbon footprint of leather products based on a knowledge graph

Abstract

The application discloses a kind of leather product carbon footprint real-time accounting method and system based on knowledge graph, specifically includes: step 1, leather product, process, resource, energy and carbon emission factor time index data are collected;Step 2, construct multi-type node knowledge graph to express entity association;Step 3, construct probability soft logic rule set and rule index structure;Step 4, construct life cycle stage set and accounting path set;Step 5, construct stage-path double-indexed carbon flow tensor and introduce rule-induced low-rank constraint;Step 6, construct target function including data consistency, rule constraint and low-rank constraint and calculate variable;Step 7, based on adjacent time index, perform prediction correction update;Step 8, output leather product carbon footprint value and path contribution result.The application realizes carbon footprint accounting with knowledge graph and low-rank constraint.

Description

Technical Field

[0001] This invention relates to the field of intelligent carbon footprint accounting technology, and in particular to a method and system for real-time carbon footprint accounting of leather products based on knowledge graphs. Background Technology

[0002] Product carbon footprint accounting has become a crucial foundation for the green transformation and refined management of the manufacturing industry. Leather products involve significant resource consumption and energy use across multiple stages, including raw material processing, tanning, finishing, and transportation. Their carbon emissions are dispersed and influenced by complex factors, resulting in carbon footprint accounting characterized by diverse data sources, intricate relationships, and significant temporal variations. Existing carbon footprint accounting methods largely rely on static inventory data or manually set rules, typically summarizing calculations on a lifecycle basis. This makes it difficult to accurately reflect the impact of different process paths, resource combinations, and changes in energy structure on carbon emission results.

[0003] On the other hand, existing models for carbon emission data often employ linear statistical models or single optimization models, lacking a unified ability to express structured information involving multiple entities and relationships. When faced with dynamically updated data, frequent recalculations are often required, limiting computational efficiency and real-time performance. Furthermore, rule constraints are typically introduced as hard constraints or empirical parameters, lacking a detailed characterization of the rule's scope and structural characteristics, which can easily lead to increased model complexity or insufficient result stability.

[0004] Furthermore, in actual production and operation, resource consumption, energy structure, and emission factors all change continuously over time. Existing methods do not adequately utilize the data correlation between adjacent time periods, making it difficult to achieve rapid updates over continuous time while ensuring computational accuracy. Therefore, how to achieve dynamic modeling, constraint expression, and real-time calculation of the carbon footprint of leather products under complex relational structures remains a problem that urgently needs to be solved in existing technologies. Summary of the Invention

[0005] One objective of this invention is to propose a real-time carbon footprint accounting method and system for leather products based on knowledge graphs. This invention introduces a multi-entity association modeling mechanism based on knowledge graphs and a rule-induced low-rank tensor constraint calculation model to perform structured expression, dynamic optimization, and predictive correction updates on resource, energy, and carbon emission data throughout the entire life cycle of leather products. This constructs a continuously evolving real-time carbon footprint accounting process, which features traceable accounting paths, strong data update continuity, and stable and consistent results.

[0006] A real-time carbon footprint calculation method for leather products based on knowledge graphs according to an embodiment of the present invention includes the following steps: Step 1: Collect time-indexed data corresponding to leather products, processes, resources, energy, and carbon emission factors, and form a dataset; Step 2: Construct a knowledge graph containing product nodes, process nodes, resource nodes, energy nodes, and factor nodes; establish the relationships between products and processes, processes and resources, processes and energy, and resources or energy and carbon emission factors, and write them into the time index attribute. Step 3: Construct a set of probabilistic soft logic rules and establish a rule index structure. The rule index structure records the corresponding mapping between rules and associated nodes and relationships. Step 4: Construct the lifecycle stage set and the accounting path set; Step 5: Construct carbon flow tensor variables with life cycle stage index and accounting path index as index dimensions, determine the tensor slice range of rule action according to the rule index structure, and introduce rule-induced low-rank constraints on the corresponding tensor slices; Step 6: Construct an objective function containing data consistency terms, rule constraint terms, and low-rank constraint terms, and use the primal-dual hybrid gradient algorithm to obtain the primal and dual variables of the current time index; Step 7: Calculate the data increment for the datasets of adjacent time indices, generate predictive variables based on the original and dual variables of the previous time index in the low-rank subspace, and obtain the updated variables through a finite-step original-dual correction operation. Step 8: Output the carbon footprint value and path contribution result of the leather product based on the updated variables.

[0007] Optionally, step 2 specifically includes: Read the product identifier set, process identifier set, resource identifier set, energy identifier set and carbon emission factor identifier set, generate product node, process node, resource node, energy node and carbon emission factor node according to each identifier set and record the corresponding node identifier; Read the set of data items containing time indexes, detect the co-occurrence relationship between product identifier and process identifier, process identifier and resource identifier, process identifier and energy identifier, resource identifier or energy identifier and carbon emission factor identifier in the data items with the same time index, generate the association relationship between product node and process node, process node and resource node, process node and energy node, resource node or energy node and carbon emission factor node according to the co-occurrence relationship, and record the association relationship identifier. Write the time index and the corresponding resource consumption value, energy consumption value or carbon emission factor value into the time index attribute field of product node, process node, resource node, energy node, carbon emission factor node or corresponding relationship to form a knowledge graph structure with time index attribute.

[0008] Optionally, step 3 includes the following steps: Read the node identifiers and relationship identifiers of product nodes, process nodes, resource nodes, energy nodes, carbon emission factor nodes, and their corresponding relationships; generate probabilistic soft logic rule entries according to the node type identifier and relationship type identifier, and record the rule number. Generate mapping records between rules and node identifiers, and between rules and association identifiers, based on the node identifiers and association identifiers referenced in each probability soft logic rule entry; The rule index structure is formed by combining rule numbers with their corresponding mapping records. The rule index structure records the correspondence between rule numbers and the set of node identifiers and the set of association identifiers.

[0009] Optionally, step 4 specifically includes: Read the product node, process node and corresponding time index attribute, sort the process nodes according to the time index order, generate life cycle stage identifiers based on the sorting results and collect them to form a life cycle stage set. Read the product nodes, process nodes, and the association relationship identifiers between product nodes and process nodes. Generate an ordered node sequence consisting of product nodes and process nodes according to the connection order of the association relationship identifiers, and record the ordered node sequence identifiers. The system aggregates ordered node sequence identifiers to form a set of accounting paths, establishes an index association between lifecycle stage identifiers and corresponding ordered node sequence identifiers, and records the associations.

[0010] Optionally, step 5 includes the following steps: Read the life cycle stage identifier set in the life cycle stage set and the accounting path identifier set in the accounting path set, generate a two-dimensional index space according to the life cycle stage identifier and the accounting path identifier, and allocate carbon flow numerical storage location for each index pair in the two-dimensional index space to form a carbon flow tensor variable. Read the rule number, node identifier set, and association identifier set recorded in the rule index structure. Locate the life cycle stage identifier subset according to the node identifier set corresponding to the rule number, locate the accounting path identifier subset according to the association identifier set corresponding to the rule number, and combine the life cycle stage identifier subset and the accounting path identifier subset to generate the tensor slice index range corresponding to the rule. Extract rule-acting tensor slices from the carbon flow tensor variable according to the tensor slice index range, and record the mapping relationship between the rule number and the rule-acting tensor slice; For each rule-action tensor slice, a low-rank constraint parameter set is constructed, which includes the rank upper limit parameter and the slice decomposition dimension parameter. The rule-action tensor slices are constructed by performing a low-rank representation operation on the low-rank constraint parameter set, and the rule-action tensor slices are represented as tensor decomposition forms under constrained rank conditions. Write the low-rank representation into the constraint description field of the carbon flow tensor variable.

[0011] Optionally, step 5, constructing the low-rank constraint parameter set, specifically includes: Read the rule number, node identifier set, and association relationship identifier set recorded in the rule index structure, and locate the corresponding rule action tensor slice identifier set according to the rule number; For each rule-effect tensor slice identifier, read the lifecycle stage identifier subset and the calculation path identifier subset corresponding to the rule-effect tensor slice, and count the number of identifiers in the lifecycle stage identifier subset and the number of identifiers in the calculation path identifier subset. The slice dimension scale parameter is generated according to the number of lifecycle stage identifiers and the number of accounting path identifiers. The slice dimension scale parameter records the size information of the rule-effect tensor slice in the lifecycle stage index dimension and the accounting path index dimension. Based on the slice dimension scale parameter, the slice decomposition dimension parameter is generated. The slice decomposition dimension parameter records the set of dimension numbers and the set of dimension order numbers of the tensor slices that participate in the tensor decomposition calculation according to the rule. The rank upper limit parameter is determined according to the dimensional scale value corresponding to the slice decomposition dimensional parameter. The rank upper limit parameter records the range of rank values ​​that the rule applies to the tensor slice during the tensor decomposition calculation. The slice decomposition dimension parameters and the rank upper limit parameters are combined to form a set of low-rank constraint parameters that correspond one-to-one with the rule action tensor slice identifier, and the correspondence between the rule number, the rule action tensor slice identifier and the low-rank constraint parameter set is recorded.

[0012] Optionally, step 6 specifically includes: Read the carbon flow tensor variable, the time index data set, and the low-rank constraint parameter set, and select the tensor elements in the carbon flow tensor variable that are consistent with the time index to form the original variable set; The data consistency items are calculated based on the resource consumption values, energy consumption values, and carbon emission factor values ​​corresponding to the original variable set and time index. The rule constraint items are calculated based on the rule number, node identifier set, and association relationship identifier set recorded in the rule index structure. The low-rank constraint items are calculated based on the low-rank constraint parameter set. The objective function is formed by combining the data consistency term, the rule constraint term, and the low-rank constraint term. The original variable set and the dual variable set are initialized for the objective function. The original variable set and the dual variable set are iteratively calculated according to the original-dual mixed gradient update order. The original variable and the dual variable corresponding to the time index are output.

[0013] Optionally, step 7 includes the following steps: Read the data set corresponding to the current time index and the previous time index, perform item-by-item numerical difference operation on the resource consumption value, energy consumption value and carbon emission factor value according to the time index order, generate a data increment vector containing the incremental components of resource consumption, energy consumption and carbon emission factors, and record the correspondence between the data increment vector and the time index. Read the original variable set and dual variable set corresponding to the previous time index, read the rule-induced low-rank expression form, rank upper limit parameter and tensor decomposition dimension parameter, within the tensor decomposition dimension space limited by the low-rank expression form, map the original variable set and dual variable set to tensor decomposition factor numerical set, perform incremental update operation on the tensor decomposition factor numerical set according to the data increment vector, generate the predicted tensor decomposition factor numerical set, and reconstruct the predicted original variable set and predicted dual variable set from the predicted tensor decomposition factor numerical set; Using the original set of predicted variables and the dual set of predicted variables as the initial state, the data consistency term, rule constraint term, and low-rank constraint term corresponding to the current time index are read. The original-dual update operation is performed on the objective function a preset number of times. In each original-dual update operation, the original set of variables and the dual set of variables are updated synchronously. After each update, the rule residual value and the low-rank consistency residual value are calculated. The rule residual value and the low-rank consistency residual value are compared with the corresponding threshold. When the comparison result meets the convergence judgment condition, the updated original set of variables and the updated dual set of variables are output as the updated variables corresponding to the current time index.

[0014] Optionally, step 8 specifically includes: Read the set of updated original variables corresponding to the current time index, and extract the tensor element values ​​corresponding to the product node identifier from the carbon flow tensor variable; Aggregation operations are performed on the tensor element values ​​according to the life cycle stage identifier and the accounting path identifier to generate the carbon footprint value corresponding to the product node identifier; Perform grouped statistical operations on the tensor element values ​​according to the accounting path identifier, generate the path contribution result corresponding to each accounting path identifier, and output it.

[0015] A real-time carbon footprint accounting system for leather products based on knowledge graphs according to an embodiment of the present invention includes the following modules: The data acquisition module is used to collect time-indexed data on leather products, processes, resources, energy, and carbon emission factors. The knowledge graph construction module is used to build a knowledge graph containing product nodes, process nodes, resource nodes, energy nodes and factor nodes, and to establish temporal relationships between nodes; The rule engine module is used to build a set of probabilistic soft logic rules and establish a rule index structure, defining the mapping relationship between rules and nodes and their associations. The Lifecycle and Path module is used to construct the lifecycle stage set and the accounting path set, and generate the lifecycle stage and accounting path index of the carbon flow tensor variable; The low-rank constraint module is used to introduce rule-induced low-rank constraints on carbon flux tensor slices. The optimization computation module is used to construct an objective function that includes data consistency terms, rule constraint terms, and low-rank constraint terms, and to execute the primal-dual hybrid gradient algorithm. The prediction and correction module is used to predict and update the original and dual variables of adjacent time indices based on the data increment, and output the carbon footprint value and path contribution results.

[0016] The beneficial effects of this invention are: (1) This invention constructs a knowledge graph structure that includes products, processes, resources, energy and carbon emission factors, and models carbon emission-related data scattered in different life cycle stages and process paths in a unified manner, so that carbon emission data has a consistent structural expression form under multiple entities and multiple relationships, thereby realizing the synchronous accounting and correlation analysis of the carbon footprint of leather products in the stage dimension and path dimension.

[0017] (2) This invention introduces a rule-induced low-rank constraint expression in the carbon flow tensor variable, and limits the probabilistic soft logic rules to a specific tensor slice range for structural constraints, so that rule constraints and data constraints work together in a unified optimization framework, reducing redundant computational dimensions and enhancing the numerical stability and structural consistency of the accounting process under complex rule conditions.

[0018] (3) The present invention adopts a time-indexed data incremental prediction-correction update mechanism to continuously update the carbon emission data of adjacent time indices in the low-rank subspace, avoiding repeated solutions to the full data, so that the carbon footprint results can continuously evolve with changes in resource consumption, energy structure and emission factors, and meet the application requirements of real-time carbon footprint accounting and path contribution analysis of leather products. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is a flowchart of a real-time carbon footprint calculation method and system for leather products based on knowledge graphs, as proposed in this invention. Figure 2 This is a schematic diagram of rule-index-driven tensor slicing and rule-induced low-rank constraints for a real-time carbon footprint accounting method and system for leather products based on knowledge graphs proposed in this invention. Figure 3 This is a schematic diagram of the prediction-correction update process in the low-rank subspace of a knowledge graph-based real-time carbon footprint accounting method and system for leather products proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-3 A method and system for real-time carbon footprint calculation of leather products based on knowledge graphs, comprising the following steps: Step 1: Collect time-indexed data corresponding to leather products, processes, resources, energy, and carbon emission factors, and form a dataset; Step 2: Construct a knowledge graph containing product nodes, process nodes, resource nodes, energy nodes, and factor nodes; establish the relationships between products and processes, processes and resources, processes and energy, and resources or energy and carbon emission factors, and write them into the time index attribute. Step 3: Construct a set of probabilistic soft logic rules and establish a rule index structure. The rule index structure records the corresponding mapping between rules and associated nodes and relationships. Step 4: Construct the lifecycle stage set and the accounting path set; Step 5: Construct carbon flow tensor variables with life cycle stage index and accounting path index as index dimensions, determine the tensor slice range of rule action according to the rule index structure, and introduce rule-induced low-rank constraints on the corresponding tensor slices; Step 6: Construct an objective function containing data consistency terms, rule constraint terms, and low-rank constraint terms, and use the primal-dual hybrid gradient algorithm to obtain the primal and dual variables of the current time index; Step 7: Calculate the data increment for the datasets of adjacent time indices, generate predictive variables based on the original and dual variables of the previous time index in the low-rank subspace, and obtain the updated variables through a finite-step original-dual correction operation. Step 8: Output the carbon footprint value and path contribution result of the leather product based on the updated variables.

[0022] In this embodiment, step 2 specifically includes: Read the product identifier set, process identifier set, resource identifier set, energy identifier set and carbon emission factor identifier set, generate product node, process node, resource node, energy node and carbon emission factor node according to each identifier set and record the corresponding node identifier; Read the set of data items containing time indexes, detect the co-occurrence relationship between product identifier and process identifier, process identifier and resource identifier, process identifier and energy identifier, resource identifier or energy identifier and carbon emission factor identifier in the data items with the same time index, generate the association relationship between product node and process node, process node and resource node, process node and energy node, resource node or energy node and carbon emission factor node according to the co-occurrence relationship, and record the association relationship identifier. Write the time index and the corresponding resource consumption value, energy consumption value or carbon emission factor value into the time index attribute field of product node, process node, resource node, energy node, carbon emission factor node or corresponding relationship to form a knowledge graph structure with time index attribute.

[0023] In this embodiment, the co-occurrence relationship in step 2 specifically includes: This refers to the association status between different identifiers that appear simultaneously in the data item set corresponding to the same time index. Specifically, under the same time index, if a product identifier and a process identifier, a process identifier and a resource identifier, a process identifier and an energy identifier, or a resource identifier or energy identifier and a carbon emission factor identifier all exist simultaneously in the data item set corresponding to that time index, then a co-occurrence relationship is determined between the corresponding identifiers. The co-occurrence relationship is determined using the time index as the boundary, obtained by detecting the simultaneous occurrence of identifier fields in the data item set, and written into the knowledge graph structure as the basis for generating the association relationship between nodes.

[0024] In this embodiment, step 3 includes the following steps: Read the node identifiers and relationship identifiers of product nodes, process nodes, resource nodes, energy nodes, carbon emission factor nodes, and their corresponding relationships; generate probabilistic soft logic rule entries according to the node type identifier and relationship type identifier, and record the rule number. Generate mapping records between rules and node identifiers, and between rules and association identifiers, based on the node identifiers and association identifiers referenced in each probability soft logic rule entry; The rule index structure is formed by combining rule numbers with their corresponding mapping records. The rule index structure records the correspondence between rule numbers and the set of node identifiers and the set of association identifiers.

[0025] In this embodiment, the specific items for generating probability soft logic rules in step 3 include: Read the set of node type identifiers and the set of association relationship type identifiers involved in each probability soft logic rule entry, sort the set of node type identifiers according to the preset node type order, and sort the set of association relationship type identifiers according to the preset association relationship type order; The sorted set of node type identifiers and the set of association relationship type identifiers are combined in a fixed splicing order to form a rule structure identifier; Perform a number mapping operation on the rule structure identifier to generate a rule number that corresponds one-to-one with the rule structure identifier, and write the rule number into the rule index structure to establish the correspondence between the rule number and the node type identifier and the association relationship type identifier.

[0026] In this embodiment, step 4 specifically includes: Read the product node, process node and corresponding time index attribute, sort the process nodes according to the time index order, generate life cycle stage identifiers based on the sorting results and collect them to form a life cycle stage set. Read the product nodes, process nodes, and the association relationship identifiers between product nodes and process nodes. Generate an ordered node sequence consisting of product nodes and process nodes according to the connection order of the association relationship identifiers, and record the ordered node sequence identifiers. The system aggregates ordered node sequence identifiers to form a set of accounting paths, establishes an index association between lifecycle stage identifiers and corresponding ordered node sequence identifiers, and records the associations.

[0027] In this embodiment, step 5 includes the following steps: Read the life cycle stage identifier set in the life cycle stage set and the accounting path identifier set in the accounting path set, generate a two-dimensional index space according to the life cycle stage identifier and the accounting path identifier, and allocate carbon flow numerical storage location for each index pair in the two-dimensional index space to form a carbon flow tensor variable. Read the rule number, node identifier set, and association identifier set recorded in the rule index structure. Locate the life cycle stage identifier subset according to the node identifier set corresponding to the rule number, locate the accounting path identifier subset according to the association identifier set corresponding to the rule number, and combine the life cycle stage identifier subset and the accounting path identifier subset to generate the tensor slice index range corresponding to the rule. Extract rule-acting tensor slices from the carbon flow tensor variable according to the tensor slice index range, and record the mapping relationship between the rule number and the rule-acting tensor slice; For each rule-action tensor slice, a low-rank constraint parameter set is constructed, which includes the rank upper limit parameter and the slice decomposition dimension parameter. The rule-action tensor slices are constructed by performing a low-rank representation operation on the low-rank constraint parameter set, and the rule-action tensor slices are represented as tensor decomposition forms under constrained rank conditions. Write the low-rank representation into the constraint description field of the carbon flow tensor variable.

[0028] In this embodiment, step 5, constructing the low-rank constraint parameter set, specifically includes: Read the rule number, node identifier set, and association relationship identifier set recorded in the rule index structure, and locate the corresponding rule action tensor slice identifier set according to the rule number; For each rule-effect tensor slice identifier, read the lifecycle stage identifier subset and the calculation path identifier subset corresponding to the rule-effect tensor slice, and count the number of identifiers in the lifecycle stage identifier subset and the number of identifiers in the calculation path identifier subset. The slice dimension scale parameter is generated according to the number of lifecycle stage identifiers and the number of accounting path identifiers. The slice dimension scale parameter records the size information of the rule-effect tensor slice in the lifecycle stage index dimension and the accounting path index dimension. Based on the slice dimension scale parameter, the slice decomposition dimension parameter is generated. The slice decomposition dimension parameter records the set of dimension numbers and the set of dimension order numbers of the tensor slices that participate in the tensor decomposition calculation according to the rule. The rank upper limit parameter is determined according to the dimensional scale value corresponding to the slice decomposition dimensional parameter. The rank upper limit parameter records the range of rank values ​​that the rule applies to the tensor slice during the tensor decomposition calculation. The slice decomposition dimension parameters and the rank upper limit parameters are combined to form a set of low-rank constraint parameters that correspond one-to-one with the rule action tensor slice identifier, and the correspondence between the rule number, the rule action tensor slice identifier and the low-rank constraint parameter set is recorded.

[0029] In this embodiment, step 5, locating the path identifier subset according to the association relationship identifier set corresponding to the rule number, specifically includes: Read the set of node identifiers and the set of association identifiers corresponding to the rule number from the rule index structure; In the lifecycle stage set, retrieve lifecycle stage identifiers that have a mapping relationship with product node identifiers or process node identifiers in the node identifier set, and gather the retrieved lifecycle stage identifiers to form a lifecycle stage identifier subset. In the set of accounting paths, retrieve the accounting path identifiers that contain any association identifier in the set of association identifiers corresponding to the rule number, and gather the retrieved accounting path identifiers to form a subset of accounting path identifiers; The subset of lifecycle stage identifiers is combined with the subset of accounting path identifiers to determine the tensor slice index range for rule application.

[0030] In this embodiment, step 5, constructing the low-rank constraint parameter set, specifically includes: Read the rule number recorded in the rule index structure and locate the corresponding rule action tensor slice based on the rule number; Read the index range of the tensor slice affected by the rule in the lifecycle stage index dimension and the accounting path index dimension, and count the number of stage identifiers of the tensor slice in the lifecycle stage index dimension and the number of path identifiers in the accounting path index dimension. The slice decomposition dimension parameters are generated based on the number of stage identifiers and the number of path identifiers. The slice decomposition dimension parameters are used to record the index dimensions and corresponding dimension order of the tensor slices that participate in the decomposition during the tensor decomposition calculation process under the effect of this rule. Based on the dimension scale value corresponding to the slice decomposition dimension parameter, determine the upper rank parameter corresponding to the tensor slice affected by the rule. The upper rank parameter is used to limit the rank value range of the tensor slice in the tensor decomposition calculation process. Finally, combine the slice decomposition dimension parameter and the upper rank parameter to form a set of low-rank constraint parameters that correspond one-to-one with the tensor slice affected by the rule, and establish an association record between the set of low-rank constraint parameters and the corresponding rule number.

[0031] In this embodiment, step 5, which involves constructing a low-rank representation of the rule-acting tensor slices according to the low-rank constraint parameter set, specifically includes: Read the rule number recorded in the rule index structure and locate the corresponding rule action tensor slice based on the rule number; Read the index range of the tensor slice affected by the rule in the lifecycle stage index dimension and the accounting path index dimension, and count the number of stage identifiers of the tensor slice in the lifecycle stage index dimension and the number of path identifiers in the accounting path index dimension. The slice decomposition dimension parameters are generated based on the number of stage identifiers and the number of path identifiers. The slice decomposition dimension parameters are used to record the index dimensions and corresponding dimension order of the tensor slices that participate in the decomposition during the tensor decomposition calculation process under the effect of this rule. Based on the dimension scale value corresponding to the slice decomposition dimension parameter, the upper rank parameter corresponding to the tensor slice affected by the rule is determined. The upper rank parameter is used to limit the range of rank values ​​of the tensor slice in the tensor decomposition calculation process. The slice decomposition dimension parameter and the rank upper limit parameter are combined to form a set of low-rank constraint parameters that correspond one-to-one with the rule action tensor slice, and the set of low-rank constraint parameters is associated with the corresponding rule number.

[0032] In this embodiment, step 6 specifically includes: Read the carbon flow tensor variable, the time index data set, and the low-rank constraint parameter set, and select the tensor elements in the carbon flow tensor variable that are consistent with the time index to form the original variable set; The data consistency items are calculated based on the resource consumption values, energy consumption values, and carbon emission factor values ​​corresponding to the original variable set and time index. The rule constraint items are calculated based on the rule number, node identifier set, and association relationship identifier set recorded in the rule index structure. The low-rank constraint items are calculated based on the low-rank constraint parameter set. The objective function is formed by combining the data consistency term, the rule constraint term, and the low-rank constraint term. The original variable set and the dual variable set are initialized for the objective function. The original variable set and the dual variable set are iteratively calculated according to the original-dual mixed gradient update order. The original variable and the dual variable corresponding to the time index are output.

[0033] In this embodiment, the data consistency item calculated in step 6 specifically includes: Read the original variable set corresponding to the current time index, and extract the variable components corresponding to resource nodes, energy nodes and carbon emission factor nodes from the original variable set; Read the resource consumption value, energy consumption value, and carbon emission factor value corresponding to the current time index, and align the original variable components with the corresponding resource consumption value, energy consumption value, and carbon emission factor value according to the node identifier consistency principle. For each pair of aligned original variable components and corresponding data values, perform numerical difference operation to obtain the difference result, and perform squaring operation on the difference result to generate the corresponding difference metric value; Summing is performed on all differential metric values ​​to form a data consistency term corresponding to the current time index, and this data consistency term is used as a component of the objective function in subsequent calculations.

[0034] In this embodiment, step 6, performing iterative calculations, specifically includes: In each iteration, the current value of the dual variable set is fixed first, and a numerical update operation is performed on the original variable set to adjust the original variable set in the direction of satisfying the data consistency term, rule constraint term and low-rank constraint term; After fixing the updated original variable set, perform a numerical update operation on the dual variable set to synchronize the degree of satisfaction of the dual variable set with the corresponding constraint terms. The numerical update of the dual variable set is to transform the deviation of the original variable set from each constraint term into the corresponding numerical adjustment of the dual variable components, which is used to record the constraint offset state under the current iteration. After completing the above two updates, a joint update result of the original variable set and the dual variable set is formed. Iterative calculation involves repeating the above joint update process several times, and outputting the corresponding original and dual variables when the preset number of iterations is reached or the convergence criterion is met.

[0035] In this embodiment, step 7 includes the following steps: Read the data set corresponding to the current time index and the previous time index, perform item-by-item numerical difference operation on the resource consumption value, energy consumption value and carbon emission factor value according to the time index order, generate a data increment vector containing the incremental components of resource consumption, energy consumption and carbon emission factors, and record the correspondence between the data increment vector and the time index. Read the original variable set and dual variable set corresponding to the previous time index, read the rule-induced low-rank expression form, rank upper limit parameter and tensor decomposition dimension parameter, within the tensor decomposition dimension space limited by the low-rank expression form, map the original variable set and dual variable set to tensor decomposition factor numerical set, perform incremental update operation on the tensor decomposition factor numerical set according to the data increment vector, generate the predicted tensor decomposition factor numerical set, and reconstruct the predicted original variable set and predicted dual variable set from the predicted tensor decomposition factor numerical set; Using the original set of predicted variables and the dual set of predicted variables as the initial state, the data consistency term, rule constraint term, and low-rank constraint term corresponding to the current time index are read. The original-dual update operation is performed on the objective function a preset number of times. In each original-dual update operation, the original set of variables and the dual set of variables are updated synchronously. After each update, the rule residual value and the low-rank consistency residual value are calculated. The rule residual value and the low-rank consistency residual value are compared with the corresponding threshold values ​​respectively. When the comparison result meets the convergence judgment condition, the updated original set of variables and the updated dual set of variables are output as the updated variables corresponding to the current time index.

[0036] In this embodiment, step 7, performing the incremental update operation, specifically includes: Read the data increment vector corresponding to the current time index, and map each increment component in the data increment vector to the corresponding factor component position in the tensor decomposition factor value set according to the correspondence between the life cycle stage identifier and the accounting path identifier. Perform numerical superposition on the mapped factor components, and combine the corresponding incremental components with the original factor component values ​​to generate updated factor component values. After completing the incremental superposition of all factor components, an updated set of tensor decomposition factor values ​​is formed, and this updated set of tensor decomposition factor values ​​is used for subsequent reconstruction operations of the original set of prediction variables and the set of prediction dual variables.

[0037] In this embodiment, step 7, performing the primitive-dual update operation a preset number of times, specifically includes: In each update round, the current value of the dual variable set obtained in the previous round is read. While keeping the value of the dual variable set unchanged, a value update operation is performed on the original variable set to adjust the value of the original variable set in the direction that simultaneously satisfies the data consistency term, the rule constraint term, and the low-rank constraint term. Read the updated original variable set, and while keeping the values ​​of the original variable set unchanged, perform a value update operation on the dual variable set to keep the values ​​of the dual variable set synchronized with the constraint residual state corresponding to the current original variable set; After completing the primary and dual variable updates, a complete primary-dual update operation round is formed. The primary-dual update operation generates the primary and dual variable sets corresponding to the current time index by repeatedly executing the above update rounds a preset number of times.

[0038] In this embodiment, step 8 specifically includes: Read the set of updated original variables corresponding to the current time index, and extract the tensor element values ​​corresponding to the product node identifier from the carbon flow tensor variable; Aggregation operations are performed on tensor element values ​​according to life cycle stage identifiers and accounting path identifiers to generate carbon footprint values ​​corresponding to product node identifiers. The aggregation operation is to perform grouping and summing of tensor element values ​​at different life cycle stages and different accounting paths corresponding to the same product node identifier to form the corresponding carbon footprint values.

[0039] Grouped statistical operations are performed on the tensor element values ​​according to the accounting path identifier, generating and outputting the path contribution result corresponding to each accounting path identifier. The grouped statistical operation is to perform summation statistics on the tensor element values ​​corresponding to the same accounting path identifier to obtain the cumulative carbon flow value corresponding to that accounting path, which is used as the path contribution result.

[0040] A real-time carbon footprint accounting system for leather products based on knowledge graphs according to an embodiment of the present invention includes the following modules: The data acquisition module is used to collect time-indexed data on leather products, processes, resources, energy, and carbon emission factors. The knowledge graph construction module is used to build a knowledge graph containing product nodes, process nodes, resource nodes, energy nodes and factor nodes, and to establish temporal relationships between nodes; The rule engine module is used to build a set of probabilistic soft logic rules and establish a rule index structure, defining the mapping relationship between rules and nodes and their associations. The Lifecycle and Path module is used to construct the lifecycle stage set and the accounting path set, and generate the lifecycle stage and accounting path index of the carbon flow tensor variable; The low-rank constraint module is used to introduce rule-induced low-rank constraints on carbon flux tensor slices. The optimization computation module is used to construct an objective function that includes data consistency terms, rule constraint terms, and low-rank constraint terms, and to execute the primal-dual hybrid gradient algorithm. The prediction and correction module is used to predict and update the original and dual variables of adjacent time indices based on the data increment, and output the carbon footprint value and path contribution results. Example

[0041] To verify the feasibility of this invention in practice, it was applied to a carbon footprint accounting scenario for the entire production process of leather products. In this scenario, leather products continuously generate carbon emission data in multiple stages, including raw material processing, manufacturing, resource input, and energy consumption. There are significant differences between different process paths, and the relevant data changes continuously over time. Traditional accounting methods struggle to simultaneously consider both structural correlation and continuous updating capabilities.

[0042] In practical applications, the process first collects records of leather product-related processes, resource consumption, energy usage, and corresponding carbon emission factors. Each data point is then assigned a unique time index to ensure consistency across all data types over time. Subsequently, products, processes, resources, energy, and carbon emission factors are abstracted into different types of nodes. The co-occurrence of these node identifiers under the same time index is then detected. Based on this, a knowledge graph structure reflecting actual production relations is constructed, allowing complex production activities to be expressed in a structured form.

[0043] Building upon knowledge graphs, probabilistic soft logic rules are introduced to constrain the relationships between processes and resources, processes and energy, and resources and carbon emission factors. The scope of these rules is limited through a rule index structure. Simultaneously, a lifecycle stage set is formed based on the chronological order of processes, and multiple accounting paths are formed based on the connection between products and processes. This constructs a carbon flow tensor variable indexed by lifecycle stages and accounting paths. For tensor slices representing the rule effects, low-rank constraint expressions are introduced to compress the carbon flow data while maintaining structural consistency.

[0044] During continuous operation, when new time index data arrives, the system calculates the incremental data of resource, energy, and carbon emission factors between adjacent time indices, incrementally updates the tensor decomposition factors in the low-rank subspace, and obtains the updated variables corresponding to the current time index through primal-dual correction operations. Based on the updated carbon flow tensor, the carbon flow values ​​at different life cycle stages and different accounting paths are aggregated to finally obtain the overall carbon footprint value of leather products and the carbon emission contribution results corresponding to each accounting path.

[0045] In terms of implementation results, this method can continuously reflect the impact of changes in production activities on carbon footprint results without repeatedly calculating historical data. During multiple consecutive calculations, the differences in carbon emissions between different calculation paths remain stable and distinguishable, and the overall carbon footprint value shows a smooth evolution trend with changes in resource and energy inputs. This verifies the feasibility and consistency of this invention in real-time carbon footprint calculation under complex production scenarios.

[0046] Table 1: Comparison of Carbon Footprint Calculation Results and Path Contribution of Leather Products: , Table 1 shows the changes in the total resource consumption, total energy consumption, and carbon emission contribution of different life cycle stages and accounting paths under continuous time indexing. As can be seen from the time index numbers, the data for each stage are not isolated calculations, but are continuously updated based on the previous time index, reflecting the continuous nature of the carbon footprint accounting process.

[0047] In the time index corresponding to stage A, the total resource consumption values ​​for path 1 and path 2 are 128.5 and 142.3, respectively, showing a stable difference. This difference stems from the different process combinations and resource input structures within the accounting paths. The corresponding total energy consumption values ​​are 256.4 and 241.8, respectively, with their numerical distribution maintaining a consistent relationship with changes in resource consumption. The calculated carbon emission contributions from these paths are 312.7 and 298.4, reflecting the differences in carbon emission distribution across different paths at the same life cycle stage.

[0048] As the time index progresses to stages B and C, the resource and energy consumption of paths 1 and 2 both show a gradual changing trend. For example, in stage C, the total resource consumption of path 1 rises to 132.9, and the total energy consumption rises to 262.8, with the corresponding carbon emission contribution value changing synchronously to 321.4. The magnitude of change in the carbon emission contribution value of each path is consistent with the changes in resource and energy inputs, and no abrupt changes or abnormal jumps occur, indicating that the carbon flow tensor variable and the low-rank constraint expression maintain stable numerical evolution characteristics during continuous updates.

[0049] By comparing the carbon emission contributions of different accounting paths at the same life cycle stage, we can intuitively identify the differences in the contribution of each path to the overall carbon footprint, providing a clear data foundation for subsequent path analysis and structural optimization.

[0050] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for real-time carbon footprint calculation of leather products based on knowledge graphs, characterized in that, Includes the following steps: Step 1: Collect time-indexed data corresponding to leather products, processes, resources, energy, and carbon emission factors, and form a dataset; Step 2: Construct a knowledge graph containing product nodes, process nodes, resource nodes, energy nodes, and factor nodes; establish the relationships between products and processes, processes and resources, processes and energy, and resources or energy and carbon emission factors, and write them into the time index attribute. Step 3: Construct a set of probabilistic soft logic rules and establish a rule index structure. The rule index structure records the corresponding mapping between rules and associated nodes and relationships. Step 4: Construct the lifecycle stage set and the accounting path set; Step 5: Construct carbon flow tensor variables with life cycle stage index and accounting path index as index dimensions, determine the tensor slice range of rule action according to the rule index structure, and introduce rule-induced low-rank constraints on the corresponding tensor slices; Step 6: Construct an objective function containing data consistency terms, rule constraint terms, and low-rank constraint terms, and use the primal-dual hybrid gradient algorithm to obtain the primal and dual variables of the current time index; Step 7: Calculate the data increment for the datasets of adjacent time indices, generate predictive variables based on the original and dual variables of the previous time index in the low-rank subspace, and obtain the updated variables through a finite-step original-dual correction operation. Step 8: Output the carbon footprint value and path contribution result of the leather product based on the updated variables.

2. The method for real-time carbon footprint calculation of leather products based on knowledge graphs according to claim 1, characterized in that, Step 2 specifically includes: Read the product identifier set, process identifier set, resource identifier set, energy identifier set and carbon emission factor identifier set, generate product node, process node, resource node, energy node and carbon emission factor node according to each identifier set and record the corresponding node identifier; Read the set of data items containing time indexes, detect the co-occurrence relationship between product identifier and process identifier, process identifier and resource identifier, process identifier and energy identifier, resource identifier or energy identifier and carbon emission factor identifier in the data items with the same time index, generate the association relationship between product node and process node, process node and resource node, process node and energy node, resource node or energy node and carbon emission factor node according to the co-occurrence relationship, and record the association relationship identifier. Write the time index and the corresponding resource consumption value, energy consumption value or carbon emission factor value into the time index attribute field of product node, process node, resource node, energy node, carbon emission factor node or corresponding relationship to form a knowledge graph structure with time index attribute.

3. The method for real-time carbon footprint calculation of leather products based on knowledge graphs according to claim 2, characterized in that, Step 3 includes the following steps: Read the node identifiers and relationship identifiers of product nodes, process nodes, resource nodes, energy nodes, carbon emission factor nodes, and their corresponding relationships; generate probabilistic soft logic rule entries according to the node type identifier and relationship type identifier, and record the rule number. Generate mapping records between rules and node identifiers, and between rules and association identifiers, based on the node identifiers and association identifiers referenced in each probability soft logic rule entry; The rule index structure is formed by combining rule numbers with their corresponding mapping records. The rule index structure records the correspondence between rule numbers and the set of node identifiers and the set of association identifiers.

4. The method for real-time carbon footprint calculation of leather products based on knowledge graphs according to claim 3, characterized in that, Step 4 specifically includes: Read the product node, process node and corresponding time index attribute, sort the process nodes according to the time index order, generate life cycle stage identifiers based on the sorting results and collect them to form a life cycle stage set. Read the product nodes, process nodes, and the association relationship identifiers between product nodes and process nodes. Generate an ordered node sequence consisting of product nodes and process nodes according to the connection order of the association relationship identifiers, and record the ordered node sequence identifiers. The system aggregates ordered node sequence identifiers to form a set of accounting paths, establishes an index association between lifecycle stage identifiers and corresponding ordered node sequence identifiers, and records the associations.

5. The method for real-time carbon footprint calculation of leather products based on knowledge graphs according to claim 4, characterized in that, Step 5 includes the following steps: Read the life cycle stage identifier set in the life cycle stage set and the accounting path identifier set in the accounting path set, generate a two-dimensional index space according to the life cycle stage identifier and the accounting path identifier, and allocate carbon flow numerical storage location for each index pair in the two-dimensional index space to form a carbon flow tensor variable. Read the rule number, node identifier set, and association identifier set recorded in the rule index structure. Locate the life cycle stage identifier subset according to the node identifier set corresponding to the rule number, locate the accounting path identifier subset according to the association identifier set corresponding to the rule number, and combine the life cycle stage identifier subset and the accounting path identifier subset to generate the tensor slice index range corresponding to the rule. Extract rule-acting tensor slices from the carbon flow tensor variable according to the tensor slice index range, and record the mapping relationship between the rule number and the rule-acting tensor slice; For each rule-action tensor slice, a low-rank constraint parameter set is constructed, which includes the rank upper limit parameter and the slice decomposition dimension parameter. The rule-action tensor slices are constructed by performing a low-rank representation operation on the low-rank constraint parameter set, and the rule-action tensor slices are represented as tensor decomposition forms under constrained rank conditions. Write the low-rank representation into the constraint description field of the carbon flow tensor variable.

6. The method for real-time carbon footprint calculation of leather products based on knowledge graphs according to claim 5, characterized in that, Step 5, constructing the low-rank constraint parameter set, specifically includes: Read the rule number, node identifier set, and association relationship identifier set recorded in the rule index structure, and locate the corresponding rule action tensor slice identifier set according to the rule number; For each rule-effect tensor slice identifier, read the lifecycle stage identifier subset and the calculation path identifier subset corresponding to the rule-effect tensor slice, and count the number of identifiers in the lifecycle stage identifier subset and the number of identifiers in the calculation path identifier subset. The slice dimension scale parameter is generated according to the number of lifecycle stage identifiers and the number of accounting path identifiers. The slice dimension scale parameter records the size information of the rule-effect tensor slice in the lifecycle stage index dimension and the accounting path index dimension. Based on the slice dimension scale parameter, the slice decomposition dimension parameter is generated. The slice decomposition dimension parameter records the set of dimension numbers and the set of dimension order numbers of the tensor slices that participate in the tensor decomposition calculation according to the rule. The rank upper limit parameter is determined according to the dimensional scale value corresponding to the slice decomposition dimensional parameter. The rank upper limit parameter records the range of rank values ​​that the rule applies to the tensor slice during the tensor decomposition calculation. The slice decomposition dimension parameters and the rank upper limit parameters are combined to form a set of low-rank constraint parameters that correspond one-to-one with the rule action tensor slice identifier, and the correspondence between the rule number, the rule action tensor slice identifier and the low-rank constraint parameter set is recorded.

7. The method for real-time carbon footprint calculation of leather products based on knowledge graphs according to claim 6, characterized in that, Step 6 specifically includes: Read the carbon flow tensor variable, the time index data set, and the low-rank constraint parameter set, and select the tensor elements in the carbon flow tensor variable that are consistent with the time index to form the original variable set; The data consistency items are calculated based on the resource consumption values, energy consumption values, and carbon emission factor values ​​corresponding to the original variable set and time index. The rule constraint items are calculated based on the rule number, node identifier set, and association relationship identifier set recorded in the rule index structure. The low-rank constraint items are calculated based on the low-rank constraint parameter set. The objective function is formed by combining the data consistency term, the rule constraint term, and the low-rank constraint term. The original variable set and the dual variable set are initialized for the objective function. The original variable set and the dual variable set are iteratively calculated according to the original-dual mixed gradient update order. The original variable and the dual variable corresponding to the time index are output.

8. The method for real-time carbon footprint calculation of leather products based on knowledge graphs according to claim 7, characterized in that, Step 7 includes the following steps: Read the data set corresponding to the current time index and the previous time index, perform item-by-item numerical difference operation on the resource consumption value, energy consumption value and carbon emission factor value according to the time index order, generate a data increment vector containing the incremental components of resource consumption, energy consumption and carbon emission factors, and record the correspondence between the data increment vector and the time index. Read the original variable set and dual variable set corresponding to the previous time index, read the rule-induced low-rank expression form, rank upper limit parameter and tensor decomposition dimension parameter, within the tensor decomposition dimension space limited by the low-rank expression form, map the original variable set and dual variable set to tensor decomposition factor numerical set, perform incremental update operation on the tensor decomposition factor numerical set according to the data increment vector, generate the predicted tensor decomposition factor numerical set, and reconstruct the predicted original variable set and predicted dual variable set from the predicted tensor decomposition factor numerical set; Using the original set of predicted variables and the dual set of predicted variables as the initial state, the data consistency term, rule constraint term, and low-rank constraint term corresponding to the current time index are read. The original-dual update operation is performed on the objective function a preset number of times. In each original-dual update operation, the original set of variables and the dual set of variables are updated synchronously. After each update, the rule residual value and the low-rank consistency residual value are calculated. The rule residual value and the low-rank consistency residual value are compared with the corresponding threshold. When the comparison result meets the convergence judgment condition, the updated original set of variables and the updated dual set of variables are output as the updated variables corresponding to the current time index.

9. A knowledge graph-based real-time carbon footprint accounting system for leather products, applied to the knowledge graph-based real-time carbon footprint accounting method for leather products as described in any one of claims 1 to 8, characterized in that, Includes the following modules: The data acquisition module is used to collect time-indexed data on leather products, processes, resources, energy, and carbon emission factors. The knowledge graph construction module is used to build a knowledge graph containing product nodes, process nodes, resource nodes, energy nodes and factor nodes, and to establish temporal relationships between nodes; The rule engine module is used to build a set of probabilistic soft logic rules and establish a rule index structure, defining the mapping relationship between rules and nodes and their associations. The Lifecycle and Path module is used to construct the lifecycle stage set and the accounting path set, and generate the lifecycle stage and accounting path index of the carbon flow tensor variable; The low-rank constraint module is used to introduce rule-induced low-rank constraints on carbon flux tensor slices. The optimization computation module is used to construct an objective function that includes data consistency terms, rule constraint terms, and low-rank constraint terms, and to execute the primal-dual hybrid gradient algorithm. The prediction and correction module is used to predict and update the original and dual variables of adjacent time indices based on the data increment, and output the carbon footprint value and path contribution results.

Images