A method and apparatus for verifying and managing carbon emission data of energy enterprises

Abstract

This application provides a method and apparatus for verifying and managing carbon emission data of energy enterprises, relating to the field of carbon emission regulatory technology, and solving the technical problem that existing technologies cannot quantify the uncertainty and reliability of carbon emission data. The method specifically includes: acquiring enterprise carbon emission data; preprocessing the enterprise carbon emission data to construct a carbon emission dataset; constructing a directed acyclic graph (DAG) based on the carbon emission dataset; using the MCMC algorithm, updating the posterior distribution of each node in the DAG based on the historical prior distribution and the carbon emission dataset, and calculating the confidence level of each node; performing multi-level verification based on the confidence level of each node, calculating the contribution of each source node to the uncertainty of the target node; ranking each source node based on the contribution level to identify key abnormal nodes; generating correction suggestion values ​​and correction confidence intervals based on the key abnormal nodes, determining the correction suggestions, and pushing them out. This application is used for the verification and management of carbon emission data of energy enterprises.

Description

Technical Field

[0001] This application relates to the field of carbon emission regulation technology, and in particular to a method and apparatus for verifying and managing carbon emission data of energy enterprises. Background Technology

[0002] Carbon emission data management is a crucial foundation for enterprises to achieve their carbon peaking and carbon neutrality goals. Currently, energy companies typically employ fixed formula calculations or single-dimensional anomaly detection methods for carbon emission data management. However, existing carbon emission accounting methods can only achieve simple statistical analysis of data, failing to quantify the inherent uncertainty and reliability of the data itself. They are ill-equipped to identify the root causes of data deviations, and existing data verification methods lack consideration for the complex dependencies between data points, resulting in low verification accuracy and an inability to pinpoint key factors affecting the final carbon emission results. When data anomalies occur, current technologies lack effective traceability mechanisms and dynamic correction schemes, hindering the formation of closed-loop management and making it difficult to continuously improve data quality. This fails to meet the high reliability requirements of enterprises' refined carbon management and environmental regulatory departments. Therefore, existing technologies suffer from the technical problem of being unable to quantify the uncertainty and reliability of carbon emission data. Summary of the Invention

[0003] This application provides a method and apparatus for verifying and managing carbon emission data of energy enterprises, which solves the technical problem that the existing technology cannot quantify the uncertainty and reliability of carbon emission data.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for verifying and managing carbon emission data of energy enterprises is provided, including: acquiring enterprise carbon emission data; preprocessing the enterprise carbon emission data to construct a carbon emission dataset; constructing a directed acyclic graph (DAG) based on the carbon emission dataset; the nodes of the DAG include source nodes, computation nodes, and target nodes, and directed edges are established between nodes based on the dependency relationship of carbon emission accounting; the dependency relationship is a logical association relationship of carbon emission accounting, which is the transmission from source nodes to computation nodes, the hierarchical transmission between computation nodes, and the transmission from computation nodes to target nodes; using the MCMC algorithm, updating the posterior distribution of each node in the DAG based on the historical prior distribution and the carbon emission dataset, and calculating the confidence level of each node; performing multi-level verification based on the confidence level of each node, and calculating the contribution of each source node to the uncertainty of the target node; the multi-level verification includes single-node self-verification, inter-node mutual verification, and global consistency verification; sorting each source node based on the contribution level to determine key abnormal nodes; and generating correction suggestion values ​​and correction confidence intervals based on key abnormal nodes, determining correction suggestions, and pushing them.

[0005] In conjunction with the first aspect mentioned above, one possible implementation involves preprocessing enterprise carbon emission data to construct a carbon emission dataset. This includes: standardizing the data format and units of measurement for enterprise carbon emission data, and establishing a standardized data dictionary; aligning the data time sequence of enterprise carbon emission data with the DCS system timestamp as a benchmark; using the 3σ criterion combined with empirical thresholds to remove abnormal data from enterprise carbon emission data; processing missing values ​​according to their duration using linear interpolation, contemporaneous mean interpolation, or LSTM neural network interpolation; integrating the preprocessed enterprise carbon emission data; labeling the processing status; and forming a carbon emission dataset.

[0006] In conjunction with the first aspect mentioned above, one possible implementation involves constructing a directed acyclic graph (DAG) based on the carbon emission dataset. This includes: dividing the data within the carbon emission dataset into three categories: source nodes, computation nodes, and target nodes; source nodes are the original measurement data nodes for enterprise carbon emission accounting, including fuel consumption, coal quality test data, CEMS monitoring data, and unit operating condition data; computation nodes are intermediate data nodes calculated based on the source node data, including daily fuel consumption, daily carbon emissions, and monthly cumulative carbon emissions; and target nodes are the total carbon emission data nodes ultimately reported by the enterprise. Based on the logical dependencies of carbon emission accounting, directed edges are established between nodes, and a directed acyclic graph is constructed based on the source nodes, computation nodes, target nodes, and directed edges; the directed edges satisfy the following conditions: source nodes point to computation nodes, computation nodes point to computation nodes, and computation nodes point to target nodes.

[0007] In conjunction with the first aspect mentioned above, one possible implementation involves using the MCMC algorithm to update the posterior distribution of each node in the directed acyclic graph based on the historical prior distribution and the carbon emission dataset, and calculating the confidence of each node. This includes: setting the historical prior distribution of each node, where the source node adopts a normal distribution, the computation node adopts a log-normal distribution, and the target node adopts a Gamma distribution; using the NUTS sampling method of the MCMC algorithm, combining the historical prior distribution and the carbon emission dataset, iteratively updating the posterior distribution of each node; and calculating the confidence of each node after the update.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the confidence level of the updated cutover point is calculated according to the following formula:

[0009] in, For the true value of node i, Let be the observation value at node i, δ be the allowable error range for node i, and D be the carbon emission dataset. Represents conditional probability. Let be the confidence level of node i.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, multi-level verification is performed based on the confidence level of each node to calculate the contribution of each source node to the uncertainty of the target node. This includes: based on the confidence level of each node, verification operations are performed sequentially in the following order: single-node self-verification, inter-node mutual verification, and global consistency verification. Single-node self-verification: The confidence level of each node is compared with a pre-set confidence level threshold and the node's historical confidence interval to determine whether the node data passes verification, triggers a verification warning, or fails verification. Inter-node mutual verification: Based on the nodes in the directed acyclic graph... Dependency analysis: This involves assessing the consistency of data changes between logically related nodes, calculating the correlation coefficient of related nodes and comparing it with a preset correlation threshold to determine if the logic between nodes is abnormal; Global consistency verification: Weight coefficients are assigned to each node according to node type, and a weighted average method is used to calculate the global confidence level. The global confidence level is then compared with a preset global verification threshold to determine the quality level of the overall carbon emission dataset; Based on the confidence level of each node and the dependencies between nodes in the directed acyclic graph, the variance ratio method is used to calculate the contribution of each source node to the uncertainty of the target node.

[0011] In conjunction with the first aspect mentioned above, one possible implementation involves sorting the source nodes based on their contribution to determine key abnormal nodes. This includes: sorting all source nodes in descending order of their contribution to the uncertainty of the target node, generating a source node contribution ranking table; selecting source nodes from the sorted nodes that have been determined to have failed verification / verification warning by single-node self-verification and have been determined to be logically abnormal by inter-node mutual verification, forming an abnormal source node set; selecting the source nodes with the highest contribution ranking from the abnormal source node set as key abnormal nodes; the selection rule is to select the top N nodes in terms of contribution ranking, or to select nodes whose contribution value exceeds a preset contribution threshold, where N is a positive integer, and the contribution threshold is set according to the enterprise's carbon emission accounting accuracy requirements and historical data characteristics.

[0012] In conjunction with the first aspect mentioned above, one possible implementation involves generating suggested correction values ​​and corrected confidence intervals based on key anomaly nodes, determining and pushing corrected suggestions, including: extracting key anomaly node data and strongly dependent, trusted node data, removing anomalies, and establishing a standardized correction benchmark dataset; predicting reasonable values ​​for key anomaly nodes using the MCMC algorithm as suggested correction values; calculating corrected confidence intervals using the Bootstrap sampling statistical method based on node data characteristics; classifying correction levels based on node influence weights to determine corrected suggestions; and pushing the suggested correction values, corrected confidence intervals, and corrected suggestions in a targeted manner.

[0013] In conjunction with the first aspect mentioned above, in one possible implementation, the contribution of each source node to the uncertainty of the target node is calculated using the variance ratio method, which satisfies the following formula:

[0014] in, The contribution of source node i. The true value of the target node. The true value of source node i. This is the variance calculation function. Given the true value of source node i, the conditional expectation of the target node.

[0015] Secondly, a device for verifying and managing carbon emission data of energy enterprises is provided, comprising: a communication unit and a processing unit; the communication unit is used to acquire enterprise carbon emission data; the processing unit is used to preprocess the enterprise carbon emission data to construct a carbon emission dataset; a directed acyclic graph is constructed based on the carbon emission dataset; the nodes of the directed acyclic graph include source nodes, computing nodes, and target nodes, and directed edges are established between nodes based on the dependency relationship of carbon emission accounting; the dependency relationship is the logical association relationship of carbon emission accounting, which is the transmission from source nodes to computing nodes, the hierarchical transmission between computing nodes, and the transmission from computing nodes to target nodes; using the MCMC algorithm, the posterior distribution of each node in the directed acyclic graph is updated based on the historical prior distribution and the carbon emission dataset, and the confidence of each node is calculated; based on the confidence of each node, multi-level verification is performed to calculate the contribution of each source node to the uncertainty of the target node; the multi-level verification includes single-node self-verification, inter-node mutual verification, and global consistency verification; each source node is sorted based on the contribution to determine key abnormal nodes; based on the key abnormal nodes, correction suggestion values ​​and correction confidence intervals are generated, correction suggestions are determined and pushed.

[0016] This application provides a method and apparatus for verifying and managing carbon emission data of energy enterprises. It can quantify the confidence level of each node of carbon emission data based on the MCMC algorithm, realize the comprehensive verification of carbon emission data through multi-level verification, and quantify the contribution of each source node to the uncertainty of the target node to accurately locate key abnormal nodes. It can also generate targeted correction suggestions and confidence intervals and push them to the relevant departments. This effectively quantifies the uncertainty and credibility of carbon emission data, realizes accurate verification, source tracing and correction of carbon emission data, improves the accuracy and credibility of carbon emission data, and meets the relevant requirements of enterprises' refined carbon management and environmental supervision.

[0017] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a method for verifying and managing carbon emission data of an energy company, provided as an embodiment of this application; Figure 2 A flowchart illustrating another method for verifying and managing carbon emission data of energy enterprises provided in this application embodiment; Figure 3 A flowchart illustrating another method for verifying and managing carbon emission data of energy enterprises provided in this application embodiment; Figure 4 This application provides a schematic diagram of the structure of a directed acyclic graph. Figure 5 A flowchart illustrating another method for verifying and managing carbon emission data of energy enterprises provided in this application embodiment; Figure 6 A flowchart illustrating another method for verifying and managing carbon emission data of energy enterprises provided in this application embodiment; Figure 7 A flowchart illustrating another method for verifying and managing carbon emission data of energy enterprises provided in this application embodiment; Figure 8 A flowchart illustrating another method for verifying and managing carbon emission data of energy enterprises provided in this application embodiment; Figure 9 This is a schematic diagram of the structure of an energy enterprise carbon emission data verification and management device provided in an embodiment of this application. Detailed Implementation

[0019] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0020] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0021] To address the technical problem of existing technologies failing to quantify the uncertainty and reliability of carbon emission data, this application provides a method for verifying and managing carbon emission data for energy enterprises. The method includes: acquiring and preprocessing enterprise carbon emission data to construct a carbon emission dataset; constructing a directed acyclic graph (DAG) containing source nodes, computation nodes, and target nodes based on the dataset; updating the posterior distribution of each node using the MCMC algorithm combined with historical prior distributions and calculating confidence levels; conducting multi-level verification based on the confidence levels and calculating the contribution of each source node to the uncertainty of the target node; identifying key anomalous nodes by ranking them according to contribution; generating correction suggestion values ​​and correction confidence intervals based on the key anomalous nodes and pushing the correction suggestions. Based on this, the uncertainty and reliability of each node in the carbon emission data can be accurately quantified. Multi-level verification enables comprehensive verification of carbon emission data, quantifies and locates key anomalous nodes affecting the final carbon emission results, and provides targeted correction suggestions with confidence intervals. This achieves reliable management of carbon emission data throughout the entire process from verification to source tracing and correction, effectively improving the accuracy and reliability of carbon emission data and meeting the requirements of refined carbon management and environmental supervision for enterprises.

[0022] like Figure 1 As shown in the embodiments of this application, the method for verifying and managing carbon emission data of energy enterprises includes: S101. Obtain enterprise carbon emission data.

[0023] Among them, carbon emission data covers various original monitoring, measurement and testing data required for carbon emission accounting by energy companies.

[0024] In this embodiment, the energy enterprise carbon emission data verification and management device acquires data through multi-channel collaboration, covering all data related to carbon emission accounting in the enterprise's production process, ensuring the comprehensiveness and real-time nature of data collection.

[0025] It should be noted that when collecting data, the data source, collection time, and corresponding collection device number will be recorded simultaneously to achieve preliminary traceability management of the collected data.

[0026] As an example, it connects to the enterprise's CEMS and DCS systems to achieve automatic data collection, while also receiving manually entered coal quality test data to complete the acquisition of carbon emission data.

[0027] Based on the above steps, comprehensive collection of carbon emission data through multiple channels was achieved, effectively improving the integrity and comprehensiveness of the data source.

[0028] S102. Preprocess the enterprise's carbon emission data to construct a carbon emission dataset.

[0029] Among them, data preprocessing is the process of standardizing and purifying the raw collected data, and the carbon emission dataset is a structured and standardized data set after being organized.

[0030] In this embodiment, the energy enterprise carbon emission data verification and management device performs unified organization, time sequence alignment, abnormal data processing, and missing value completion on the collected carbon emission data, integrates the processed data, and marks the processing status to form a structured carbon emission dataset.

[0031] It should be noted that processing logs are kept for each step during the preprocessing process to record the data processing methods and results, which facilitates subsequent traceability of the data processing process.

[0032] Based on the above steps, the format differences and noise interference of the original data are eliminated, the usability of the data is improved, and the problem of strong heterogeneity of carbon emission data and difficulty in direct use is solved.

[0033] S103. Construct a directed acyclic graph based on the carbon emission dataset.

[0034] The directed acyclic graph consists of source nodes, computation nodes, target nodes, and directed edges between nodes. The directed edges between nodes are established based on the logical dependencies of carbon emission accounting.

[0035] In this embodiment, the energy enterprise carbon emission data verification and management device first classifies the data in the carbon emission dataset into nodes, then sorts out the logical relationships of carbon emission accounting between various types of nodes, and establishes directed edges between nodes to construct a complete directed acyclic graph.

[0036] It should be noted that the constructed directed acyclic graph can be flexibly updated according to the adjustment of the enterprise's carbon emission accounting process, adapting to the dynamic changes in the enterprise's production and accounting requirements.

[0037] As an example, fuel consumption data is set as the source node, daily fuel consumption is set as the calculation node, and directed edges are established from the source node to the calculation node to gradually complete the construction of the entire graph.

[0038] Based on the above steps, the carbon emission accounting logic is transformed into a visualized node relationship graph, which clarifies the accounting dependencies between data and overcomes the limitation of not being able to intuitively represent the relationship between carbon emission data.

[0039] S104. Using the MCMC algorithm, update the posterior distribution of each node in the directed acyclic graph based on the historical prior distribution and the carbon emission dataset, and calculate the confidence of each node.

[0040] Among them, the MCMC algorithm is the Markov chain Monte Carlo algorithm. The prior distribution is the node data distribution characteristics obtained based on historical data, the posterior distribution is the node data distribution updated by combining the real-time dataset, and the confidence level is an indicator that quantifies the credibility of the node data.

[0041] In this embodiment, the energy enterprise carbon emission data verification and management device first sets an appropriate historical prior distribution for each node, and then iteratively updates the posterior distribution of each node by combining the carbon emission dataset with the MCMC algorithm, and calculates the confidence level of each node based on the updated distribution.

[0042] It should be noted that the device will update the posterior distribution and confidence level of the nodes in real time, and the update frequency will match the frequency of data acquisition and processing to ensure the real-time nature of the confidence level.

[0043] As an example, a normal prior distribution is set for the source node, and the posterior distribution is updated by iterative calculation using the sampling method of the MCMC algorithm, and the confidence value of the source node is calculated.

[0044] Based on the above steps, a quantitative assessment of the reliability of data at each node was achieved, solving the core problem that existing technologies cannot quantify the uncertainty and reliability of carbon emission data, and providing a quantitative basis for subsequent data verification.

[0045] S105. Based on the confidence level of each node, perform multi-level verification and calculate the contribution of each source node to the uncertainty of the target node.

[0046] Among them, the multi-level verification includes single-node self-verification, inter-node mutual verification, and global consistency verification. The contribution is an indicator that quantifies the degree of uncertainty of the source node data on the target node data.

[0047] In this embodiment of the application, the energy enterprise carbon emission data verification and management device uses the confidence level of each node as the core basis to perform verification operations at different levels in sequence. At the same time, it combines the dependency relationship between nodes and uses a quantitative algorithm to calculate the uncertainty contribution of each source node to the target node.

[0048] It should be noted that during the multi-level verification process, the verification results and abnormal node information at each level will be recorded to form a complete verification record archive.

[0049] As an example, we first determine whether the confidence level of a single node meets the requirements, then verify the logical consistency between the fuel consumption and daily carbon emission nodes, and finally calculate the global confidence level to complete the verification and calculate the contribution value of each source node.

[0050] Based on the above steps, a multi-dimensional and comprehensive verification of carbon emission data was achieved. At the same time, the influence of source nodes on target nodes was quantified, solving the problem that existing technologies have a single verification dimension and cannot locate the influencing nodes.

[0051] S106. Sort each source node based on its contribution and identify key abnormal nodes.

[0052] Among them, the key abnormal nodes are the source nodes that have a significant impact on the uncertainty of the target node and have been determined to have data anomalies through multi-level verification.

[0053] In this embodiment, the energy enterprise carbon emission data verification and management device sorts each source node from largest to smallest contribution, and combines the results of multi-level verification to screen out nodes with abnormalities from the sorted source nodes, thereby identifying key abnormal nodes.

[0054] It should be noted that the screening rules for identifying key anomaly nodes can be flexibly adjusted according to the company's requirements for carbon emission accounting accuracy.

[0055] Based on the above steps, the abnormal nodes that have the greatest impact on the final carbon emission results were accurately located, avoiding blind investigation and greatly improving the efficiency and accuracy of abnormal node location.

[0056] S107. Based on key anomaly nodes, generate suggested correction values ​​and correction confidence intervals, determine correction suggestions, and push them out.

[0057] Among them, the suggested correction value is a reasonable data value for the key abnormal node, the confidence interval of the correction is the credible range of the suggested correction value, and the correction suggestion is a specific rectification guide for the abnormal node.

[0058] In this embodiment, the energy enterprise carbon emission data verification and management device generates corresponding correction suggestion values ​​and correction confidence intervals based on the data of key abnormal nodes and associated trusted node data, determines correction suggestions in combination with node abnormalities, and pushes them to the corresponding management terminal.

[0059] It should be noted that when pushing out correction suggestions, relevant information about the abnormal nodes and the basis for correction will be attached simultaneously to provide data support for managers to make rectifications.

[0060] As an example, for the key anomaly node of belt scale fuel consumption, a reasonable correction suggestion value and corresponding confidence interval are generated, the correction suggestion for equipment calibration is determined and pushed to the metering management terminal.

[0061] Based on the above steps, quantitative correction references and specific rectification suggestions are provided for key abnormal nodes, solving the problem of no effective correction plan after data anomalies and realizing targeted and accurate correction of abnormal data.

[0062] Based on the above technical solutions, the method of this application embodiment realizes the full-process reliable management of carbon emission data of energy enterprises from collection, preprocessing, modeling, verification, anomaly location to correction. It not only quantifies the uncertainty and reliability of carbon emission data, but also realizes the comprehensive verification of data through multi-level verification, accurately locates key anomaly nodes and provides correction suggestions with confidence intervals, effectively improving the accuracy and reliability of carbon emission data and meeting the relevant requirements of refined carbon management and environmental supervision of energy enterprises.

[0063] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 2 As shown, the above S102 can be specifically implemented through the following S201, S202, S203 and S204, which are explained in detail below: S201. Standardize the data format and unit of measurement for enterprise carbon emission data, and establish a standardized data dictionary.

[0064] The standardized data dictionary contains core information such as data field names, types, uniform units, and value ranges.

[0065] In this embodiment, the energy enterprise carbon emission data verification and management device parses the carbon emission data from different sources, unifies the time format to a standard timestamp, converts similar data from different units of measurement into a unified standard, and simultaneously develops and archives a standardized data dictionary.

[0066] It should be noted that the data dictionary can be dynamically updated according to the enterprise's new data sources or accounting requirements to ensure adaptability.

[0067] As an example, the lower heating value of coal is uniformly converted from "kcal / kg" to "kJ / kg", and the unified unit and conversion rules for this field are clearly defined in the data dictionary.

[0068] Based on the above steps, the differences in format and unit of multi-source data are eliminated, the problem of strong data heterogeneity in the prior art is solved, and the compatibility and comparability of data are improved.

[0069] S202. Based on the DCS system timestamp, align the data time sequence of enterprise carbon emission data.

[0070] In this embodiment, the energy enterprise carbon emission data verification and management device extracts the timestamp information of each data source, uses the high-precision timestamp of the DCS system as a benchmark, adjusts the time dimension of other data sources through linear interpolation and other methods, and associates non-real-time data with the corresponding accounting period in batches.

[0071] As an example, the data collected by the belt scale every 10 minutes is aligned with the time grid of the DCS system every minute to ensure that the data can be matched in the same time dimension.

[0072] Based on the above steps, the problem of time sequence misalignment of multi-source data is solved, ensuring the consistency of data in the time dimension and providing reliable support for subsequent correlation calculations.

[0073] S203. Abnormal data in enterprise carbon emission data are removed by using the 3σ criterion combined with empirical thresholds.

[0074] In this embodiment, the energy enterprise carbon emission data verification and management device first calculates the mean μ and standard deviation σ of each data sequence, removes statistical outliers that exceed the range of [μ-3σ, μ+3σ], and then compares the remaining data with preset industry experience thresholds or enterprise production parameter thresholds to remove logical outliers.

[0075] It should be noted that the experience threshold can be dynamically adjusted according to the company's production process, equipment characteristics, and industry standards.

[0076] As an example, outliers in fuel consumption that are outside the statistical range are first removed, and then the remaining data are compared with the reasonable consumption threshold corresponding to the maximum load of the equipment to further filter out invalid data.

[0077] Based on the above steps, the dual screening method can accurately remove abnormal data, which effectively reduces the risk of accidental or missed deletion and improves the data purification effect compared with the single screening method.

[0078] S204. Based on the duration of the missing data, linear interpolation, synchronous mean interpolation, or LSTM neural network interpolation are used to process the missing values. The preprocessed enterprise carbon emission data are then integrated, and the processing status is labeled to form a carbon emission dataset.

[0079] In this embodiment, the energy enterprise carbon emission data verification and management device first determines the duration of missing data, selects the corresponding interpolation method according to preset rules, completes the missing value filling, integrates all preprocessed data, and marks the processing status of each data (original / unit conversion / time sequence alignment / anomaly removal / interpolation).

[0080] It should be noted that all interpolation operations are logged, recording the interpolation method, parameters, and basis, which facilitates subsequent traceability.

[0081] As an example, the missing 40 minutes of operating data is imputed using the average value of adjacent time periods, and the missing 3 hours of fuel consumption data is imputed using an LSTM neural network.

[0082] Based on the above steps, the imputation strategy is flexibly adapted according to the duration of the missing data, which ensures the accuracy of the completion while controlling the computational cost. At the same time, the data can be traced throughout the entire process through status labeling.

[0083] Based on the above technical solutions, a complete multi-source carbon emission data preprocessing process is formed. The device effectively solves the problems of strong heterogeneity, disordered time sequence, noise interference and insufficient integrity of carbon emission data in the prior art through the coordinated processing of unified format, time sequence alignment, anomaly removal and missing value imputation. It generates a standardized and high-quality carbon emission dataset, which provides a reliable data foundation for subsequent construction of directed acyclic graphs and confidence assessment.

[0084] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 3 As shown, the above S103 can be implemented through the following S301 and S302, which are explained in detail below: S301. Divide the data in the carbon emission dataset into three categories: source nodes, computing nodes, and target nodes.

[0085] Among them, the source node is the original measurement data node for the enterprise's carbon emission accounting, including fuel consumption, coal quality test data, CEMS monitoring data and unit operating data; the calculation node is the intermediate data node calculated based on the source node data, including daily fuel consumption, daily carbon emissions and monthly cumulative carbon emissions; the target node is the total carbon emission data node finally reported by the enterprise.

[0086] In this embodiment, the carbon emission data verification and management device for energy enterprises classifies carbon emission datasets according to the data source attributes and accounting roles, and clarifies the core data scope and functional positioning of each type of node.

[0087] Specifically, source nodes represent the physical source of data acquisition. Their data characteristics typically originate directly from sensors or laboratory reports, possessing high originality and independence. For example, fuel consumption data usually comes from metering equipment such as belt scales and track scales; coal quality analysis data comes from elemental analysis results in laboratories; CEMS monitoring data comes from online monitoring instruments within the flue; and unit operating condition data comes from parameters such as load, temperature, and pressure recorded by the DCS system. It should be understood that the specific content of source nodes is not limited to the above examples. Depending on the company's production process, source nodes can also include other raw parameters such as fuel oil density and natural gas composition analysis data. Calculation nodes represent intermediate states in the data processing process. Their values ​​depend on the input of the parent node and are calculated using specific accounting formulas. For example, daily fuel consumption is a cumulative calculation of instantaneous fuel consumption, and daily carbon emissions are calculated based on the product of fuel consumption and carbon content. Target nodes are the final output of carbon emission accounting, typically corresponding to key indicators in the company's carbon emission report, such as the annual total carbon emissions. This hierarchical classification deconstructs the complex accounting system into a clear hierarchical structure, facilitating subsequent tracking of the transmission path of data deviations.

[0088] Based on the above steps, hierarchical classification of data is achieved, avoiding confusion between different types of data, and improving the organization and targeting of data management compared with existing technologies.

[0089] S302. Based on the logical dependencies of carbon emission accounting, establish directed edges between nodes, and construct a directed acyclic graph based on the source node, computing node, target node, and directed edges.

[0090] Among them, directed edges are used to represent the computational dependency direction between nodes, that is, the logical order of data flow and computation.

[0091] In this embodiment, the energy enterprise carbon emission data verification and management device sorts out the accounting logic relationship of various nodes, establishes directed edges according to the rules of source node to computing node, hierarchical transmission between computing nodes, and computing node to target node, and integrates nodes and directed edges to construct a complete directed acyclic graph.

[0092] Specifically, directed edges represent the physical logic of data flow and dependencies, and the direction of the arrow indicates the relationship of "who depends on whom," that is, the node pointed to by the arrow depends on the node at the tail of the arrow. For example... Figure 4As shown, during the construction process, the device first identifies the direct dependencies between source nodes and computing nodes. For example, the source node "belt scale fuel consumption" points to the computing node "daily fuel consumption," indicating that the calculation of daily consumption depends on the real-time reading of the belt scale. A directed edge is established between the source node "coal quality as received carbon content" and the computing node "daily carbon emissions," indicating that the calculation of carbon emissions depends on coal quality parameters. Secondly, the device identifies the hierarchical dependencies between computing nodes. For example, the computing node "daily fuel consumption" points to the computing node "daily carbon emissions," indicating that the calculation of daily emissions uses daily consumption as input. The computing node "daily carbon emissions" points to the computing node "monthly cumulative carbon emissions," indicating that monthly data is derived from the aggregation of daily data. Finally, the device establishes directed edges between computing nodes and target nodes. For example, the computing node "monthly cumulative carbon emissions" points to the target node "total carbon emissions," indicating that the final reported data is generated by aggregating intermediate calculation results. The directed acyclic graph constructed using the aforementioned pointing rules clearly demonstrates the entire chain logic of carbon emission data from initial collection and intermediate calculation to final output. This allows data fluctuations at any node to be traced along the directed edges to their impact on downstream nodes and even the final result, laying a topological foundation for subsequent confidence propagation and source tracing analysis based on Bayesian networks.

[0093] Based on the above steps, the abstract accounting logic is transformed into a visual topological structure, clearly presenting the dependencies between data, solving the problem of ambiguous data associations, and providing an intuitive model for subsequent uncertainty analysis.

[0094] Based on the above technical solution, by classifying nodes and constructing directed edges, multi-source carbon emission data is organized into a structured directed acyclic graph according to the accounting logic. This clarifies the hierarchical relationship and flow path of the data, solves the pain points of chaotic carbon emission data association and unintuitive accounting logic, and provides solid model support for subsequent confidence assessment, multi-level verification and source tracing.

[0095] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 5 As shown, the above S104 can be implemented through the following S501, S502 and S503, which are explained in detail below: S501, Set the historical prior distribution of each node.

[0096] Among them, the historical prior distribution is the initial probability distribution of nodes determined based on the statistical characteristics of the enterprise's past carbon emission data, which is used to provide an initial basis for posterior inference.

[0097] In this embodiment of the application, the energy enterprise carbon emission data verification and management device configures appropriate prior distribution types for source nodes, computing nodes and target nodes according to the data type, physical characteristics and historical fluctuation patterns of each node, and determines the distribution parameters based on 1-2 years of historical data training.

[0098] Specifically, the prior distribution is not chosen arbitrarily, but is based on a deep understanding of the physical characteristics of the data. For source nodes, such as fuel consumption and coal quality analysis data, their data fluctuations are usually caused by the superposition of a large number of random, small factors. According to the Central Limit Theorem, their distribution shape approximately follows a normal distribution. Therefore, this embodiment uses a normal distribution N(μ,σ²) to describe the uncertainty of the source nodes, where μ reflects the average level of the data and σ reflects the dispersion of the data. For calculation nodes, such as daily carbon emissions, their values ​​are usually obtained from the source node data through multiplication operations (e.g., consumption multiplied by carbon content). In probability theory, multiplication operations tend to produce a right-skewed distribution, and the calculation result must be non-negative. The log-normal distribution LN(μ, σ²) can fit this non-negative and right-skewed data characteristic very well, and is therefore selected as the prior distribution for the calculation nodes. For the target node, i.e., the final total carbon emissions, as the sum of multiple intermediate variables, it has a significant cumulative effect and its values ​​are continuous and non-negative. The Gamma distribution has flexible shape parameters, adapting to various forms from exponential to approximately normal distributions, making it well-suited for describing the uncertainty of such cumulative data. Through this targeted distribution setting, this embodiment can more accurately characterize the statistical regularities of data at different levels, providing a reasonable prior basis for subsequent posterior inference.

[0099] Based on the above steps, personalized configuration of the prior distribution is realized, avoiding the blindness of setting a uniform distribution and improving the initial rationality of the posterior inference.

[0100] S502. The NUTS sampling method using the MCMC algorithm combines historical prior distributions with carbon emission datasets to iteratively update the posterior distributions of each node.

[0101] Among them, the NUTS sampling method is an adaptive Hamiltonian Monte Carlo sampling method in the MCMC algorithm. It achieves efficient sampling by simulating Hamiltonian dynamics and does not require manual adjustment of the step size.

[0102] In this embodiment, due to the complex dependencies between nodes in a directed acyclic graph, the posterior distribution often cannot be directly calculated using analytical solutions. This embodiment introduces the Markov Chain Monte Carlo (MCMC) method to simulate samples of the target distribution by constructing a Markov chain. Specifically, this embodiment uses the NUTS (No-U-Turn Sampler) sampling algorithm, an adaptive Hamiltonian Monte Carlo (HMC) sampling method. Traditional random walk Metropolis algorithms are inefficient in high-dimensional spaces, while the NUTS algorithm, by simulating Hamiltonian dynamics, can make long-distance jumps along the contour lines of the probability density function, greatly improving sampling efficiency. More importantly, the NUTS algorithm can automatically adjust the step size, reaching the optimal sampling state without manual intervention, effectively avoiding low sampling efficiency or divergence caused by improper step size settings. In implementation, the system first initializes the parameter states of each node, and then runs the NUTS sampler iteratively. Typically, the number of iterations is set to several thousand to tens of thousands. The first portion of the samples is discarded as a "burning period" to eliminate the influence of the initial values, while the latter portion is used to approximate the posterior distribution. In this way, the posterior sample set of each node parameter can be obtained, thereby realizing dynamic modeling of data uncertainty.

[0103] It should be noted that the convergence of the sampling process is determined by the R-hat test. When R-hat < 1.01, the sampling is considered to be converged. If it is not converged, the number of samplings is increased or the prior distribution parameters are adjusted.

[0104] As an example, we set the iteration sampling to 10,000 times, with the first 2,000 times being the combustion period. The posterior distribution of each node is updated using the effective sampling results of the last 8,000 times, and convergence is confirmed by R-hat test.

[0105] Based on the above steps, the problems of low sampling efficiency and difficulty in adjusting step size in traditional MCMC are solved, and the posterior distribution is updated quickly and reliably, thus improving the inference efficiency.

[0106] S503. Calculate the confidence level of each node after the update.

[0107] Among them, confidence level is an indicator that quantifies the credibility of node data, reflecting the probability that the true value of a node falls within the allowable error range of the observed value.

[0108] In this embodiment of the application, the energy enterprise carbon emission data verification and management device calculates the confidence level of each node by integral calculation based on the updated posterior distribution of each node and the preset allowable error range δ.

[0109] Optionally, the confidence level satisfies the following formula:

[0110] in, Let be the actual value of node i, which is an objectively existing but unknown parameter. Let be the observed value of node i, i.e., the actual data obtained through sensors or calculations. δ is the allowable error range for node i, which is a preset threshold reflecting the business's tolerance for data accuracy. For example, for critical emission data, δ can be set to 2%, while for auxiliary parameters, δ can be appropriately relaxed. D is the carbon emission dataset. Represents conditional probability. Let be the confidence level of node i.

[0111] It should be noted that the physical meaning of this formula lies in calculating the probability that the true value of a node falls within an interval centered on the observed value and with a half-width of δ, based on a comprehensive consideration of historical prior knowledge (represented by the prior distribution) and current observational data (represented by the likelihood function). This is the confidence level, and its value ranges from [0, 1]. The closer it is to 1, the better the observed value. Compared with the true value The greater the likelihood of a high degree of agreement, the more reliable the data; conversely, the less likely a match is, the more reliable the data. The lower the confidence level, the higher the risk of significant data bias. For example, if the confidence level of a belt scale reading... A value of 0.95 means that, with a 95% probability level, the deviation between this reading and the actual fuel consumption will not exceed the preset δ value. This confidence assessment method based on probability intervals provides richer information than the traditional "yes / no" binary judgment, quantifies data risk, and provides a scientific quantitative basis for subsequent multi-level verification and traceability.

[0112] As an example, the δ value is set differently according to the node type, typically 5% for source nodes, 3% for compute nodes, and 2% for target nodes, which can be adjusted according to the enterprise's accounting accuracy requirements.

[0113] Based on the above steps, the abstract probability distribution is transformed into an intuitive and reliable quantifiable indicator, solving the problem of the inability to accurately quantify the reliability of data.

[0114] Based on the above technical solution, the uncertainty of carbon emission data nodes is quantitatively assessed through personalized prior distribution setting, efficient NUTS sampling and updating, and accurate confidence calculation. Compared with existing technologies, this significantly improves the accuracy and efficiency of data credibility assessment, laying a quantitative foundation for subsequent multi-level verification and source tracing.

[0115] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 6As shown, the above S105 can be implemented through the following S601 and S602, which are explained in detail below: S601. Based on the confidence level of each node, perform verification operations in the following order: single node self-verification, inter-node mutual verification, and global consistency verification.

[0116] Among them, single-node self-verification focuses on the data credibility of a single node, inter-node mutual verification verifies the logical consistency of related nodes, and global consistency verification evaluates the overall dataset quality.

[0117] In this embodiment, the energy enterprise carbon emission data verification and management device first compares the confidence level of each node with the classification threshold and historical confidence interval to determine the verification result of a single node; then, based on the dependency relationship of the directed acyclic graph, it calculates the correlation coefficient of the associated nodes and compares it with the association threshold to verify the logical consistency; finally, it configures weights according to node type, calculates the global confidence level by weighted average method, and determines the quality level of the dataset.

[0118] Specifically, for single-node self-verification, the device has preset confidence level thresholds, typically divided into high confidence (e.g., CI ≥ 0.9), medium confidence (e.g., 0.7 ≤ CI < 0.9), and low confidence (CI < 0.7). When a node's confidence level is in the high confidence range and does not exceed the fluctuation range of its historical confidence range, the device determines that the node has "passed verification," indicating that the data is reliable. When the confidence level is in the medium confidence range, or although it is in the high confidence range but exceeds the historical fluctuation range, it is determined as a "verification warning," prompting management personnel to pay attention to the potential risks of the data. When the confidence level is below the low confidence threshold, or deviates significantly from the historical confidence range, it is determined as a "verification failure," confirming that the node's data has significant anomalies. It should be understood that the above thresholds of 0.9 and 0.7 are only examples; in actual applications, they can be flexibly adjusted according to the enterprise's requirements for data accuracy. For example, for critical emission source nodes, the warning threshold can be increased to 0.95.

[0119] For inter-node cross-verification, the dependencies (i.e., directed edges) defined in the directed acyclic graph are used to verify whether the physical logic between parent and child nodes holds true. For example, in the directed edge "fuel consumption → daily carbon emissions," according to the principle of carbon emission accounting, an increase in fuel consumption must lead to an increase in carbon emissions, and the two should show a positive correlation. The device calculates the Pearson correlation coefficient of the data sequences of these two nodes in real time. If the correlation coefficient is lower than a preset correlation threshold (e.g., 0.8), it is determined that the logical consistency between the two is insufficient, indicating a logical anomaly. Similarly, for the correlation "unit load → fuel consumption," if the unit load increases while fuel consumption decreases, the correlation coefficient is negative, which is directly determined to be a logical anomaly, possibly indicating a metering equipment malfunction or data transmission error. Through this logic verification based on physical mechanisms, hidden anomalies can be identified where individual values ​​appear normal but the logical relationship is incorrect.

[0120] For global consistency verification, to assess the overall quality of the enterprise's carbon emission data, this embodiment uses a weighted average method to calculate the global confidence level. The weighting coefficients are set based on the importance of the nodes: the target node (total carbon emissions) has the highest weight (e.g., 0.4), followed by the calculation nodes (e.g., 0.4), and the source nodes have relatively lower weights (e.g., 0.2). The calculation formula is: CI_global = Σ(w_i × CI_i), where w_i is the weight of node i. The device presets a global verification threshold, for example, setting 0.85 as the excellent threshold and 0.75 as the acceptable threshold. If the global confidence level is higher than 0.85, the overall data quality level is determined to be "excellent"; if it is between 0.75 and 0.85, it is determined to be "good"; if it is lower than 0.75, it is determined to be "unacceptable," requiring the triggering of a global review mechanism. This global perspective assessment helps management quickly grasp the overall health of the enterprise's carbon data assets.

[0121] It should be noted that the three levels of verification are progressive, and the verification results of the previous level can provide a basis for targeted analysis of the next level, avoiding invalid calculations.

[0122] Based on the above steps, the device achieves a three-dimensional verification from local to global perspectives, which significantly improves the comprehensiveness and accuracy of data anomaly identification compared to the single-dimensional verification of existing technologies.

[0123] S602. Based on the confidence level of each node and the dependency relationship between nodes in the directed acyclic graph, the variance ratio method is used to calculate the contribution of each source node to the uncertainty of the target node.

[0124] The contribution score is used to quantify the weight of the impact of data fluctuations in a single source node on the uncertainty of the final result of the target node.

[0125] In this embodiment, after completing the above verification, the influence of each source node on the final result is further quantified. The core logic is that the uncertainty of the target node (total carbon emissions) is accumulated by the uncertainty of each source node through directed edge propagation. Using the variance ratio method, the contribution ratio of each source node to the target node's variance can be calculated. A higher contribution indicates a greater impact of the source node's data fluctuations on the final total carbon emissions, making it a key tool for data quality control. This quantitative indicator directly guides subsequent anomaly tracing and sorting, achieving a leap from qualitative verification to quantitative tracing. The energy company's carbon emission data verification and management device first determines the transmission path between the source node and the target node based on the confidence level and dependency relationship of each node. Then, it uses the variance calculation function to solve for the total variance of the target node and the expected variance of the target node under the source node conditions, substituting these values ​​into the variance ratio formula to obtain the contribution of each source node.

[0126] Optionally, the contribution satisfies the following formula:

[0127] In the formula, The contribution of source node i is denoted by , and its value ranges from [0, 1]. The larger the value, the higher the proportion of the source node's contribution to the uncertainty of the target node. This refers to the true value at the target node, that is, the objective true value of the total carbon emissions ultimately reported by the enterprise. The true value of source node i, such as the fuel consumption of a belt scale or the objective true value of the carbon content of coal received. This is a variance calculation function used to measure the dispersion of data, i.e., the magnitude of uncertainty. Given the true value of source node i, this is the conditional expectation of the target node, which is the Bayesian optimal estimate of the total carbon emissions of the target node given the data of source node i.

[0128] Specifically, the statistical principle of this formula is based on the theory of variance decomposition, with the denominator... This represents the total uncertainty at the target node (total carbon emissions), which is the overall reflection of the fluctuations in the final data quality of enterprises. This represents the uncertainty component of the target node caused by the source node i. Physically, it represents the conditional expectation of the target node when the value of the source node i is fixed. It will still fluctuate with changes in other source nodes; however, when we let source node i change freely, the variance of the conditional expectation... This reflects the magnitude of the impact of changes in source node i on the target node through the propagation path of the directed acyclic graph. Therefore, the ratio of the two is... The proportion of source node i in the total uncertainty was precisely quantified.

[0129] As an example, suppose that in a thermal power plant, the contribution of the source node "belt scale fuel consumption" is... The calculated result is 0.65, meaning that 65% of the uncertainty in the company's total carbon emission data originates from measurement fluctuations in the belt scale. In contrast, if the contribution of the source node "coal quality as received carbon content" is 0.15, it indicates that its impact on the total uncertainty is relatively small. Through this quantitative calculation, managers can intuitively identify key bottlenecks affecting data quality, thereby prioritizing limited calibration resources to the nodes with the highest contribution, achieving a leap from qualitative investigation to quantitative source tracing.

[0130] In terms of specific computational implementation, this embodiment utilizes the posterior sampling results of the MCMC algorithm for numerical integration. The device first obtains samples of the posterior distribution of the target nodes based on the constructed directed acyclic graph model using the PyMC3 equal probability programming library, and calculates its variance to obtain the denominator. Subsequently, for each source node i, the trajectory of the change in the conditional expectation of the target node when the node takes different values ​​is recorded during the sampling process, and then the numerator is calculated. It should be understood that although this embodiment provides a specific calculation formula, in other embodiments, methods based on sensitivity analysis or mutual information can also be used to quantify the contribution between nodes, as long as they can reflect the influence weight of the source node on the uncertainty of the target node.

[0131] Based on the above steps, the influence weight of the source node was quantified, which solved the problem that existing technologies could not locate key influencing nodes and provided a quantitative basis for accurate source tracing.

[0132] Based on the above technical solution, data quality issues are comprehensively investigated through multi-level verification. The influence weight of source nodes is quantified by combining the variance ratio method. This not only ensures the comprehensiveness of data verification but also enables quantifiable traceability, providing scientific support for the subsequent location of key anomalies and significantly improving the level of precision in carbon emission data management.

[0133] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 7 As shown, the above S106 can be implemented through the following S701, S702 and S703, which are explained in detail below: S701. Sort all source nodes in descending order of their contribution to the uncertainty of the target node, and generate a source node contribution ranking table.

[0134] The source node contribution ranking table includes the source node name, contribution value, and ranking information, which intuitively presents the influence weight of each source node.

[0135] In this embodiment of the application, the energy enterprise carbon emission data verification and management device extracts the contribution calculation results of all source nodes, sorts them in descending order of value, integrates the basic information of the nodes and the sorting results, generates a structured source node contribution sorting table and stores it.

[0136] It should be noted that the sorting table supports reverse sorting by contribution value or filtering by node type, adapting to query needs in different scenarios.

[0137] As an example, in the generated ranking table, "belt scale fuel consumption" ranks first with a contribution of 0.72, "coal quality as received carbon content" ranks second with a contribution of 0.35, and "CEMS CO2 concentration" ranks third with a contribution of 0.18.

[0138] Based on the above steps, the influence weight of each source node is made explicit, which solves the problem that existing technologies cannot distinguish the priority of node influence and provides clear guidance for subsequent accurate investigation.

[0139] S702. Based on the results of multi-level verification, select source nodes from the sorted source nodes that are determined to be verification failures / verification warnings by single-node self-verification and to be logically abnormal by inter-node mutual verification, and form a set of abnormal source nodes.

[0140] Among them, the abnormal source node set is a set of source nodes that simultaneously meet the dual conditions of "high contribution" and "data abnormality", providing a candidate pool for the selection of key abnormal nodes.

[0141] In this embodiment, the energy enterprise carbon emission data verification and management device retrieves the historical results of multi-level verification, matches the sorted source nodes one by one, filters out the source nodes whose single node self-verification is "verification failure" or "verification warning" and whose mutual verification between nodes is "logical abnormal", integrates them to form an abnormal source node set and marks the abnormal type.

[0142] It should be noted that the screening process uses logical AND to determine whether to retain only double abnormal nodes, effectively eliminating interference nodes that have "high contribution but normal data".

[0143] As an example, from the top 5 source nodes in the sorting table, select 2 nodes that simultaneously satisfy the self-check warning and mutual check logic exceptions to form an abnormal source node set.

[0144] Based on the above steps, accurate screening of abnormal nodes is achieved. Compared with indiscriminate screening, the scope of key attention is greatly narrowed, and the pertinence of anomaly location is improved.

[0145] S703. Select the source nodes with the highest contribution ranking from the set of abnormal source nodes as key abnormal nodes.

[0146] Among them, key abnormal nodes are the core nodes that have the greatest impact on the uncertainty of the target nodes and are indeed abnormal. They are the key targets for subsequent correction. The selection rule is to select the N nodes with the highest contribution or select the nodes whose contribution value exceeds the preset contribution threshold, where N is a positive integer. The contribution threshold is set according to the enterprise's carbon emission accounting accuracy requirements and historical data characteristics.

[0147] In this embodiment of the application, the carbon emission data verification and management device for energy enterprises can select selection rules according to the actual needs of the enterprise. If the Top N rule is selected, a positive integer N is set. If the threshold rule is selected, a contribution threshold is set based on the calculation accuracy and historical data. Nodes that meet the rules are selected from the set of abnormal source nodes as key abnormal nodes.

[0148] It should be noted that the N value and contribution threshold can be dynamically adjusted. For example, carbon trading companies can set stricter thresholds or smaller N values ​​to improve data accuracy.

[0149] As an example, N=3 is set, and the top 3 nodes in terms of contribution are selected from the set of abnormal source nodes as key abnormal nodes; or the contribution threshold is set to 0.2, and the 2 nodes in the set with a contribution of more than 0.2 are selected as key abnormal nodes.

[0150] Based on the above steps, key abnormal nodes were accurately identified, solving the problem of blind anomaly investigation and providing a clear target for subsequent targeted corrections.

[0151] Based on the above technical solution, the key abnormal nodes with the greatest impact on carbon emission results can be accurately located from all source nodes. This not only ensures the accuracy of the location but also improves the efficiency of anomaly handling, laying a core foundation for subsequent dynamic correction.

[0152] In one possible implementation of the embodiments of this application, combined with Figure 1 ,like Figure 8 As shown, the above S107 can be specifically implemented through the following S801 to S805, which are explained in detail below: S801. Extract key abnormal node data and strongly dependent and related trusted node data, remove anomalies, and establish a standardized correction benchmark dataset.

[0153] Among them, a strongly dependent and associated trusted node refers to a node that has a direct dependency relationship with a key abnormal node, has a high confidence level, and has passed multi-level verification.

[0154] In this embodiment, the energy enterprise carbon emission data verification and management device first identifies the parent / child dependent nodes of key abnormal nodes, filters out credible nodes with a confidence level ≥ 0.95 and that have passed verification, extracts data of the two types of nodes, removes data from abnormal periods, performs standardization processing, and establishes a structured correction benchmark dataset.

[0155] As an example, the data of the key anomaly node "daily carbon emissions" and the trusted associated node "coal quality based carbon content" are extracted, and after removing outliers that exceed the 3σ range, they are standardized into a benchmark dataset with a unified format.

[0156] Based on the above steps, a high-quality correction data foundation was constructed, avoiding the limitations of traditional corrections that rely on single data, and providing reliable support for accurate prediction of correction values.

[0157] S802. Predict reasonable values ​​for key abnormal nodes using the MCMC algorithm, and use these values ​​as suggested correction values.

[0158] The suggested correction values ​​are the optimal and reasonable values ​​for key anomaly nodes inferred from probability models and reliable data.

[0159] In this embodiment, the energy enterprise carbon emission data verification and management device inputs the standardized correction benchmark dataset into the Bayesian network model, fixes the values ​​of trusted nodes, iteratively samples the posterior distribution of key abnormal nodes through the MCMC algorithm, and takes the mean or median of the sampling results as the correction suggestion value.

[0160] As an example, for the abnormal node of "belt scale fuel consumption", the MCMC algorithm was used to sample 10,000 times (2,000 times during the combustion period), and the posterior distribution mean of 118 t / h was taken as the suggested correction value.

[0161] Based on the above steps, data-driven probabilistic prediction is realized. Compared with traditional interpolation or mean filling, the corrected suggested values ​​are more in line with the actual working conditions and logical connections, and the rationality is greatly improved.

[0162] S803. Combining the characteristics of node data, the Bootstrap sampling statistical method is used to calculate the corrected confidence interval.

[0163] Among them, the revised confidence interval is an indicator that quantifies the confidence range of the revised suggested value and reflects the uncertainty boundary of the revised result.

[0164] In this embodiment, the energy enterprise carbon emission data verification and management device performs 1,000 resamplings with replacement from the posterior distribution samples of key anomaly nodes obtained by the MCMC algorithm, calculates the sample mean each time, determines the 95% corrected confidence interval based on the empirical distribution, and clarifies the credible upper and lower limits of the corrected suggested value.

[0165] It should be noted that the number of samplings can be adjusted according to the fluctuation characteristics of the node data. For nodes with large fluctuations, the number of samplings can be increased to 2000 to improve the reliability of the interval.

[0166] As an example, Bootstrap sampling of the revised suggested value of 118 t / h yields a revised confidence interval of [116, 122] t / h, indicating that the value falls within this range with a 95% probability.

[0167] Based on the above steps, the uncertainty of the correction result is quantified, solving the problem that traditional corrections only provide single-point values ​​and lack risk references, thus providing managers with a more comprehensive basis for decision-making. S804. Based on the influence weight of nodes, classify the correction level and determine the correction recommendations.

[0168] Among them, the node impact weight is the contribution of each key abnormal node to the uncertainty of the target node (CU value), and the correction level reflects the urgency and priority of the correction work.

[0169] In this embodiment of the application, the energy enterprise carbon emission data verification and management device uses contribution as the core basis to divide the correction level: CU≥0.4 is level 1 (immediate processing), 0.2≤CU<0.4 is level 2 (time-limited processing), and CU<0.2 is level 3 (delayed processing). Combined with the anomaly type (such as metering drift, data delay), it generates correction suggestions containing operation guidelines.

[0170] It should be noted that the revised recommendations should be tailored to the actual production scenarios of enterprises to ensure operability and avoid vague guidance.

[0171] As an example, a node's contribution is 0.65 (Level 1 correction), the anomaly type is metering drift, and the correction suggestion is "immediately calibrate the zero point of the belt scale, review the fuel consumption data for the past 24 hours and re-enter it".

[0172] Based on the above steps, a differentiated classification of correction work was achieved, prioritizing high-impact anomalies, improving resource allocation efficiency, and avoiding blindly carrying out correction operations.

[0173] S805, push the corrected suggested value, corrected confidence interval and corrected suggestion to a targeted audience.

[0174] Among them, targeted push refers to accurately pushing information to the corresponding management terminal based on the department or responsible entity to which the key abnormal node belongs, so as to ensure that the responsibility is assigned to the person.

[0175] In this embodiment, the energy enterprise carbon emission data verification and management device identifies the domain of key abnormal nodes (such as the operation and maintenance department corresponding to metering equipment, and the laboratory corresponding to test data), integrates the correction suggestion value, confidence interval, correction level and operation guide, and pushes them to the relevant responsible persons through system pop-ups or SMS.

[0176] As an example, the correction information for "belt scale fuel consumption" is pushed to the fuel department maintenance personnel terminal. The content is "Level 1 correction warning: belt scale fuel consumption data is abnormal. The correction suggestion value is 118 t / h (confidence interval [116,122] t / h). It is recommended to immediately calibrate the equipment zero point."

[0177] Based on the above steps, a closed loop of "detecting anomalies - generating suggestions - pushing for execution" is established, which significantly improves the correction response speed and execution efficiency compared to generalized push.

[0178] Based on the above technical solutions, a complete correction loop of "benchmark construction - probability prediction - interval quantification - hierarchical suggestions - targeted push" was constructed. Through data-driven and differentiated management, the accurate correction of key anomaly nodes was achieved, which effectively improved the accuracy and credibility of carbon emission data and solved the problems of subjectivity, inefficiency and lack of quantitative support in traditional correction methods.

[0179] The above primarily describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, such as an energy enterprise carbon emission data verification and management device, includes at least one of the hardware structures and software modules corresponding to the execution of each function in order to achieve the above-mentioned functions. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0180] This application embodiment can divide the energy enterprise carbon emission data verification and management device into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or software functional units. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0181] When using integrated units, Figure 9 A possible structural schematic diagram of the energy enterprise carbon emission data verification and management device (referred to as energy enterprise carbon emission data verification and management device 90) involved in the above embodiments is shown. The energy enterprise carbon emission data verification and management device 90 includes a processing unit 901 and a communication unit 902, and may also include a storage unit 903. Figure 9 The structural diagram shown can be used to illustrate the structure of the energy enterprise carbon emission data verification and management device involved in the above embodiments.

[0182] when Figure 9 The schematic diagram shown illustrates the structure of the energy enterprise carbon emission data verification and management device involved in the above embodiments. The processing unit 901 is used to control and manage the operation of the energy enterprise carbon emission data verification and management device, the communication unit 902 is used for the energy enterprise carbon emission data verification and management device to communicate with other devices, and the storage unit 903 is used to store the program code and data of the energy enterprise carbon emission data verification and management device.

[0183] For example, communication unit 902 is used to acquire corporate carbon emission data; Processing unit 901 is used to preprocess enterprise carbon emission data and construct a carbon emission dataset; construct a directed acyclic graph (DAG) based on the carbon emission dataset; the nodes of the DAG include source nodes, computation nodes, and target nodes, and directed edges are established between nodes according to the dependency relationship of carbon emission accounting; the dependency relationship is the logical association relationship of carbon emission accounting, which is the transmission from source nodes to computation nodes, the hierarchical transmission between computation nodes, and the transmission from computation nodes to target nodes; using the MCMC algorithm, the posterior distribution of each node in the DAG is updated based on the historical prior distribution and the carbon emission dataset, and the confidence of each node is calculated; based on the confidence of each node, multi-level verification is performed to calculate the contribution of each source node to the uncertainty of the target node; the multi-level verification includes single-node self-verification, inter-node mutual verification, and global consistency verification; the source nodes are sorted according to the contribution and key abnormal nodes are identified; based on the key abnormal nodes, correction suggestion values ​​and correction confidence intervals are generated, correction suggestions are determined and pushed.

[0184] In one possible implementation, the processing unit 901 is also used to unify the data format and measurement unit of enterprise carbon emission data, and establish a standardized data dictionary; align the data time sequence of enterprise carbon emission data with the DCS system timestamp as a benchmark; use the 3σ criterion combined with empirical thresholds to remove abnormal data from enterprise carbon emission data; process missing values ​​according to the missing duration using linear interpolation, synchronous mean interpolation or LSTM neural network interpolation, integrate the preprocessed enterprise carbon emission data, label the processing status, and form a carbon emission dataset.

[0185] In one possible implementation, the processing unit 901 is further used to divide the data within the carbon emission dataset into three categories: source nodes, computation nodes, and target nodes. The source nodes are the original measurement data nodes for enterprise carbon emission accounting, including fuel consumption, coal quality test data, CEMS monitoring data, and unit operating data. The computation nodes are intermediate data nodes calculated based on the source node data, including daily fuel consumption, daily carbon emissions, and monthly cumulative carbon emissions. The target nodes are the total carbon emission data nodes finally reported by the enterprise. Based on the logical dependencies of carbon emission accounting, directed edges are established between nodes, and a directed acyclic graph is constructed based on the source nodes, computation nodes, target nodes, and directed edges. The direction of the directed edges satisfies the following: source node points to computation node, computation node points to computation node, and computation node points to target node.

[0186] In one possible implementation, the processing unit 901 is further configured to set the historical prior distribution of each node, wherein the source node adopts a normal distribution, the computation node adopts a log-normal distribution, and the target node adopts a Gamma distribution; the NUTS sampling method of the MCMC algorithm is used to iteratively update the posterior distribution of each node by combining the historical prior distribution with the carbon emission dataset; and the confidence of each node after the update is calculated.

[0187] In one possible implementation, the confidence level of the updated cutover point is calculated according to the following formula:

[0188] in, For the true value of node i, Let be the observation value at node i, δ be the allowable error range for node i, and D be the carbon emission dataset. Represents conditional probability. Let be the confidence level of node i.

[0189] In one possible implementation, the processing unit 901 is further configured to perform verification operations sequentially based on the confidence level of each node, in the following order: single-node self-verification, inter-node mutual verification, and global consistency verification. Single-node self-verification: The confidence level of each node is compared with a preset confidence level threshold and the node's historical confidence interval to determine if the node data passes verification, triggers a warning, or fails verification. Inter-node mutual verification: Based on the node dependencies in the directed acyclic graph, the consistency of data changes between logically related nodes is determined, the correlation coefficient of related nodes is calculated and compared with a preset correlation threshold to determine if the logic between nodes is abnormal. Global consistency verification: Weight coefficients are configured for each node according to node type, a weighted average method is used to calculate the global confidence level, and the global confidence level is compared with a preset global verification threshold to determine the quality level of the overall carbon emission dataset. Based on the confidence level of each node and the dependencies between nodes in the directed acyclic graph, the variance ratio method is used to calculate the contribution of each source node to the uncertainty of the target node.

[0190] In one possible implementation, the processing unit 901 is further configured to sort all source nodes in descending order of their contribution to the uncertainty of the target node, generating a source node contribution ranking table; combining the results of multi-level verification, selecting source nodes from the sorted source nodes that have been determined to have failed verification / verification warning by single-node self-verification and have been determined to be logically abnormal by inter-node mutual verification, forming an abnormal source node set; selecting the source nodes with the highest contribution ranking from the abnormal source node set as key abnormal nodes; the selection rule is to select the nodes with the highest contribution ranking, or to select the nodes whose contribution value exceeds a preset contribution threshold, where N is a positive integer, and the contribution threshold is set according to the enterprise's carbon emission accounting accuracy requirements and historical data characteristics.

[0191] In one possible implementation, the processing unit 901 is further configured to extract key abnormal node data and strongly dependent and associated trusted node data, remove anomalies, and establish a standardized correction benchmark dataset; predict reasonable values ​​for key abnormal nodes using the MCMC algorithm as correction suggestion values; calculate correction confidence intervals using the Bootstrap sampling statistical method based on node data characteristics; classify correction levels based on node influence weights to determine correction suggestions; and push the correction suggestion values, correction confidence intervals, and correction suggestions to specific users.

[0192] In one possible implementation, the contribution of each source node to the uncertainty of the target node is calculated using the variance ratio method, which satisfies the following formula:

[0193] in, The contribution of source node i. The true value of the target node. The true value of source node i. This is the variance calculation function. Given the true value of source node i, the conditional expectation of the target node.

[0194] The processing unit 901 can be a processor or a controller, and the communication unit 902 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 903 can be a memory. When the energy enterprise carbon emission data verification and management device 90 is a chip, the processing unit 901 can be a processor or a controller, and the communication unit 902 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 903 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.).

[0195] The communication unit can also be called a transceiver unit. The antenna and control circuit with transceiver functions in the energy enterprise carbon emission data verification and management device 90 can be considered as the communication unit 902 of the energy enterprise carbon emission data verification and management device 90, and the processor with processing functions can be considered as the processing unit 901 of the energy enterprise carbon emission data verification and management device 90. Optionally, the device in the communication unit 902 used to implement the receiving function can be considered as a communication unit, which is used to execute the receiving steps in the embodiments of this application. The communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 902 used to implement the transmitting function can be considered as a transmitting unit, which is used to execute the transmitting steps in the embodiments of this application. The transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0196] Figure 9 If the integrated units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0197] Figure 9 The units in the process can also be called modules; for example, a processing unit can be called a processing module.

[0198] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0199] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0200] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A method for verifying and managing carbon emission data of energy enterprises, characterized in that, include: Obtain corporate carbon emission data; The carbon emission data of the enterprises are preprocessed to construct a carbon emission dataset; A directed acyclic graph is constructed based on the carbon emission dataset; the nodes of the directed acyclic graph include source nodes, computation nodes and target nodes, and directed edges are established between nodes based on the dependency relationship of carbon emission accounting. The dependency relationship is a logical association for carbon emission accounting, which is transmitted from the source node to the computing node, transmitted level by level between computing nodes, and transmitted from the computing node to the target node. Using the MCMC algorithm, the posterior distribution of each node in the directed acyclic graph is updated based on the historical prior distribution and the carbon emission dataset, and the confidence of each node is calculated. Based on the confidence level of each node, a multi-level verification is performed to calculate the contribution of each source node to the uncertainty of the target node; the multi-level verification includes single-node self-verification, inter-node mutual verification, and global consistency verification. Based on the contribution, each source node is sorted to identify key abnormal nodes; Based on the key anomaly nodes, a correction suggestion value and a correction confidence interval are generated, and the correction suggestion is determined and pushed.

2. The method according to claim 1, characterized in that, The preprocessing of the enterprise's carbon emission data to construct a carbon emission dataset includes: Unify the data format and units of measurement for the carbon emission data of the aforementioned enterprises, and establish a standardized data dictionary; Based on the DCS system timestamp, align the data time sequence of the enterprise's carbon emission data; Abnormal data in the enterprise's carbon emission data were removed using the 3σ criterion combined with empirical thresholds. Missing values ​​are processed using linear interpolation, synchronous mean interpolation, or LSTM neural network interpolation according to the duration of the missing values. The preprocessed enterprise carbon emission data are then integrated, and the processing status is labeled to form the carbon emission dataset.

3. The method according to claim 1, characterized in that, The construction of the directed acyclic graph based on the carbon emission dataset includes: The carbon emission dataset is divided into three categories: source nodes, calculation nodes, and target nodes. The source nodes are the original measurement data nodes for enterprise carbon emission accounting, including fuel consumption, coal quality test data, CEMS monitoring data, and unit operating data. The calculation nodes are intermediate data nodes calculated based on the source node data, including daily fuel consumption, daily carbon emissions, and monthly cumulative carbon emissions. The target nodes are the total carbon emission data nodes that enterprises ultimately report. Based on the logical dependencies of carbon emission accounting, directed edges are established between nodes, and the directed acyclic graph is constructed based on the source node, computing node, target node, and directed edges; the direction of the directed edges satisfies: source node points to computing node, computing node points to computing node, and computing node points to target node.

4. The method according to claim 1, characterized in that, The method utilizes the MCMC algorithm to update the posterior distribution of each node in the directed acyclic graph based on the historical prior distribution and the carbon emission dataset, and calculates the confidence score of each node, including: Define the historical prior distribution of each node, where the source node adopts a normal distribution, the computation node adopts a log-normal distribution, and the target node adopts a Gamma distribution; The NUTS sampling method using the MCMC algorithm combines historical prior distributions with carbon emission datasets to iteratively update the posterior distributions of each node. Calculate the confidence level of each node after the update.

5. The method according to claim 4, characterized in that, The confidence level of the updated cutover point satisfies the following formula: in, For the true value of node i, Let be the observation value at node i, δ be the allowable error range for node i, and D be the carbon emission dataset. Represents conditional probability. Let be the confidence level of node i.

6. The method according to claim 1, characterized in that, The multi-level verification based on the confidence level of each node, calculating the contribution of each source node to the uncertainty of the target node, includes: Based on the confidence level of each node, the verification operations are performed sequentially according to the hierarchy of single-node self-verification, inter-node mutual verification, and global consistency verification: Single-node self-verification: The confidence level of each node is compared with the preset confidence level threshold and the historical confidence interval of the node to determine whether the node data passes the verification, issues a verification warning, or fails the verification. Inter-node mutual verification: Based on the node dependency relationship in the directed acyclic graph, the consistency of data changes between logically related nodes is judged, the correlation coefficient of related nodes is calculated and compared with the preset correlation threshold to determine whether the logic between nodes is abnormal. Global consistency verification: Configure weight coefficients for each node according to node type, calculate the global confidence level using the weighted average method, compare the global confidence level with the preset global verification threshold, and determine the quality level of the overall carbon emission dataset. Based on the confidence level of each node and the dependencies between nodes in the directed acyclic graph, the variance ratio method is used to calculate the contribution of each source node to the uncertainty of the target node.

7. The method according to claim 1, characterized in that, The process of sorting source nodes based on their contribution to determine key abnormal nodes includes: Sort all source nodes in descending order of their contribution to the uncertainty of the target node, and generate a source node contribution ranking table. Based on the results of multi-level verification, source nodes that are determined to be verification failures / verification warnings by single-node self-verification and to be logically abnormal by inter-node mutual verification are selected from the sorted source nodes, forming a set of abnormal source nodes; From the set of abnormal source nodes, the source nodes with the highest contribution ranking are selected as key abnormal nodes; the selection rule is to select the N nodes with the highest contribution ranking, or to select the nodes whose contribution value exceeds the preset contribution threshold, where N is a positive integer, and the contribution threshold is set according to the enterprise's carbon emission accounting accuracy requirements and historical data characteristics.

8. The method according to claim 1, characterized in that, The process of generating suggested correction values ​​and correction confidence intervals based on the key anomaly nodes, determining correction suggestions, and pushing them includes: Extract key abnormal node data and strongly dependent and trusted node data, remove anomalies, and establish a standardized and corrected benchmark dataset. The reasonable values ​​of key abnormal nodes are predicted using the MCMC algorithm, and these values ​​are used as suggested correction values. Based on the characteristics of node data, the Bootstrap sampling statistical method is used to calculate the corrected confidence interval; The correction level is determined by combining the influence weight of nodes, and correction recommendations are made accordingly. The suggested correction value, the confidence interval for correction, and the suggested correction are pushed to a targeted audience.

9. The method according to claim 6, characterized in that, The contribution of each source node to the uncertainty of the target node calculated using the variance ratio method satisfies the following formula: in, The contribution of source node i. The true value of the target node. The true value of source node i. This is the variance calculation function. Given the true value of source node i, the conditional expectation of the target node.

10. A device for verifying and managing carbon emission data of energy enterprises, characterized in that, The device includes: a communication unit and a processing unit; The communication unit is used to acquire enterprise carbon emission data; The processing unit is used to preprocess the enterprise carbon emission data to construct a carbon emission dataset; construct a directed acyclic graph (DAG) based on the carbon emission dataset; the nodes of the DAG include source nodes, computation nodes, and target nodes, and directed edges are established between nodes according to the dependency relationship of carbon emission accounting; the dependency relationship is a logical association relationship of carbon emission accounting that is transmitted from source nodes to computation nodes, transmitted level by level between computation nodes, and transmitted from computation nodes to target nodes; using the MCMC algorithm, the posterior distribution of each node in the DAG is updated based on the historical prior distribution and the carbon emission dataset, and the confidence score of each node is calculated; based on the confidence scores of each node, multi-level verification is performed to calculate the contribution of each source node to the uncertainty of the target node; the multi-level verification includes single-node self-verification, inter-node mutual verification, and global consistency verification; the source nodes are sorted based on the contribution scores to determine key abnormal nodes; based on the key abnormal nodes, correction suggestion values ​​and correction confidence intervals are generated, correction suggestions are determined, and pushed out.

Images