Enterprise carbon emission analysis method and system based on ESG comprehensive evaluation model

Abstract

The invention relates to the technical field of enterprise carbon emission analysis, and discloses an enterprise carbon emission analysis method and system based on an ESG comprehensive evaluation model, and the system comprises a data collection layer, an analysis processing layer and a decision output layer. The data acquisition layer acquires carbon emission related data in the whole production and operation process of an enterprise in real time through multiple devices; the analysis processing layer constructs a credible data processing environment, and realizes data cleaning verification, secure transmission, classified storage and dynamic synchronization of data and a model; and the decision output layer establishes a multi-dimensional associated ESG comprehensive evaluation model by using a mode recognition technology and an integrated learning algorithm, and realizes carbon emission intensity calculation, supply chain carbon footprint tracing and emission reduction path optimization decision. The method and the system can comprehensively and accurately analyze the carbon emission condition of the enterprise, provide scientific decision support for low-carbon management of the enterprise, and assist in realizing a carbon neutralization target.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise carbon emission analysis, and particularly to a method and system for enterprise carbon emission analysis based on an ESG comprehensive evaluation model. Background Art

[0002] In the context of the global active promotion of carbon neutrality and carbon peak, enterprises, as the main entities of carbon emissions, are facing increasingly strict environmental supervision requirements and the responsibility of social low-carbon development. Traditional enterprise carbon emission analysis methods have many limitations and are difficult to meet the current complex carbon emission management needs.

[0003] From the perspective of data collection, traditional methods often rely on manual input or single-device monitoring, with limited data coverage and unable to achieve real-time and comprehensive monitoring of the entire process of enterprise production and operation. For example, only partial data on direct emissions of enterprises can be obtained, and there is a lack of effective means to collect indirect carbon emission information in all links of the supply chain, resulting in incomplete carbon emission analysis. At the same time, manual data collection is prone to errors, and data updates are lagging, making it difficult to meet the needs of real-time analysis and decision-making.

[0004] In terms of data processing and storage, most traditional systems adopt centralized storage and simple data processing methods. Centralized storage has the risk of single-point failure, with low data security and reliability. Moreover, for multi-source heterogeneous carbon emission-related data, traditional processing methods lack effective cleaning, verification, and integration capabilities, unable to build a credible data processing environment, resulting in uneven data quality and affecting the accuracy of subsequent analysis. In addition, the security of data transmission cannot be effectively guaranteed, and problems such as data leakage are likely to occur.

[0005] In the construction of evaluation models, most existing carbon emission evaluation models are based on a single dimension or a fixed index system and lack the ability of dynamic adjustment. The production and operation activities of enterprises are dynamically changing, and factors such as energy consumption and production processes at different stages will affect the carbon emission situation. Traditional models cannot calibrate the weights of evaluation indicators in real time according to the actual operation data of enterprises and are difficult to accurately reflect the true carbon emission status of enterprises. At the same time, there is a lack of effective integration of industry benchmark data and best practice cases, resulting in the evaluation results of the models lacking comparability and reference.

[0006] In terms of decision-making support, traditional methods mainly rely on empirical judgment or simple statistical analysis and cannot provide scientific and accurate decision-making on optimizing emission reduction paths. For the traceability of the carbon footprint of the supply chain, it is difficult to conduct a full-chain analysis of the carbon emission contribution degree and accurately identify key carbon emission nodes. When formulating emission reduction strategies, there is a lack of comprehensive consideration of various information such as the energy efficiency of enterprise equipment, process improvement plans, application of clean technologies, and the carbon trading market, making it difficult to find the optimal balance between carbon emission constraints and economic cost limitations.

[0007] With the wide spread of the ESG (Environmental, Social, Governance) concept, enterprises not only need to pay attention to their own carbon emissions, but also need to consider the environmental responsibilities and governance levels of the supply chain. Traditional carbon emission analysis methods cannot meet the requirements of ESG comprehensive assessment. There is an urgent need for a new method and system that can integrate multi-dimensional data, build a dynamic assessment model, achieve full-chain carbon emission analysis and scientific decision-making. Summary of the Invention

[0008] The purpose of the present invention is to provide an enterprise carbon emission analysis method and system based on an ESG comprehensive assessment model to solve the problems raised in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solution: An enterprise carbon emission analysis system based on an ESG comprehensive assessment model, the system includes: A data acquisition layer, an analysis and processing layer, and a decision-making and output layer; The data acquisition layer includes environmental monitoring devices, energy metering instruments, a supply chain tracking module, and a compliance record unit, and is used for real-time monitoring and raw data acquisition of carbon emission-related elements in the whole process of enterprise production and operation; The analysis and processing layer includes edge computing nodes, a distributed storage cluster, and a blockchain verification module, and deploys a dynamic optimization algorithm and a hybrid encryption protocol, and is used for building a trusted data processing environment, performing data cleaning and verification operations, using a hybrid encryption protocol for secure transmission and classified storage for different business modules and data types, and realizing dynamic synchronization of enterprise operation data and the ESG comprehensive assessment model through a standardized interface; The decision-making and output layer is used for integrating enterprise historical operation data and real-time monitoring data by using pattern recognition technology, dynamically calibrating the evaluation index system, establishing an ESG comprehensive assessment model with multi-dimensional associations, and calculating carbon emission intensity, tracing the carbon footprint of the supply chain, and optimizing the emission reduction path decision based on this model by using an ensemble learning algorithm.

[0010] Preferably, the environmental monitoring devices at least include waste gas emission sensors, waste water detection probes, solid waste metering devices, and noise monitors, and are used for obtaining environmental parameters at the enterprise production and operation site; the energy metering instruments at least include electric energy meters, gas flow meters, steam meters, and fuel consumption recorders, and are used for counting enterprise energy consumption data; the supply chain tracking module at least includes logistics information collection terminals, raw material traceability databases, and transport vehicle positioning devices, and is used for recording carbon emission information of each link of the supply chain.

[0011] Preferably, in the analysis and processing layer, the data obtained by each collection terminal is transmitted by the edge computing node to the distributed storage cluster through a dedicated channel. After the data format conversion by the distributed storage cluster, it is jointly stored in the blockchain verification module with the supervision information obtained by the compliance record unit for feature extraction. The data cleaning and verification operation uses outlier detection technology, and the collected raw data is input into the verification models established by different verification rules for cross-verification. The dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model is realized through a standardized interface, including: establishing a standardized data interface to realize the interactive communication between the business system and the ESG comprehensive evaluation model, synchronizing the actual operation status data of the enterprise, performing associated mapping on the structured data, and simultaneously updating the model parameter configuration to complete data version management, model iteration update, and decision instruction verification.

[0012] Preferably, the establishment of the ESG comprehensive evaluation model with multi-dimensional associations includes: Constructing a two-way association channel and a dynamic matching mechanism between the actual operation data of the enterprise and the parameters of the ESG comprehensive evaluation model; By means of data mapping and feature association of the enterprise's production and operation activities, based on the enterprise's on-site monitoring data, supply chain record data, and compliance information, establishing an enterprise carbon emission inventory and an industry benchmark comparison framework, and initializing the model parameters according to the business characteristics, simulating the operation status of multiple scenarios, to form a dynamically adjustable enterprise ESG comprehensive evaluation model; Optimizing the parameters of the enterprise ESG comprehensive evaluation model, importing the multi-dimensional operation data of the enterprise into the established ESG comprehensive evaluation model, and using the Monte Carlo simulation method to perform sensitivity analysis and correction on the model output results to obtain an optimized enterprise ESG comprehensive evaluation model; Among them, the enterprise ESG comprehensive evaluation model includes operation entity data, an evaluation rule engine, a dynamic knowledge base, and the interaction relationships between various modules; The operation entity data is the basic data source of the ESG comprehensive evaluation model, including the enterprise's direct emission data and indirect emission data; the evaluation rule engine establishes a mapping relationship with the actual business data, and quantitatively evaluates the carbon emission characteristics through a multi-dimensional index system and a weight distribution mechanism; the dynamic knowledge base integrates industry standard data, policy and regulation texts, and best practice cases, and forms an evaluation benchmark database through continuous update and maintenance; the interaction relationships between various modules realize information interconnection through a data pipeline, and the operation data and the evaluation rule engine are standardized through a data conversion interface, and the evaluation rule engine and the dynamic knowledge base realize rule matching through semantic analysis technology.

[0013] Preferably, the calculation of the carbon emission intensity using an integrated learning algorithm based on this model includes: Collect the energy consumption data, production process record data, equipment operation logs, and product carbon footprint reports of the enterprise in previous years to construct a carbon emission characteristic data set; After filling in missing values and correcting outliers in the characteristic data set, divide it into a training set and a validation set; Configure the initial parameters of the XGBoost-GRU joint model, input the training set into the joint model, sort the feature importance through the XGBoost component, perform time series modeling through the GRU network, and eliminate the impact of emission fluctuations in different production cycles through the feature weighting mechanism until the model error reaches the preset threshold; Input the validation set into the trained joint model, calculate the predicted value of carbon emission intensity, and generate an evaluation result of the enterprise carbon emission intensity level in combination with the industry benchmark value; Match the preset improvement strategy library according to the evaluation result to generate targeted carbon emission management suggestions.

[0014] Preferably, the supply chain carbon footprint tracing using the ensemble learning algorithm based on this model includes: Obtain the logistics transportation data, raw material procurement records, production and processing information, and product distribution paths of each link in the supply chain to construct a supply chain carbon flow data set; Extract the key features of the supply chain carbon flow through the principal component analysis method to establish a carbon footprint tracing feature vector; Construct a random forest regression model and optimize the model parameter combination using the adaptive particle swarm algorithm; Input the feature vector into the optimized random forest regression model and output the ranking of the carbon emission contribution degrees of each link in the supply chain; Identify the key carbon footprint nodes according to the contribution degree ranking result and generate an optimized path map of the supply chain carbon flow.

[0015] Preferably, the emission reduction path optimization decision-making using the ensemble learning algorithm based on this model includes: Integrate the enterprise equipment energy efficiency data, process improvement plan library, clean technology application cases, and carbon trading market information to construct an emission reduction strategy knowledge graph; Extract the node relationships in the knowledge graph through graph convolutional network features to generate an emission reduction plan feature matrix; Establish a multi-objective optimization model and solve the Pareto optimal solution set using the NSGA-II algorithm; Input the feature matrix into the optimization model and output a set of feasible solutions that meet the carbon emission constraints and economic cost limitations; Select the final implementation plan according to the decision-making preference to generate a phased emission reduction route planning table.

[0016] Preferably, the specific implementation method of the data cleaning and verification operation in the analysis and processing layer includes: Build a data quality inspection rule library, including numerical range verification rules, data integrity verification rules, and logical consistency verification rules; Use a streaming computing engine to verify real-time data item by item, and trigger an early warning mechanism for data that does not conform to the verification rules; Perform a full scan of historical batch data, and use a data repair algorithm to automatically correct repairable abnormal records; Record the data verification process and exception handling logs through blockchain smart contracts, and generate an immutable data quality audit report.

[0017] Preferably, the construction method of the evaluation rule engine includes: Collect domestic and foreign carbon emission accounting standards, industry technical specifications, and policy and regulatory requirements, and build a knowledge base of standard clauses; Use natural language processing technology to parse standard documents, and extract key clauses to form structured evaluation rules; Build a rule matching engine, and realize the automatic association of enterprise actual data with standard clauses through semantic similarity calculation; Set up a dynamic adjustment mechanism for rule weights, and automatically optimize rule priorities according to the policy update frequency and industry practice changes.

[0018] Preferably, the present invention further includes an enterprise carbon emission analysis method based on an ESG comprehensive evaluation model, and the method includes the following steps: Real-time collect waste gas emission concentration, wastewater pollutant indicators, solid waste generation amount, and noise level data through environmental monitoring devices deployed at the production and operation site, and at the same time use energy metering instruments to obtain the consumption of electric energy, gas, steam, and fuel; After filtering out outliers and standardizing the format of the collected raw data through edge computing nodes, transmit it to a distributed storage cluster for multi-source data fusion, and use a blockchain verification module to perform encrypted verification on data integrity and time series consistency; Based on pattern recognition technology, perform feature correlation analysis on historical operation data and real-time monitoring data, dynamically adjust the weight parameters of ESG evaluation indicators, and build a multi-dimensional ESG comprehensive evaluation model including environmental compliance, energy utilization efficiency, and supply chain carbon footprint; Use a transfer learning framework to perform cross-industry knowledge transfer on the ESG comprehensive evaluation model, and generate carbon emission intensity prediction data under different emission reduction scenarios through Monte Carlo simulation; Combined with the logistics path, raw material source, and production and processing information recorded by the supply chain tracking module, establish a full-chain carbon flow mapping relationship, and use a graph neural network to identify key carbon emission nodes; According to the prediction results of carbon emission intensity and the carbon flow mapping relationship, a multi-objective optimization algorithm is used to solve the set of emission reduction paths that meet economic constraints, and a decision-making report including equipment transformation plans, process optimization strategies, and carbon trading suggestions is output.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: In terms of data collection and processing, the system realizes the real-time monitoring and original data collection of carbon emission-related elements in the whole process of enterprise production and operation through multi-source devices such as environmental monitoring devices, energy metering instruments, supply chain tracking modules, and compliance record units. The environmental monitoring device covers various devices such as waste gas emission sensors and wastewater detection probes, which can comprehensively obtain environmental parameters; the energy metering instrument can accurately count various energy consumption data; the supply chain tracking module can record the carbon emission information of each link in the supply chain to ensure the comprehensiveness and real-time nature of data collection. The analysis and processing layer constructs a trusted data processing environment through edge computing nodes, distributed storage clusters, and blockchain verification modules, uses outlier detection technology for data cleaning and verification, ensures data transmission and storage security through a hybrid encryption protocol, and realizes the dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model through a standardized interface, effectively improving data quality and processing efficiency, and providing a reliable data basis for subsequent analysis.

[0020] In terms of the construction of the ESG comprehensive evaluation model, the system establishes a multi-dimensional associated ESG comprehensive evaluation model. By constructing a two-way association channel and a dynamic matching mechanism between the actual operation data of the enterprise and the parameters of the ESG comprehensive evaluation model, based on the on-site monitoring data, supply chain record data, and compliance information of the enterprise, a carbon emission inventory and an industry benchmark comparison framework are established, and the model parameters are initialized and simulated in multiple scenarios to form an ESG comprehensive evaluation model that can be dynamically adjusted. At the same time, the Monte Carlo simulation method is used to perform sensitivity analysis and correction on the model output results to ensure the accuracy and adaptability of the model. This model integrates operation entity data, an evaluation rule engine, a dynamic knowledge base, and the interaction relationships between modules, and can comprehensively and dynamically evaluate the carbon emission status of the enterprise, meeting the requirements of ESG comprehensive evaluation.

[0021] In terms of data analysis and decision support, the system uses a variety of integrated learning algorithms based on the ESG comprehensive evaluation model to achieve carbon emission intensity calculation, supply chain carbon footprint tracing, and emission reduction path optimization decision-making. Through the XGBoost-GRU joint model for carbon emission intensity calculation, it can effectively eliminate the impact of emission fluctuations in different production cycles, generate accurate carbon emission intensity level evaluation results and targeted management suggestions; use the random forest regression model and adaptive particle swarm algorithm for supply chain carbon footprint tracing, which can identify key carbon footprint nodes and generate an optimization path map; by constructing an emission reduction strategy knowledge graph and a multi-objective optimization model, and using the NSGA-II algorithm to solve the Pareto optimal solution set, it can output a set of feasible solutions that meet carbon emission constraints and economic cost limitations, generate a phased emission reduction route planning table, and provide scientific and accurate decision support for enterprises.

[0022] In terms of data security and management, the analysis and processing layer records the data verification process and exception handling logs through blockchain smart contracts, generates an immutable data quality audit report, and ensures the integrity and credibility of the data. The evaluation rule engine parses standard documents through natural language processing technology, establishes a rule matching engine and a weight dynamic adjustment mechanism, and can respond in a timely manner to policy updates and changes in industry practices, ensuring the scientific nature and timeliness of the evaluation rules. Brief Description of the Drawings

[0023] Figure 1 It is the working principle diagram of the enterprise carbon emission analysis system based on the ESG comprehensive evaluation model of the present invention; Figure 2 It is the working principle diagram of data transmission and verification in the analysis and processing layer; Figure 3 It is the schematic diagram of the construction of the multi-dimensional associated ESG comprehensive evaluation model; Figure 4 It is the working principle diagram of supply chain carbon footprint tracing and contribution analysis. Detailed Embodiments

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0025] Please refer to Figures 1-4 , an enterprise carbon emission analysis system based on the ESG comprehensive evaluation model involved in the present invention, the system includes: a data collection layer, an analysis and processing layer, and a decision output layer, and each layer works together to achieve a comprehensive analysis and management of enterprise carbon emissions. The specific implementation steps are as follows: The data acquisition layer uses environmental monitoring devices, energy metering instruments, supply chain tracking modules, and compliance record units to monitor carbon emission-related elements in the entire process of an enterprise's production and operation in real time and collect raw data. The environmental monitoring devices obtain the environmental parameters at the enterprise's production and operation site, the energy metering instruments count the enterprise's energy consumption data, the supply chain tracking modules record the carbon emission information of each link in the supply chain, and the compliance record units collect the enterprise's compliance information.

[0026] The analysis and processing layer uses edge computing nodes, distributed storage clusters, and blockchain verification modules to build a trusted data processing environment. The edge computing nodes perform preliminary processing on the data obtained by the acquisition terminals and transmit it to the distributed storage cluster through a dedicated channel. After the distributed storage cluster performs data format conversion, it stores the data together with the regulatory information obtained by the compliance record units in the blockchain verification module for feature extraction. This layer also deploys dynamic optimization algorithms and hybrid encryption protocols to perform data cleaning and verification operations, uses hybrid encryption protocols for secure transmission and classified storage for different business modules and data types, and realizes the dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model through standardized interfaces.

[0027] Refer to the appendix Figure 2 For the data processing flow of the data obtained by each acquisition terminal in the analysis and processing layer, it first involves the process of data transmission from the edge computing nodes to the distributed storage cluster through a dedicated channel. As the front end of data processing, the edge computing nodes are responsible for performing preliminary processing on the raw data obtained by the acquisition terminals. This preliminary processing includes data format conversion, data cleaning and verification operations using outlier detection techniques, and inputting the collected raw data into the verification models established by different verification rules for cross-verification.

[0028] In terms of data transmission, the establishment of a dedicated channel ensures the security and stability of data transmission. This channel uses multi-layer encryption technology to encrypt the transmitted data in real time to prevent the data from being stolen or tampered with during transmission. At the same time, the channel also has a data transmission monitoring function, which can monitor the status of data transmission in real time. Once an abnormal situation is found, it can take timely measures to handle it to ensure the secure transmission of data.

[0029] After receiving the data transmitted by the edge computing nodes, the distributed storage cluster will perform further format conversion on the data. This process is to ensure that the data can be recognized and processed by subsequent processing modules. The distributed storage cluster uses a distributed file system, which can store data dispersedly on multiple nodes, improving the storage reliability and availability of data. At the same time, the distributed storage cluster also has a data redundancy backup function, which can back up important data multiple times to prevent data loss.

[0030] The regulatory information obtained by the data and compliance record unit is jointly stored in the blockchain verification module for feature extraction. The blockchain verification module adopts blockchain technology, which can ensure the immutability and traceability of data. Before the data is stored in the blockchain verification module, it will be hashed to generate a unique hash value. Then, the hash value is stored in the blockchain to ensure the integrity and authenticity of the data. When performing feature extraction, the blockchain verification module will conduct multi-dimensional analysis and processing on the data to extract the key features and information in the data.

[0031] The data cleaning and verification operation adopts outlier detection technology. The collected raw data is input into the verification models established by different verification rules for cross-verification. This process is to ensure the accuracy and reliability of the data. The outlier detection technology adopts a variety of algorithms, including statistical-based methods, machine learning-based methods, etc., which can effectively detect outliers in the data. In terms of the establishment of the verification model, corresponding verification models will be established according to different data types and business requirements. These verification models include numerical range verification rules, data integrity verification rules, and logical consistency verification rules, etc., which can conduct comprehensive verification and validation on the data.

[0032] The dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model is achieved through a standardized interface. Specifically, a standardized data interface is established to realize the interactive communication between the business system and the ESG comprehensive evaluation model, synchronize the actual business status data of the enterprise, conduct associated mapping on the structured data, and at the same time update the model parameter configuration to complete data version management, model iteration update, and decision instruction verification. The establishment of the standardized interface is the key to realizing the dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model. This interface adopts unified standards and specifications, which can ensure data interaction and sharing between different systems. In terms of data synchronization, the actual business status data of the enterprise will be obtained in real time and transmitted to the ESG comprehensive evaluation model for analysis and processing. At the same time, associated mapping will also be conducted on the structured data to ensure the consistency and accuracy of the data.

[0033] In terms of the update of model parameter configuration, the parameter configuration of the ESG comprehensive evaluation model will be updated in real time according to the actual business situation of the enterprise and market changes. This process can ensure the accuracy and reliability of the ESG comprehensive evaluation model and improve the decision-making level of the enterprise. Data version management is an important link to achieve the dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model. Through data version management, the historical versions of data can be recorded and managed to ensure the traceability and consistency of data. In terms of model iteration and update, the ESG comprehensive evaluation model will be continuously iterated and updated according to the actual needs of the enterprise and market changes. This process can ensure the advancement and practicality of the ESG comprehensive evaluation model and improve the competitiveness of the enterprise. Decision instruction verification is the last line of defense to achieve the dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model. Through decision instruction verification, the decision instructions output by the ESG comprehensive evaluation model can be verified and audited to ensure the rationality and feasibility of the decision instructions.

[0034] The specific implementation method of data cleaning and verification operations is to establish a data quality detection rule library, which includes numerical range verification rules, data integrity verification rules, and logical consistency verification rules; use a streaming computing engine to verify real-time data item by item, and trigger an early warning mechanism for data that does not meet the verification rules; perform a full scan of historical batch data and use a data repair algorithm to automatically correct repairable abnormal records; record the data verification process and exception handling logs through blockchain smart contracts to generate an immutable data quality audit report.

[0035] The establishment of a data quality detection rule library is the basis of data cleaning and verification operations. The rule library contains various data quality detection rules, which can comprehensively detect and verify data. The numerical range verification rule can detect and verify the numerical range of data to ensure that the data values are within a reasonable range. The data integrity verification rule can detect and verify the integrity of data to ensure that the data is not missing or duplicated. The logical consistency verification rule can detect and verify the logical consistency of data to ensure that the logical relationships between data are correct.

[0036] The adoption of a streaming computing engine is the key to achieving item-by-item verification of real-time data. The engine can perform real-time processing and analysis of real-time data and trigger an early warning mechanism for data that does not meet the verification rules. The early warning mechanism can promptly notify relevant personnel to process abnormal data to ensure the accuracy and reliability of the data.

[0037] Perform a full - scale scan of historical batch data and use a data repair algorithm to automatically correct repairable abnormal records. This process can comprehensively clean and repair historical data, improving data quality and usability. The data repair algorithm can automatically select appropriate repair methods based on data characteristics and abnormal conditions to ensure data accuracy and reliability.

[0038] Record the data verification process and exception handling logs through blockchain smart contracts to generate an immutable data quality audit report. Blockchain smart contracts can ensure the immutability and traceability of the data verification process and exception handling logs. The data quality audit report can comprehensively evaluate and analyze data quality, providing strong support for an enterprise's data management and decision - making.

[0039] The decision - making output layer uses pattern recognition technology to integrate an enterprise's historical operation data and real - time monitoring data. After dynamically calibrating the evaluation index system, a multi - dimensional associated ESG comprehensive evaluation model is established. Based on this model, an integrated learning algorithm is used to calculate carbon emission intensity, trace the carbon footprint of the supply chain, and optimize decision - making on emission reduction paths, providing scientific carbon emission management decisions for the enterprise.

[0040] Refer to the appendix Figure 3 To establish a multi - dimensional associated ESG comprehensive evaluation model, it is necessary to go through links such as the construction of a two - way association channel, the establishment of a multi - dimensional model, parameter optimization, and model structure construction. First, it is necessary to construct a two - way association channel and a dynamic matching mechanism between the enterprise's actual operation data and the parameters of the ESG comprehensive evaluation model. The establishment of the two - way association channel is based on the enterprise's internal data transmission network. Through standardized data interfaces, it realizes the real - time transmission of operation data from business systems such as production, energy, and supply chain to the ESG comprehensive evaluation model. At the same time, it allows model parameters to act on business systems in reverse, forming an interactive closed - loop between data and the model. The dynamic matching mechanism automatically matches key indicators in operation data (such as energy consumption, waste gas emissions, supply chain node information, etc.) with model parameters (such as emission factors, weight coefficients, industry benchmark values, etc.) through preset data mapping rules, ensuring that the correspondence between data and parameters is dynamically adjusted according to changes in business scenarios.

[0041] The enterprise's production and operation activities are used to establish an enterprise carbon emission inventory and an industry benchmark comparison framework based on the enterprise's on-site monitoring data, supply chain record data, and compliance information through data mapping and feature association. The model parameters are initialized according to business characteristics, and the operation status of multiple scenarios is simulated to form a dynamically adjustable enterprise ESG comprehensive evaluation model. During the data mapping process, on-site monitoring data (such as concentration data collected by exhaust gas emission sensors and energy consumption data recorded by energy metering instruments) is preprocessed through edge computing nodes and mapped to the environmental dimension parameters of the model in a unified data format; supply chain record data (such as logistics transportation routes and raw material procurement sources) is verified by the blockchain verification module and then mapped to the supply chain dimension parameters of the model; compliance information (such as environmental policy compliance records and carbon emission quota implementation) is accessed through the compliance record unit and mapped to the governance dimension parameters of the model. Feature association uses the correlation analysis algorithm in machine learning to identify potential connections between data in different dimensions, such as the linear relationship between energy consumption and exhaust gas emissions, and the positive correlation between supply chain transportation distance and carbon footprint, to construct a multi-dimensional data feature network.

[0042] The establishment of the enterprise carbon emission inventory is based on the principle of full-life-cycle carbon emission accounting, covering the enterprise's direct emissions (such as emissions generated by the combustion of fossil fuels during the production process) and indirect emissions (such as emissions generated by raw material production in the upstream supply chain and product transportation in the downstream supply chain). Through data aggregation, an inventory table containing the total carbon emissions and intensity of each link is formed. The industry benchmark comparison framework is based on the carbon emission benchmark data released by industry associations and the public data of peer benchmarking enterprises in the same industry. A horizontal comparison coordinate system is established to compare the indicators in the enterprise carbon emission inventory with the industry benchmark values to identify the enterprise's carbon emission level positioning in the industry. Model parameter initialization assigns initial values to the parameters of each dimension of the model according to business characteristics such as the enterprise's industry characteristics, production scale, and process level. For example, the initial value of the energy consumption emission factor in high-energy-consuming industries is set at a relatively high level, while that in clean energy industries is set at a relatively low level. Multi-scenario simulation adjusts the model parameters (such as assuming scenarios of optimized energy structure and changed supply chain paths), runs the model, and outputs the carbon emission prediction results under different scenarios to form a scenario analysis report, providing a basis for the dynamic adjustment of the model.

[0043] When optimizing the parameters of the enterprise ESG comprehensive evaluation model, import the multi-dimensional operation data of the enterprise into the established ESG comprehensive evaluation model, and use the Monte Carlo simulation method to conduct sensitivity analysis and correction on the model output results to obtain the optimized enterprise ESG comprehensive evaluation model. The multi-dimensional operation data includes the energy consumption ledger for the past three years, carbon flow records for each link in the supply chain, environmental compliance inspection reports, etc., and is batch-imported into the model through the interface of the distributed storage cluster. The Monte Carlo simulation generates a large number of random number sequences to simulate the random changes of data parameters, calculates the probability distribution of the model output results, and identifies the key parameters that have a significant impact on the model results (such as the energy consumption coefficient of a certain type of equipment, the carbon intensity of a specific transportation route). Sensitivity analysis determines the priority and amplitude of parameter adjustment by calculating the sensitivity index of each parameter. For example, key calibration is performed on parameters with a high sensitivity index, more accurate parameter values are obtained through historical data regression analysis, and then re-entered into the model for verification until the stability and accuracy of the model output results reach the expected standards.

[0044] The enterprise ESG comprehensive evaluation model includes operation entity data, an evaluation rule engine, a dynamic knowledge base, and the interaction relationships between each module. The operation entity data, as the basic data source of the ESG comprehensive evaluation model, includes the enterprise's direct emission data and indirect emission data. The direct emission data is collected in real time through environmental monitoring devices and energy metering instruments, and after being cleaned and verified by the analysis and processing layer, it is stored in the direct emission database of the distributed storage cluster; the indirect emission data is collected by the supply chain tracking module to collect information on each link in the supply chain, and is calculated in combination with the emission factors in the industry database and stored in the indirect emission database. The two types of data are associated through timestamps and business process numbers to form a complete enterprise carbon emission data chain.

[0045] The evaluation rule engine establishes a mapping relationship with the actual business data. Its construction method is to collect domestic and foreign carbon emission accounting standards, industry technical specifications, and policy and regulatory requirements to build a knowledge base of standard terms; use natural language processing technology to parse standard documents and extract key terms to form structured evaluation rules; establish a rule matching engine to automatically associate enterprise actual data with standard terms through semantic similarity calculation; set up a dynamic adjustment mechanism for rule weights to automatically optimize rule priorities according to the policy update frequency and industry practice changes. The knowledge base of standard terms covers international standards (such as ISO14064), national standards (such as GB / T32151), local standards, and industry association specifications. The latest policy documents are captured in real time through web crawler technology and stored in the knowledge base after manual review. Natural language processing technology performs word segmentation, part-of-speech tagging, named entity recognition, etc. on standard documents, extracts core terms (such as emission accounting boundaries, data quality requirements, report format specifications, etc.), and transforms them into structured evaluation rules, which are stored in the rule database. The rule matching engine calculates the semantic similarity between enterprise actual data (such as carbon emission accounting scope, data collection frequency) and evaluation rules through the cosine similarity algorithm. When the similarity exceeds the preset threshold, the rule association is automatically triggered to generate a compliance evaluation result. The dynamic adjustment mechanism for rule weights is based on the policy update log and industry practice case library. When the update frequency of a certain type of policy and regulation increases or a new emission reduction technology is widely adopted in the industry, the weight coefficient of the evaluation rule is adjusted accordingly. For example, the weight of rules related to clean energy utilization is increased, and the weight of rules corresponding to backward processes is reduced.

[0046] The dynamic knowledge base integrates industry standard data, policy and regulatory texts, and best practice cases, and forms an evaluation benchmark database through continuous update and maintenance. Industry standard data includes indicators such as industry average carbon emission intensity, energy efficiency level, and supply chain carbon efficiency, which are obtained through industry statistical annual reports and research reports of third-party institutions; policy and regulatory texts synchronize in real time the carbon emission-related laws, regulations, and policy documents issued by the state and localities; best practice cases collect the successful emission reduction experiences of domestic and foreign enterprises (such as the specific measures of a certain enterprise to reduce carbon emissions through equipment transformation, and the carbon footprint reduction effect of a certain supply chain optimization case), and are stored in the case library after expert review. The update mechanism of the dynamic knowledge base includes scheduled automatic crawling (such as updating policy documents at 0:00 every day), manual review and entry (such as summarizing industry data every month), and user feedback and supplementation (such as enterprises submitting their own emission reduction cases), to ensure the timeliness and practicality of the evaluation benchmark data.

[0047] The interaction relationships between modules achieve information interconnection through data pipelines. Standardized processing is carried out between operation data and the evaluation rule engine through a data conversion interface, and rule matching is achieved between the evaluation rule engine and the dynamic knowledge base through semantic parsing technology. The data pipeline adopts message queue technology to establish a transmission link for operation data from the acquisition layer to the analysis and processing layer and then to the decision-making output layer, ensuring the orderly flow of data between modules. The data conversion interface between operation data and the evaluation rule engine is responsible for converting the original operation data (such as analog signals collected by sensors and structured data in the database) into a format recognizable by the rule engine (such as rule input parameters in JSON format), and at the same time converting the evaluation results output by the rule engine into the data format required for visual reports. The semantic parsing technology between the evaluation rule engine and the dynamic knowledge base realizes the semantic matching of evaluation rules with standard terms and cases in the knowledge base through algorithms such as text classification and entity linking in natural language processing. For example, when the rule engine needs to query the emission reduction standards for a certain type of equipment, the semantic parsing technology automatically retrieves relevant industry standards and best practice cases in the dynamic knowledge base to provide a reference basis for rule evaluation.

[0048] The steps for calculating the carbon emission intensity using the integrated learning algorithm based on the ESG comprehensive evaluation model are as follows: First, collect the enterprise's historical annual energy consumption data, production process record data, equipment operation logs, and product carbon footprint reports to construct a carbon emission feature dataset. The energy consumption data covers detailed information such as the consumption quantity, consumption time, and consumption department of various types of energy (such as coal, electricity, natural gas, etc.) within the historical year of the enterprise, and is collected through the enterprise's internal energy metering instruments and energy management systems. The production process record data includes raw material input quantity, product output quantity, production process parameters, production equipment operation status, etc., and these data come from the enterprise's production management system and on-site monitoring equipment. The equipment operation logs record information such as the start time, stop time, operation load, and maintenance records of various production equipment, and can be obtained from the equipment control system and maintenance management system. The product carbon footprint report details the carbon emissions throughout the entire life cycle of the product from raw material procurement, production and processing, transportation and sales to final consumption, and is generated through the supply chain tracking module and life cycle assessment tool. Integrate these multi-source heterogeneous data, correlate them according to the time series and business processes, and construct a carbon emission feature dataset containing multiple feature dimensions.

[0049] After filling in missing values and correcting outliers in the feature dataset, the training set and validation set are divided. The missing value filling adopts the multiple imputation method. Based on the distribution characteristics and correlations of the data, other feature variables are used to estimate and fill in the missing values. The outlier correction is based on statistical analysis methods. The mean and standard deviation of the data are calculated, and data points that deviate from the mean by more than a certain multiple of the standard deviation are regarded as outliers, and are corrected by replacing with boundary values or based on model predicted values. When dividing the training set and validation set, the processed feature dataset is randomly divided according to the ratio of 8:2. The training set is used for model training and parameter optimization, and the validation set is used to evaluate the performance and generalization ability of the ESG comprehensive evaluation model.

[0050] Configure the initial parameters of the XGBoost-GRU joint model, input the training set into the joint model. Through the XGBoost component, the feature importance is sorted, through the GRU network, time series modeling is carried out, and the impact of emission fluctuations in different production cycles is eliminated through the feature weighting mechanism until the model error reaches the preset threshold. The initial parameter configuration of the XGBoost component includes the maximum depth of the tree, learning rate, subsample ratio, etc. The optimal parameter combination is determined by the methods of grid search and cross-validation. After the training set is input into the XGBoost component, the model will automatically calculate the importance score of each feature for carbon emission intensity, sort the features according to the score, and screen out the key features that have a greater impact on carbon emission intensity.

[0051] The GRU network is used to process time series data and capture the trends and patterns of carbon emission data changing over time. The initial parameter configuration of the GRU network includes the number of neurons in the hidden layer, time step, dropout rate, etc. The selected key features are input into the GRU network in time series format, and the network will automatically learn the temporal patterns and dependencies in the data. The feature weighting mechanism assigns different weights to the features at each time step according to the characteristics of different production cycles. During the peak production season, energy consumption and carbon emissions are usually high, and higher weights are assigned to the relevant features at this time; during the off-peak production season, lower weights are assigned. In this way, the impact of emission fluctuations in different production cycles on model prediction can be effectively eliminated. During the model training process, the backpropagation algorithm is used to continuously adjust the model parameters to minimize the error between the predicted value and the actual value until the error reaches the preset threshold.

[0052] Input the validation set into the trained joint model, calculate the predicted carbon emission intensity value, and generate the evaluation result of the enterprise carbon emission intensity level in combination with the industry benchmark value. After processing the data in the validation set in the same preprocessing manner as the training set, input it into the trained XGBoost-GRU joint model, and the model will output the predicted carbon emission intensity value of each sample. The industry benchmark value comes from the industry carbon emission intensity standard issued by the industry association and the carbon emission data of peer enterprises in the same industry. Compare the predicted carbon emission intensity value of the enterprise with the industry benchmark value to generate the evaluation result of the enterprise carbon emission intensity level. The evaluation results are divided into multiple levels, such as excellent, good, medium, poor, etc., and each level corresponds to a different carbon emission intensity interval.

[0053] Match the preset improvement strategy library according to the evaluation result to generate targeted carbon emission management suggestions. The improvement strategy library stores improvement strategies for different carbon emission intensity levels and different business scenarios. These strategies include energy structure optimization plans, production process improvement measures, equipment upgrade suggestions, supply chain management optimization strategies, etc. According to the evaluation result of the enterprise's carbon emission intensity level, match the corresponding improvement strategy from the improvement strategy library, and combine the actual business situation and resource constraints of the enterprise to generate specific and operable carbon emission management suggestions. These suggestions include short-term, medium-term, and long-term emission reduction targets and implementation paths, providing a scientific basis for the enterprise's carbon emission management decision-making.

[0054] In the process of calculating the carbon emission intensity, the prediction result of the XGBoost-GRU joint model can be expressed as: where represents the predicted carbon emission intensity value at time , represents the input feature vector at time , including key features such as energy consumption data and production process data, represents the set of model parameters, represents the mapping function of the XGBoost-GRU joint model, which converts the input feature vector into the predicted carbon emission intensity value through the feature processing of XGBoost and the time series modeling of GRU.

[0055] Refer to the appendix Figure 4, the process of tracing the carbon footprint of the supply chain using an ensemble learning algorithm based on the ESG comprehensive evaluation model is as follows: First, it is necessary to obtain the logistics transportation data, raw material procurement records, production and processing information, and product distribution paths of each link in the supply chain to construct a supply chain carbon flow dataset. The logistics transportation data includes the type of transportation vehicle (such as trucks, ships, airplanes), the mileage of the transportation route, the total fuel consumption, the carbon emission factor, etc., which are obtained in real time through the logistics information collection terminal and the transportation vehicle positioning device. For example, the GPS positioning system installed on a truck can record the driving route and mileage, and the on-vehicle fuel gauge can collect fuel consumption data, and the carbon emissions of the transportation link can be calculated by combining the carbon emission factor of diesel. The raw material procurement records cover the name of the raw material, the procurement quantity, the supplier information, the raw material production process, etc., which are obtained through the raw material traceability database. The supplier information includes the region where the supplier is located and the production scale, and the raw material production process information is used to evaluate the carbon emission intensity of the upstream production link. The production and processing information includes the production equipment model, the processing time, the energy consumption, the amount of auxiliary materials used, etc., which come from the equipment operation logs and energy metering instruments in the production workshop. For example, the total energy consumption of a production line can be obtained by multiplying the equipment operation time of a production line by the energy consumption per unit time, and then the carbon emissions of the production and processing link can be calculated by combining the carbon emission factor corresponding to the energy type. The product distribution path data includes the distribution channel (such as dealers, e-commerce platforms), the sales area, the transportation mode conversion nodes, etc., which are obtained through the product distribution management system and the logistics tracking platform, and are used to analyze the carbon emission distribution in the entire distribution process of the product from the factory to the hands of consumers.

[0056] Associate and integrate the above multi-source data according to the business process of the supply chain (procurement → production → transportation → distribution) in terms of time series and spatial nodes, eliminate duplicate records and invalid data, and unify the data format (such as using JSON or CSV format) to construct a supply chain carbon flow dataset containing fields such as carbon emissions, timestamps, node locations, and business types of each link in the supply chain. The dataset needs to ensure the integrity and accuracy of the data. For example, the missing transportation mileage data is filled with the average value of adjacent nodes, and the abnormal raw material procurement quantity data is corrected by checking with the supplier.

[0057] Extract the key features of the supply chain carbon flow through the principal component analysis method to establish a carbon footprint tracking feature vector. The principal component analysis method (PCA) is a dimensionality reduction technique used to convert high-dimensional supply chain carbon flow data into low-dimensional key features. First, standardize the dataset to eliminate the influence of different variable dimensions and value ranges. The calculation formula is: Among them, represents the th sample's th original eigenvalue, is the mean value of the th feature. The standard deviation of the th feature. After standardization, calculate the feature covariance matrix, solve the eigenvalues and eigenvectors of the covariance matrix, determine the number of principal components according to the cumulative contribution rate of the eigenvalues (usually select the principal components with a cumulative contribution rate exceeding 85%), and linearly combine the original features into new principal component features. For example, if the original features include 10 variables such as transportation mileage, fuel consumption, and raw material procurement volume, 3 principal components can be extracted through PCA, representing comprehensive features such as transportation intensity, production scale, and distribution complexity respectively. Arrange the principal component eigenvalues of each sample in order to form a carbon footprint tracking feature vector for subsequent model input.

[0058] Construct a random forest regression model and optimize the model parameter combination using the adaptive particle swarm optimization algorithm. The random forest regression model consists of multiple decision trees, and the base learners are constructed through bootstrap sampling and random feature selection, with strong anti-overfitting ability and good fitting effect for non-linear data. The initial parameters of the model include the number of decision trees ( ), the maximum number of features considered when splitting each node ( ), the maximum depth of the decision tree ( ), etc. The adaptive particle swarm optimization algorithm (APSO) is used to optimize the model parameters. The particle swarm optimization algorithm finds the optimal solution by simulating the foraging behavior of a bird flock. Each particle represents a set of parameter combinations, and iteratively updates the speed and position in the search space to find the parameter combination that minimizes the model loss function. The adaptive mechanism is reflected in dynamically adjusting the inertia weight and learning factor according to the fitness value of the particle. For example, when the particle approaches the optimal solution, reduce the inertia weight to improve the local search accuracy; when the particle falls into the local optimum, increase the learning factor to enhance the global search ability.

[0059] During the model training process, input the feature vector (independent variable) of the supply chain carbon flow dataset and the corresponding total carbon footprint (dependent variable) into the random forest model. Each decision tree makes an independent prediction on the sample, and finally takes the average of all decision tree prediction values as the model output. Continuously adjust the parameters through the APSO algorithm to minimize the mean squared error (MSE) of the model on the training set. The optimized parameter combinations are such as , , etc. Input the feature vector into the optimized random forest regression model to output the ranking of the carbon emission contribution degrees of each link in the supply chain. After the model training is completed, use the feature importance evaluation mechanism of the random forest to calculate the contribution degree of each principal component feature to the total carbon footprint. The feature importance is measured by calculating the total sum of the purity improvements caused by each feature in all decision trees. The higher the contribution degree of a feature, the greater the impact of that link on the overall carbon footprint of the supply chain. For example, if the principal component feature importance score corresponding to the transportation mileage is the highest, it indicates that the transportation link is the main contributor to the supply chain carbon footprint. Normalize the feature importance scores corresponding to each link (procurement, production, transportation, distribution), convert them into contribution degree values in percentage form, and sort them in descending order to form a list of the carbon emission contribution degrees of each link in the supply chain.

[0060] Identify the key carbon footprint nodes based on the ranking results of the contribution degrees and generate an optimized path map of the supply chain carbon flow. The identification criterion for the key nodes is the link or specific business node whose contribution degree exceeds a preset threshold (such as 20%). For example, if the contribution degree of the transportation link is 35%, then determine the transportation link as a key node, and further analyze the contribution degrees of specific sub-nodes such as the specific transportation routes and transportation tool types in this link, and identify the transportation route with the highest contribution degree (such as a certain long-distance truck transportation route) as the key sub-node. For the key nodes, formulate optimization strategies in combination with the actual business processes of the supply chain and industry best practices. For example, for a transportation route with a high contribution degree, the transportation path can be optimized to shorten the mileage, or it can be replaced with a new energy transportation tool to reduce the carbon emission factor.

[0061] When generating the optimized path map of the supply chain carbon flow, use the supply chain nodes as vertices and the carbon flow direction as edges, and use different colors and thicknesses to represent the contribution degrees and carbon flow volumes of each node. For example, represent the key nodes with red circles, represent the non-key nodes with blue circles, draw the edges with large carbon flow volumes with thick lines, and draw the edges with small carbon flow volumes with thin lines. Suggested optimization measures can also be marked in the optimized path map, such as marking "Replace with an electric truck" next to the transportation link node and marking "Upgrade energy-saving equipment" next to the production link node, etc., to provide a visual decision-making basis for enterprises to optimize the supply chain carbon footprint. The entire process is data-driven to achieve accurate traceability and scientific optimization of the supply chain carbon footprint, ensuring that enterprises can take effective emission reduction measures for high-carbon emission links and improve the environmental sustainability of the supply chain.

[0062] The process of optimizing the emission reduction path using an ensemble learning algorithm based on the ESG comprehensive evaluation model is as follows: First, it is necessary to integrate enterprise equipment energy efficiency data, process improvement solution libraries, clean technology application cases, and carbon trading market information to construct an emission reduction strategy knowledge graph. Enterprise equipment energy efficiency data covers the models, operating years, rated power, actual energy efficiency values, etc. of various production equipment, which are collected in real time through the equipment management system and energy metering instruments. For example, the actual energy efficiency value of a certain injection molding machine can be calculated by comparing its energy consumption per unit output with the rated energy consumption. The process improvement solution library stores the process optimization solutions accumulated by the enterprise over time, including injection molding process parameter adjustment, heat treatment process improvement, etc. Each solution contains information such as implementation steps, expected emission reduction effects, and cost inputs, which are jointly sorted and entered by the production department and the technology R & D department. Clean technology application cases collect the successful experiences of domestic and foreign enterprises in the same industry using clean technologies (such as solar photovoltaic power generation, wastewater recycling treatment systems), including technical principles, implementation cycles, investment returns, etc., which are obtained through channels such as industry reports and technical seminars and incorporated into the case library after expert review. Carbon trading market information synchronizes in real time the price trends, quota supply and demand situations, policy dynamics, etc. of the carbon emission rights trading platform, which are automatically captured through data interfaces and stored in the system database.

[0063] Structurally process the above multi-source heterogeneous information and transform it into the nodes and edges of the knowledge graph. The equipment models, energy efficiency values, etc. in the equipment energy efficiency data are used as entity nodes, the optimization measures, costs, expected effects, etc. in the process improvement solutions are used as attribute nodes, the technical types, applying enterprises, etc. in the clean technology cases are used as case nodes, and the prices, quotas, etc. of the carbon trading market are used as market nodes. The association relationships between the nodes are represented by edges. For example, the equipment node and the process improvement solution node are connected by an "applicable process" edge, the clean technology case node and the equipment node are connected by a "technical adaptation" edge, and the carbon trading market node and the enterprise's total carbon emissions node are connected by a "quota impact" edge. Visualize these nodes and edges through a knowledge graph construction tool (such as Neo4j) to form an emission reduction strategy knowledge graph containing multi-dimensional information such as enterprise internal resources, external technologies, and market environments.

[0064] Extract the features of the node relationships in the knowledge graph using a graph convolutional network to generate a feature matrix for emission reduction solutions. The graph convolutional network (GCN) is a neural network specifically designed for processing graph-structured data. By aggregating the feature information of nodes and their adjacent nodes, it learns the hidden representations of the nodes. First, standardize the knowledge graph, convert the node features into numerical vectors (such as processing the equipment model through one-hot encoding), and construct an adjacency matrix to represent the connection relationships between nodes. Then, input the adjacency matrix and the node feature matrix into the GCN model. Through the operations of multiple convolutional layers, extract the high-order correlation features of the nodes, such as the synergy features between equipment energy efficiency and process improvement solutions, and the linkage features between clean technology applications and carbon trading quotas. The output of each convolutional layer is used as the input of the next layer, and a non-linear transformation is introduced through an activation function (such as ReLU). Finally, obtain the low-dimensional feature vectors of each node in the output layer. Arrange the feature vectors of all nodes in order to form a feature matrix for emission reduction solutions. This matrix contains the potential features and correlation relationships of various emission reduction strategies in the knowledge graph, providing input data for subsequent optimization models.

[0065] Establish a multi-objective optimization model and use the NSGA-II algorithm to solve the Pareto optimal solution set. The multi-objective optimization model aims to maximize carbon emission reduction, minimize economic cost, and minimize the implementation cycle, while considering constraints such as the production capacity of the enterprise, technical feasibility, and policy compliance. The carbon emission reduction target is measured by calculating the impact of different emission reduction solutions on the enterprise's carbon emission intensity and total amount. The economic cost target includes equipment procurement costs, technology transformation costs, personnel training costs, etc. The implementation cycle target is determined based on the complexity of the solution and resource requirements. Constraint conditions such as equipment transformation cannot affect the normal production schedule, and clean technology applications need to meet environmental protection policy requirements.

[0066] The NSGA-II algorithm is a non-dominated sorting genetic algorithm that searches for Pareto optimal solutions in the solution space by simulating operations such as selection, crossover, and mutation in the biological evolution process. The algorithm first generates an initial population, where each individual represents a combination of emission reduction solutions (such as implementing equipment upgrades and process improvements simultaneously). The gene encoding of an individual contains parameters such as solution selection, implementation order, and resource allocation. Then, non-dominated sorting is performed on the population, dividing individuals into different Pareto ranks, and calculating the crowding distance of each individual to maintain population diversity. The next generation population is generated through selection operations (such as roulette wheel selection), crossover operations (such as single-point crossover), and mutation operations (such as randomly changing the implementation order of a certain solution), and the above process is repeated until the preset number of iterations or convergence conditions are reached. The finally obtained Pareto optimal solution set contains multiple non-dominated emission reduction solution combinations, and each solution achieves an optimal balance in objectives such as carbon emissions, cost, and cycle.

[0067] The feature matrix is input into the optimization model, and a set of feasible solutions that meet the carbon emission constraints and economic cost limits is output. The feature matrix, as an input parameter of the optimization model, is used to characterize the potential performance and associated features of each emission reduction solution. The model evaluates and screens each solution in the feature matrix according to the preset objective function and constraint conditions, excludes solutions that do not meet the constraint conditions (such as solutions with costs exceeding the enterprise budget), retains the eligible solutions, and calculates their scores on each objective. Each solution in the set of feasible solutions meets the minimum threshold of carbon emission reduction and the maximum upper limit of economic cost set by the enterprise. For example, it is required that carbon emissions be reduced by at least 10% and the cost does not exceed 5 million yuan.

[0068] The final implementation plan is selected according to the decision-making preference, and a phased emission reduction route plan is generated. The decision-making preference setting is determined by the enterprise management layer according to its own strategic goals and resource status. For example, some enterprises give priority to short-term emission reduction effects and are willing to bear higher costs; some enterprises pay more attention to cost control and tend to implement low-cost solutions in phases. The system provides a visual solution comparison tool to display the performance of the solutions in the set of feasible solutions in dimensions such as carbon emissions, costs, and cycles in the form of charts (such as radar charts and Gantt charts), helping decision-makers intuitively compare the advantages and disadvantages of different solutions. The decision-maker selects the preferred solution combination through the interactive interface (such as selecting a solution with a 15% reduction in carbon emissions, a cost of 4.5 million yuan, and a cycle of 12 months), and the system automatically generates a phased emission reduction route plan according to the implementation steps and resource requirements of the solution.

[0069] The planning table divides the stages in chronological order (such as the preparation stage, the implementation stage, and the acceptance stage), and each stage specifies specific tasks, responsible departments, time nodes, and resource requirements. For example, the tasks in the preparation stage include equipment procurement bidding and technical team formation, the time node is the 1st - 2nd month, and the responsible departments are the procurement department and the technical department; the tasks in the implementation stage include the removal of old equipment, the installation and commissioning of new equipment, and the adjustment of process parameters, the time node is the 3rd - 10th month, and the responsible departments are the production department and the engineering department; the tasks in the acceptance stage include carbon emission monitoring and assessment, and economic benefit analysis, the time node is the 11th - 12th month, and the responsible departments are the environmental protection department and the finance department. The planning table can also link relevant document materials (such as equipment procurement contracts, technical implementation plans) to facilitate access and execution by each department, ensuring the implementation of the optimized decision-making on the emission reduction path. The whole process realizes the intelligent generation and scientific planning of emission reduction strategies by integrating multi-source data, applying graph neural networks, and optimization algorithms, providing an emission reduction solution that combines environmental and economic benefits for the enterprise.

[0070] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0071] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An enterprise carbon emission analysis system based on an ESG comprehensive evaluation model, characterized in that, It includes: a data acquisition layer, an analysis and processing layer, and a decision-making and output layer; The data acquisition layer includes an environmental monitoring device, an energy metering instrument, a supply chain tracking module, and a compliance record unit, and is used for real-time monitoring and raw data acquisition of carbon emission-related elements in the whole process of an enterprise's production and operation; The analysis and processing layer includes edge computing nodes, a distributed storage cluster, and a blockchain verification module, and deploys a dynamic optimization algorithm and a hybrid encryption protocol, and is used for building a trusted data processing environment, performing data cleaning and verification operations, adopting a hybrid encryption protocol for secure transmission and classified storage for different business modules and data types, and realizing the dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model through a standardized interface; The decision-making and output layer is used for integrating the enterprise's historical operation data and real-time monitoring data by using pattern recognition technology, dynamically calibrating the evaluation index system, establishing a multi-dimensional associated ESG comprehensive evaluation model, and calculating the carbon emission intensity, tracing the carbon footprint of the supply chain, and optimizing the decision-making of the emission reduction path based on this model by using an integrated learning algorithm.

2. An enterprise carbon emission analysis system based on an ESG comprehensive evaluation model according to claim 1, characterized in that, The environmental monitoring device at least includes an exhaust gas emission sensor, a wastewater detection probe, a solid waste metering device, and a noise monitor, and is used for obtaining the environmental parameters at the enterprise's production and operation site; the energy metering instrument at least includes an electric energy meter, a gas flowmeter, a steam meter, and a fuel consumption recorder, and is used for counting the enterprise's energy consumption data; the supply chain tracking module at least includes a logistics information acquisition terminal, a raw material traceability database, and a transport vehicle positioning device, and is used for recording the carbon emission information of each link in the supply chain.

3. The enterprise carbon emission analysis system based on the ESG comprehensive evaluation model according to claim 1, wherein, In the analysis and processing layer, the data obtained by each acquisition terminal is transmitted by the edge computing node to the distributed storage cluster through a dedicated channel, and after the data format is converted by the distributed storage cluster, it is jointly stored in the blockchain verification module with the supervision information obtained by the compliance record unit for feature extraction; the data cleaning and verification operation adopts an outlier detection technology, and the collected raw data is input into the verification models established by different verification rules for cross-verification; The realization of the dynamic synchronization of enterprise operation data and the ESG comprehensive evaluation model through a standardized interface includes: establishing a standardized data interface to realize the interactive communication between the business system and the ESG comprehensive evaluation model, synchronizing the actual operation status data of the enterprise, performing an associated mapping on the structured data, and simultaneously updating the model parameter configuration to complete data version management, model iteration update, and decision instruction verification.

4. An enterprise carbon emission analysis system based on an ESG comprehensive evaluation model according to claim 1, characterized in that The establishment of a multi-dimensional associated ESG comprehensive evaluation model includes: constructing a two-way association channel and a dynamic matching mechanism between the enterprise's actual operation data and the parameters of the ESG comprehensive evaluation model; By means of data mapping and feature association of the enterprise's production and operation activities, based on the enterprise's on-site monitoring data, supply chain record data, and compliance information, establishing an enterprise carbon emission inventory, an industry benchmark comparison framework, and initializing the model parameters according to the business characteristics, simulating the operation status of multiple scenarios, and forming a dynamically adjustable enterprise ESG comprehensive evaluation model; Optimize the parameters of the enterprise ESG comprehensive evaluation model, import the multi-dimensional operation data of the enterprise into the established ESG comprehensive evaluation model, and use the Monte Carlo simulation method to correct the sensitivity analysis of the model output results to obtain the optimized enterprise ESG comprehensive evaluation model; Among them, the enterprise ESG comprehensive evaluation model includes operation entity data, an evaluation rule engine, a dynamic knowledge base, and the interaction relationships between modules; The operation entity data is the basic data source of the ESG comprehensive evaluation model, including the enterprise's direct emission data and indirect emission data; the evaluation rule engine establishes a mapping relationship with the actual business data, and quantitatively evaluates the carbon emission characteristics through a multi-dimensional index system and a weight allocation mechanism; the dynamic knowledge base integrates industry standard data, policy and regulation texts, and best practice cases, and forms an evaluation benchmark database through continuous update and maintenance; the interaction relationships between modules achieve information interconnection through a data pipeline, and the operation data and the evaluation rule engine are standardized through a data conversion interface, and the evaluation rule engine and the dynamic knowledge base achieve rule matching through semantic parsing technology.

5. An enterprise carbon emission analysis system based on the ESG comprehensive evaluation model according to claim 1, characterized in that, Using the integrated learning algorithm to calculate the carbon emission intensity based on this model, including: Collect the enterprise's historical annual energy consumption data, production process record data, equipment operation logs, and product carbon footprint reports to construct a carbon emission characteristic data set; After filling in the missing values and correcting the outliers in the characteristic data set, divide it into a training set and a validation set; Configure the initial parameters of the XGBoost-GRU joint model, input the training set into the joint model, sort the feature importance through the XGBoost component, perform time series modeling through the GRU network, and eliminate the impact of emission fluctuations in different production cycles through a feature weighting mechanism until the model error reaches the preset threshold; Input the validation set into the trained joint model, calculate the predicted value of the carbon emission intensity, and generate the evaluation result of the enterprise carbon emission intensity level in combination with the industry benchmark value; Match the preset improvement strategy library according to the evaluation result to generate targeted carbon emission management suggestions.

6. An enterprise carbon emission analysis system based on an ESG comprehensive evaluation model according to claim 1, characterized in that, Using the integrated learning algorithm to trace the carbon footprint of the supply chain based on this model, including: Obtain the logistics transportation data, raw material procurement records, production and processing information, and product distribution paths of each link in the supply chain to construct a supply chain carbon flow data set; Extract the key features of the supply chain carbon flow through the principal component analysis method to establish a carbon footprint tracking feature vector; Construct a random forest regression model and optimize the model parameter combination using the adaptive particle swarm algorithm; Input the feature vector into the optimized random forest regression model to output the ranking of the carbon emission contribution of each link in the supply chain; Identify the key nodes of the carbon footprint according to the contribution ranking result and generate an optimized path map of the supply chain carbon flow.

7. An enterprise carbon emission analysis system based on an ESG comprehensive evaluation model according to claim 1, characterized in that Using the integrated learning algorithm to optimize the decision-making of the emission reduction path based on this model, including: Integrate the enterprise's equipment energy efficiency data, process improvement plan library, clean technology application cases, and carbon trading market information to construct a knowledge graph of emission reduction strategies; Extract the graph convolution network features of the node relationships in the knowledge graph to generate an emission reduction plan feature matrix; A multi-objective optimization model is established, and the NSGA-II algorithm is used to solve the Pareto optimal solution set; The feature matrix is input into the optimization model, and a set of feasible solutions that meet the carbon emission constraints and economic cost limitations is output; According to the decision-making preference, the final implementation plan is selected, and a phased emission reduction route planning table is generated.

8. An enterprise carbon emission analysis system based on an ESG comprehensive evaluation model according to claim 1, characterized in that The specific implementation method of the data cleaning and verification operation in the analysis and processing layer includes: Establish a data quality detection rule library, including numerical range verification rules, data integrity verification rules, and logical consistency verification rules; Use a streaming computing engine to verify real-time data item by item, and trigger an early warning mechanism for data that does not conform to the verification rules; Perform a full-scale scan of historical batch data, and use a data repair algorithm to automatically correct repairable abnormal records; Record the data verification process and exception handling logs through a blockchain smart contract, and generate an immutable data quality audit report.

9. An enterprise carbon emission analysis system based on an ESG comprehensive evaluation model according to claim 4, characterized in that, The construction method of the evaluation rule engine includes: Collect domestic and foreign carbon emission accounting standards, industry technical specifications, and policy and regulatory requirements, and build a standard clause knowledge base; Use natural language processing technology to parse standard documents and extract key clauses to form structured evaluation rules; Establish a rule matching engine to realize the automatic association of enterprise actual data and standard clauses through semantic similarity calculation; Set up a dynamic adjustment mechanism for rule weights to automatically optimize rule priorities according to the policy update frequency and industry practice changes.

10. An enterprise carbon emission analysis method based on an ESG comprehensive evaluation model, characterized in that, Include the following steps: Real-time collect data on waste gas emission concentration, wastewater pollutant indicators, solid waste generation volume, and noise level through environmental monitoring devices deployed at the production and operation site. At the same time, use energy metering instruments to obtain the consumption of electric energy, gas, steam, and fuel; After the collected raw data is filtered for outliers and standardized in format by the edge computing node, it is transmitted to the distributed storage cluster for multi-source data fusion, and a blockchain verification module is used to perform encrypted verification on the data integrity and temporal consistency; Based on pattern recognition technology, conduct feature correlation analysis on historical operation data and real-time monitoring data, dynamically adjust the weight parameters of the ESG evaluation indicators, and build a multi-dimensional ESG comprehensive evaluation model including environmental compliance, energy utilization efficiency, and supply chain carbon footprint; Use a transfer learning framework to perform cross-industry knowledge transfer on the ESG comprehensive evaluation model, and generate carbon emission intensity prediction data under different emission reduction scenarios through Monte Carlo simulation; Combine the logistics path, raw material source, and production and processing information recorded by the supply chain tracking module to establish a full-chain carbon flow mapping relationship, and use a graph neural network to identify key carbon emission nodes; According to the carbon emission intensity prediction results and the carbon flow mapping relationship, use a multi-objective optimization algorithm to solve the set of emission reduction paths that meet the economic constraints, and output a decision-making report including equipment transformation plans, process optimization strategies, and carbon trading suggestions.

Images