A carbon quota management method and system for carbon asset transaction

Abstract

This invention discloses a carbon quota management method and system for carbon asset trading, relating to the field of data management. The method includes: constructing a knowledge graph linking enterprises to the environment, and using a cross-modal attention mechanism to fuse multimodal data to classify enterprises by industry; using a data envelopment analysis algorithm to determine carbon emission risk thresholds for each industry, and constructing a two-layer game framework combined with a multi-head self-attention mechanism to generate a technological progress coefficient sequence to determine carbon quota ranges; constructing dynamic feature vectors and simulating carbon emission evolution paths under different scenarios using a Monte Carlo algorithm to output carbon emission accounting results; predicting the supply and demand trends of carbon quotas, calculating the supply and demand gap under different scenarios, and comparing the supply and demand gap with preset thresholds to trigger an adaptive adjustment mechanism. This invention can identify carbon emission risks, dynamically capture industry technological evolution and market demand changes, and improve the accuracy and flexibility of carbon quota management.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data management, in particular to a carbon quota management method and system for carbon asset transactions. BACKGROUND

[0002] With the continuous expansion of the carbon trading market and the increasing diversification of market participants, traditional carbon quota management methods face many challenges. Existing carbon quota management models usually rely on static quota allocation models, which fail to fully consider market fluctuations, the dynamics and uncertainties of corporate behavior, and therefore cannot adjust carbon quotas in a timely manner to reflect market demand and actual emission reduction capacity. In addition, current carbon quota adjustment mechanisms are mostly based on historical data and fixed quota thresholds, lacking dynamic assessment of corporate and industry carbon emission risks, resulting in inaccurate prediction of future carbon quota supply-demand gaps; such inflexible and responsive mechanisms make carbon quota market adjustment less frequent, making it difficult to respond to environmental changes and market fluctuations, and thus affecting the effectiveness and fairness of the market.

[0003] The prior art has not yet proposed an effective solution to the problems in the related art. SUMMARY

[0004] To solve the problems in the related art, the present application proposes a carbon quota management method and system for carbon asset transactions to overcome the above technical problems existing in the prior art.

[0005] To this end, the specific technical solutions adopted by the present application are as follows:

[0006] According to one aspect of the present application, a carbon quota management method for carbon asset transactions is provided, which comprises:

[0007] S1, collecting multi-modal data of enterprises participating in carbon asset transactions, constructing an enterprise-environment associated knowledge graph, and fusing multi-modal data using a cross-modal attention mechanism to classify enterprises by industry;

[0008] S2, based on the industry classification results, determining the carbon emission risk threshold of each industry using a data envelopment analysis algorithm, and constructing a double-layer game framework, combining a multi-head self-attention mechanism to generate a technology progress coefficient sequence, and determining a carbon quota interval;

[0009] S3, combining historical carbon emission data of the industry with the carbon quota interval, constructing a dynamic feature vector, and simulating carbon emission evolution paths under different scenarios through a Monte Carlo algorithm to output carbon emission accounting results;

[0010] S4, according to the carbon emission accounting result, predict the supply and demand trend of carbon quota, calculate the supply and demand gap under different scenarios, and compare the supply and demand gap with the preset threshold, trigger the adaptive adjustment mechanism to dynamically adjust the carbon quota.

[0011] Further, multi-modal data of enterprises participating in carbon asset transactions are collected, a knowledge graph associated with the environment of the enterprises is constructed, and a cross-modal attention mechanism is used to fuse the multi-modal data to classify the enterprises by industry, including:

[0012] S11, multi-modal data of enterprises participating in carbon asset transactions are collected, and the multi-modal data are preprocessed, and a mode layer of a knowledge graph is designed in combination with the interaction between the enterprises and the environment;

[0013] S12, according to the designed mode layer, the preprocessed multi-modal data are mapped into structured triples to construct a knowledge graph associated with the environment of the enterprises, and a knowledge graph embedding algorithm is used to map entities and relationships in the knowledge graph to a target vector space to generate structure embedding vectors;

[0014] S13, based on the constructed knowledge graph, a heterogeneous graph neural network is established, and the structure embedding vectors are used as initial nodes, a cross-modal attention mechanism is used to fuse the multi-modal data to generate node embedding vectors;

[0015] S14, the generated node embedding vectors are input into a preset classifier to predict the industry categories of the enterprises, and the industry classification probability corresponding to each enterprise is output, and the industry category corresponding to the maximum probability is selected as the industry classification result.

[0016] Further, based on the industry classification result, a data envelopment analysis algorithm is used to determine the carbon emission risk threshold of each industry, a double-layer game framework is constructed, a multi-head self-attention mechanism is used to generate a technical progress coefficient sequence, and a carbon quota interval is determined, including:

[0017] S21, based on the industry classification result, input and output data of each industry are collected to construct a data envelopment analysis model;

[0018] S22, the technical efficiency value of each industry is calculated by using the data envelopment analysis model, and the carbon emission risk is quantified under different confidence levels by combining the collected historical carbon emission data of the industry and a preset conditional value at risk model to determine the carbon emission risk threshold of each industry;

[0019] S23, according to the determined carbon emission risk threshold, a double-layer game framework is constructed, an optimal technical progress coefficient is output by combining a multi-head self-attention mechanism, and the optimal technical progress coefficient is input into a preset Bayesian dynamic linear model to generate a technical progress coefficient sequence;

[0020] S24. Use digital twin technology to simulate the sequence of technological progress coefficients, simulate the carbon emission probability distribution under different scenarios, and calculate the corresponding carbon emission risk value.

[0021] S25. Based on the carbon emission probability distribution and carbon emission risk value, use fuzzy mathematics theory to define fuzzy sets, establish a bi-objective optimization model, and combine it with the preset risk preference to output the carbon quota range.

[0022] Furthermore, based on the determined carbon emission risk threshold, a two-layer game framework is constructed, which combines a multi-head self-attention mechanism to output the optimal technological progress coefficient. This optimal technological progress coefficient is then input into a pre-defined Bayesian dynamic linear model to generate a sequence of technological progress coefficients, including:

[0023] S231. Using the carbon emission risk threshold as a dynamic risk constraint, construct a two-layer game framework consisting of a regulatory meta-agent and several industry agents.

[0024] S232. Embed a multi-head self-attention mechanism in a two-layer game framework to obtain the dynamic correlation between industries, and combine it with the risk constraint information in the regulatory meta-agent to generate a cross-industry fusion vector.

[0025] S233. Based on the generated cross-industry fusion vector, the strategy network in the two-layer game framework is optimized using a centralized training and decentralized execution mechanism. The collected historical time series data is input into the optimized two-layer game framework, and the optimal technological progress coefficient of each industry agent at different time steps is output.

[0026] S234. The output optimal technological progress coefficient is used as the observation value and input into the preset Bayesian dynamic linear model. The state is estimated by recursive filtering algorithm, and a technological progress coefficient sequence is generated based on the evaluation results.

[0027] Furthermore, a multi-head self-attention mechanism is embedded in the two-layer game framework to obtain the dynamic correlation between various industries, and combined with the risk constraint information in the regulatory meta-agent, a cross-industry fusion vector is generated, including:

[0028] S2321. Based on the constructed two-layer game framework, a multi-head self-attention mechanism is embedded in the policy network of each industry agent, a multi-dimensional feature vector is defined, and the dynamic correlation between industries is obtained by calculating the attention weights between the industry agents.

[0029] S2322. Based on the dynamic correlation between industries, each industry intelligence agent uses a query, key-value and value mapping mechanism to calculate the influence weight of each industry intelligence agent on the remaining industry intelligence agents.

[0030] S2323: Combine the influence weights between industry intelligent agents with the dynamic risk constraints in the regulatory meta-intelligent agent and input them into the gating fusion mechanism to generate a cross-industry fusion vector.

[0031] Furthermore, the influence weights among industry intelligent agents and the dynamic risk constraints in the regulatory meta-agent are jointly input into the gating fusion mechanism to generate a cross-industry fusion vector, including:

[0032] S23231. The influence weights among industry intelligent agents are used as target features, and the dynamic risk constraints of regulatory meta-intelligent agents are used as supplementary features. A fusion input vector is generated by concatenating the features.

[0033] S23231. Combining the fusion input vector with the gating fusion mechanism, dynamically adjust the fusion ratio between the target feature and the supplementary feature to generate a technology synergy matrix and a risk constraint matrix;

[0034] S23231. Based on the repetitive features in the technology collaboration matrix and risk constraint matrix, a selection gating mechanism is used for screening, and a cross-industry integration vector is generated based on the screening results.

[0035] Furthermore, based on the carbon emission probability distribution and carbon emission risk value, fuzzy sets are defined using fuzzy mathematics theory, a bi-objective optimization model is established, and combined with a preset risk preference, the output carbon quota range includes:

[0036] S251. Combining the carbon emission probability distribution and the corresponding carbon emission risk value under various scenarios, the carbon quota is defined as a fuzzy set using fuzzy mathematics theory, and the fuzzy membership degree of different carbon quotas is quantified by a preset membership function.

[0037] S252. Based on the fuzzy membership quantification results of carbon quotas, and combined with the preset economic constraints and emission reduction targets, the membership function is dynamically adjusted to establish a dual-objective optimization model.

[0038] S253. Solve the bi-objective optimization model using a multi-objective evolutionary algorithm to generate a Pareto optimal solution set. Combine this with a preset risk preference to determine the weights of the two objectives. Use a weighted sum method to select the optimal solution from the Pareto frontier to output the carbon quota range.

[0039] Furthermore, by combining historical carbon emission data from the industry with carbon quota ranges, a dynamic feature vector is constructed, and the carbon emission evolution path under different scenarios is simulated using the Monte Carlo algorithm. The output carbon emission accounting results include:

[0040] S31. Combining historical carbon emission data and carbon quota ranges in the industry, using time-series causal extraction algorithms, identify the dynamic causal relationships and evolution patterns among carbon emission behaviors, and construct dynamic feature vectors.

[0041] S32. Calculate the matching degree between the dynamic feature vector and the carbon quota range, dynamically adjust the carbon quota in combination with the preset carbon quota adjustment rules, and use the hash algorithm to encode the adjusted carbon quota to generate a digital fingerprint.

[0042] S33. Write the generated digital fingerprint into the distributed ledger of the consortium blockchain, generate the corresponding carbon quota digital certificate through smart contract, and combine it with dynamic feature vector to form a dual blockchain evidence storage.

[0043] S34. Based on the formed dual blockchain evidence storage, the Monte Carlo algorithm is used to simulate the carbon emission evolution path under different scenarios, and the simulation results are aggregated by the federated learning algorithm to output the carbon emission accounting results.

[0044] Furthermore, based on the established dual blockchain notarization, the Monte Carlo algorithm is used to simulate carbon emission evolution paths under different scenarios, and the simulation results are aggregated using a federated learning algorithm to output carbon emission accounting results, including:

[0045] S341. Using the carbon quota digital certificate and dynamic feature vector in the dual blockchain notarization as input parameters, the carbon emission simulation task is divided into layers according to the time scale and spatial scale.

[0046] S342. Based on the hierarchical partitioning results, set differentiated sampling densities, use the Monte Carlo algorithm for random sampling, simulate the carbon emission evolution path under different scenarios, and generate the posterior probability distribution of carbon emissions at each level.

[0047] S343. Using the federated learning algorithm, calculate the statistical characteristics of the posterior distribution of carbon emissions at each level, and aggregate the global statistical characteristics through the federated averaging mechanism. Using the aggregated results as observations, use the Bayesian update rule to iteratively correct the global prior carbon emission distribution.

[0048] S344. Repeat the iterative process from S341 to S343 until the difference between the prior carbon emission distributions after two consecutive updates is less than the preset convergence threshold. Then terminate the iteration and output the carbon emission accounting results.

[0049] According to another aspect of the present invention, a carbon quota management system for carbon asset trading is also provided, the system comprising:

[0050] The industry classification module is used to collect multimodal data from companies participating in carbon asset trading, construct a knowledge graph linking companies to the environment, and use a cross-modal attention mechanism to fuse multimodal data to classify companies by industry.

[0051] The carbon quota determination module is used to determine the carbon emission risk threshold of each industry based on the industry classification results and using the data envelopment analysis algorithm. It also constructs a two-layer game framework and combines a multi-head self-attention mechanism to generate a sequence of technological progress coefficients to determine the carbon quota range.

[0052] The carbon emission accounting module is used to combine historical carbon emission data of the industry with carbon quota ranges to construct dynamic feature vectors, and simulate the carbon emission evolution path under different scenarios through Monte Carlo algorithm to output carbon emission accounting results.

[0053] The carbon quota adjustment module is used to predict the supply and demand trend of carbon quotas based on carbon emission accounting results, calculate the supply and demand gap under different scenarios, compare the supply and demand gap with preset thresholds, and trigger an adaptive adjustment mechanism to dynamically adjust carbon quotas.

[0054] The beneficial effects of this invention are as follows:

[0055] 1. This invention can identify carbon emission risks in an industry by integrating multimodal data from enterprises and combining it with a cross-modal attention mechanism; it uses data envelopment analysis to assess the carbon emission efficiency of various industries, and combines a two-layer game framework and a technology progress coefficient to dynamically capture industry technology evolution and changes in market demand, thereby improving the accuracy and flexibility of carbon quota management.

[0056] 2. This invention generates carbon emission evolution paths under multiple scenarios by constructing dynamic feature vectors and combining them with Monte Carlo simulation, and predicts future emission trends; based on carbon emission accounting results, it dynamically responds to corporate behavior and environmental changes, providing a basis for the precise and intelligent adjustment of carbon quotas.

[0057] 3. This invention combines carbon quota supply and demand forecasting with an adaptive adjustment mechanism to monitor market supply and demand dynamics in real time, accurately identify gaps and trigger automatic adjustments, thereby improving response speed and enhancing the flexibility, timeliness and overall market stability of carbon quota management. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of a carbon quota management method for carbon asset trading according to an embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of a carbon quota management system for carbon asset trading according to an embodiment of the present invention.

[0061] In the picture:

[0062] 1. Industry classification module; 2. Carbon quota determination module; 3. Carbon emission accounting module; 4. Carbon quota adjustment module. Detailed Implementation

[0063] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0064] According to embodiments of the present invention, a carbon quota management method and system for carbon asset trading are provided.

[0065] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a carbon quota management method for carbon asset trading includes:

[0066] S1. Collect multimodal data from companies participating in carbon asset trading, construct a knowledge graph linking companies to the environment, and use cross-modal attention mechanisms to fuse multimodal data to classify companies by industry.

[0067] In this optional embodiment, multimodal data of companies participating in carbon asset trading is collected, a knowledge graph relating companies to the environment is constructed, and the multimodal data is fused using a cross-modal attention mechanism to classify companies by industry, including:

[0068] S11. Collect multimodal data from companies participating in carbon asset trading, preprocess the multimodal data, and design the pattern layer of the knowledge graph by combining the interaction between companies and the environment.

[0069] It should be further explained that multimodal data includes structured data, such as carbon emissions and industry codes; semi-structured data, such as report texts and policy summaries; and unstructured data, such as corporate website content, news reports, and remote sensing images. The multimodal data undergoes unified formatting and standardized preprocessing. Combining the interaction between enterprises and the environment, such as the spatial mapping between emission sources and emission locations, and the constraint relationship between policies and industries, a schema layer of the knowledge graph is designed, clarifying node types, edge types, and semantic relationships between entities.

[0070] S12. Based on the designed pattern layer, the preprocessed multimodal data is mapped into structured triples to construct a knowledge graph relating enterprises and the environment. The knowledge graph embedding algorithm is then used to map the entities and relationships in the knowledge graph to the target vector space to generate structured embedding vectors.

[0071] It should be further explained that, according to the designed pattern layer, the preprocessed multimodal data is mapped into structured triples, such as "Company A, Emissions to, Region B" or "Company C, Belongs to, Industry D", to construct a knowledge graph relating enterprises and the environment; knowledge graph embedding algorithms, such as TransE, RotatE, or ComplEx, are used to encode the entities and relationships in the graph into a unified low-dimensional vector space to generate structured embedding vectors.

[0072] S13. Based on the constructed knowledge graph, a heterogeneous graph neural network is established, and the structure embedding vector is used as the initial node. Multimodal data is fused using a cross-modal attention mechanism to generate node embedding vectors.

[0073] S14. Input the generated node embedding vector into the preset classifier to predict the industry category of the enterprise, and output the industry classification probability corresponding to each enterprise. Select the industry category with the highest probability as the industry classification result.

[0074] It should be further explained that the node embedding vector is input into a preset classifier, such as a Softmax classifier, a fully connected neural network, or a graph attention-enhanced classification module, to predict the industry to which the enterprise belongs, output the industry classification probability distribution for each enterprise, and select the industry category corresponding to the highest probability as the final classification result; for example, if the classification output of a certain enterprise is "Manufacturing: 0.62, Energy: 0.25, Transportation: 0.13", then its industry category is determined to be manufacturing.

[0075] S2. Based on the industry classification results, the carbon emission risk threshold of each industry is determined by the data envelopment analysis algorithm, and a two-layer game framework is constructed. Combined with the multi-head self-attention mechanism, the technological progress coefficient sequence is generated to determine the carbon quota range.

[0076] In this optional embodiment, based on the industry classification results, the carbon emission risk threshold for each industry is determined using the Data Envelopment Analysis (DEA) algorithm. A two-layer game framework is constructed, and a technological progress coefficient sequence is generated by combining a multi-head self-attention mechanism to determine the carbon quota range, which includes:

[0077] S21. Based on the industry classification results, collect input and output data for each industry and construct a data envelopment analysis model.

[0078] S22. Calculate the technical efficiency value of each industry using the data envelopment analysis model, and combine the collected historical carbon emission data of the industry with the preset conditional value at risk model to quantify carbon emission risk at different confidence levels and determine the carbon emission risk threshold of each industry.

[0079] S23. Based on the determined carbon emission risk threshold, construct a two-layer game framework, combine a multi-head self-attention mechanism to output the optimal technological progress coefficient, and input the optimal technological progress coefficient into a preset Bayesian dynamic linear model to generate a technological progress coefficient sequence.

[0080] In this optional embodiment, a two-layer game framework is constructed based on a determined carbon emission risk threshold. A multi-head self-attention mechanism is used to output the optimal technological progress coefficient. This optimal technological progress coefficient is then input into a preset Bayesian dynamic linear model to generate a sequence of technological progress coefficients, including:

[0081] S231. Using the carbon emission risk threshold as a dynamic risk constraint, construct a two-layer game framework consisting of a regulatory meta-agent and several industry agents.

[0082] It should be further explained that the two-layer game framework consists of "upper-level policy-making game" and "lower-level corporate behavior game"; the upper-level intelligent agent (i.e., the regulatory meta-intelligent agent) represents the regulatory agency, whose goal is to set the optimal quota and policy intensity, while the lower-level intelligent agent (i.e., the industry intelligent agent) represents enterprises or industry organizations, whose goal is to maximize economic benefits and technical efficiency while complying with policies.

[0083] S232. Embed a multi-head self-attention mechanism in a two-layer game framework to obtain the dynamic correlation between industries, and combine it with the risk constraint information in the regulatory meta-agent to generate a cross-industry fusion vector.

[0084] In this optional embodiment, a multi-head self-attention mechanism is embedded in the two-layer game framework to obtain the dynamic correlation between various industries, and combined with the risk constraint information in the regulatory meta-agent, a cross-industry fusion vector is generated, including:

[0085] S2321. Based on the constructed two-layer game framework, a multi-head self-attention mechanism is embedded in the policy network of each industry agent, a multi-dimensional feature vector is defined, and the dynamic correlation between industries is obtained by calculating the attention weights between industry agents.

[0086] S2322. Based on the dynamic correlation between industries, each industry agent uses a query, key-value and value mapping mechanism to calculate the influence weight of each industry agent on the remaining industry agents.

[0087] S2323: Combine the influence weights between industry intelligent agents with the dynamic risk constraints in the regulatory meta-intelligent agent and input them into the gating fusion mechanism to generate a cross-industry fusion vector.

[0088] In this optional embodiment, the influence weights among industry intelligent agents and the dynamic risk constraints in the regulatory meta-intelligent agent are jointly input into the gating fusion mechanism to generate a cross-industry fusion vector, including:

[0089] S23231. The influence weights among industry intelligent agents are used as target features, and the dynamic risk constraints of regulatory meta-intelligent agents are used as supplementary features. A fusion input vector is generated by concatenating the features.

[0090] S23231. Combining the fusion input vector with the gating fusion mechanism, dynamically adjust the fusion ratio between the target feature and the supplementary feature to generate a technology synergy matrix and a risk constraint matrix;

[0091] S23231. Based on the repetitive features in the technology collaboration matrix and risk constraint matrix, a selection gating mechanism is used for screening, and a cross-industry integration vector is generated based on the screening results.

[0092] S233. Based on the generated cross-industry fusion vector, the strategy network in the two-layer game framework is optimized using a centralized training and decentralized execution mechanism. The collected historical time series data is input into the optimized two-layer game framework, and the optimal technological progress coefficient of each industry agent at different time steps is output.

[0093] It should be further explained that centralized training uses global data, such as historical carbon emissions and industry GDP share, to jointly update network parameters and ensure that the model learns universal strategies across industries. The decentralized execution phase allows agents in each industry to make differentiated decisions based on their local state, such as enterprise size and technological reserves. The optimized game framework receives historical time-series data input and outputs the optimal technological progress coefficients for each industry at each time step through multiple rounds of strategy game iteration. These coefficients reflect both the industry's own technological upgrading potential and its response to strategies from other industries.

[0094] S234. The output optimal technological progress coefficient is used as the observation value and input into the preset Bayesian dynamic linear model. The state is estimated by recursive filtering algorithm, and a technological progress coefficient sequence is generated based on the evaluation results.

[0095] S24. Use digital twin technology to simulate the sequence of technological progress coefficients, simulate the carbon emission probability distribution under different scenarios, and calculate the corresponding carbon emission risk value.

[0096] S25. Based on the carbon emission probability distribution and carbon emission risk value, use fuzzy mathematics theory to define fuzzy sets, establish a bi-objective optimization model, and combine it with the preset risk preference to output the carbon quota range.

[0097] In this optional embodiment, based on the carbon emission probability distribution and carbon emission risk value, a fuzzy set is defined using fuzzy mathematics theory to establish a bi-objective optimization model. Combined with a preset risk preference, the output carbon quota range includes:

[0098] S251. Combining the carbon emission probability distribution and the corresponding carbon emission risk value under various scenarios, the carbon quota is defined as a fuzzy set using fuzzy mathematics theory, and the fuzzy membership degree of different carbon quotas is quantified by a preset membership function.

[0099] S252. Based on the fuzzy membership quantification results of carbon quotas, and combined with the preset economic constraints and emission reduction targets, the membership function is dynamically adjusted to establish a dual-objective optimization model.

[0100] It should be further explained that the two objectives of the bi-objective optimization model are usually: maximizing the fuzzy rationality (i.e., membership degree) of carbon quotas and minimizing economic costs or maximizing emission reduction benefits; economic constraints can include total carbon quota limits, marginal emission reduction costs, industrial distribution constraints, etc., and can also include carbon neutrality paths set by policies as hard or soft constraints; on this basis, the parameters of the membership function, such as the mean and standard deviation of the Gaussian function, are dynamically adjusted so that the model can adapt to different objective weights or changes in the external environment.

[0101] S253. Solve the bi-objective optimization model using a multi-objective evolutionary algorithm to generate a Pareto optimal solution set. Combine this with a preset risk preference to determine the weights of the two objectives. Use a weighted sum method to select the optimal solution from the Pareto frontier to output the carbon quota range.

[0102] It should be further explained that when solving a bi-objective optimization model using a multi-objective evolutionary algorithm, such as NSGA-II or MOEA / D, the result is a set of Pareto optimal solutions, i.e., solutions where one objective cannot be further improved by sacrificing the other. After obtaining the Pareto frontier solution set, based on a pre-defined risk preference (e.g., setting the environmental safety weight to 0.6 and the economic benefit weight to 0.4), the weighting coefficients of the two objectives are determined. Then, a weighted sum method is used to select the unique optimal solution from the Pareto frontier. The carbon quota value corresponding to this optimal solution is the final recommended carbon quota range, which can be applied to the quota policy formulation of different industries or entities.

[0103] S3. Combining historical carbon emission data and carbon quota ranges in the industry, a dynamic feature vector is constructed, and the carbon emission evolution path under different scenarios is simulated using the Monte Carlo algorithm to output the carbon emission accounting results.

[0104] In this optional embodiment, a dynamic feature vector is constructed by combining historical carbon emission data of the industry with carbon quota ranges, and the carbon emission evolution path under different scenarios is simulated using the Monte Carlo algorithm. The output carbon emission accounting results include:

[0105] S31. Combining historical carbon emission data and carbon quota ranges in the industry, and using time-series causal extraction algorithms, identify the dynamic causal relationships and evolutionary patterns among carbon emission behaviors, and construct dynamic feature vectors.

[0106] S32. Calculate the matching degree between the dynamic feature vector and the carbon quota range, dynamically adjust the carbon quota in combination with the preset carbon quota adjustment rules, and use a hash algorithm to encode the adjusted carbon quota to generate a digital fingerprint.

[0107] S33. Write the generated digital fingerprint into the distributed ledger of the consortium blockchain, generate the corresponding carbon quota digital certificate through smart contract, and combine it with dynamic feature vector to form a dual blockchain notarization.

[0108] S34. Based on the formed dual blockchain evidence storage, the Monte Carlo algorithm is used to simulate the carbon emission evolution path under different scenarios, and the simulation results are aggregated by the federated learning algorithm to output the carbon emission accounting results.

[0109] In this optional embodiment, based on the formed dual blockchain notarization, the Monte Carlo algorithm is used to simulate carbon emission evolution paths under different scenarios, and the simulation results are aggregated using a federated learning algorithm to output carbon emission accounting results, including:

[0110] S341. Using the carbon quota digital certificate and dynamic feature vector in the dual blockchain notarization as input parameters, the carbon emission simulation task is divided into layers according to the time scale and spatial scale.

[0111] S342. Based on the hierarchical partitioning results, set differentiated sampling densities, use the Monte Carlo algorithm for random sampling, simulate the carbon emission evolution path under different scenarios, and generate the posterior probability distribution of carbon emissions at each level.

[0112] S343. Using the federated learning algorithm, calculate the statistical characteristics of the posterior distribution of carbon emissions at each level, and aggregate the global statistical characteristics through the federated averaging mechanism. Using the aggregated results as observations, use the Bayesian update rule to iteratively correct the global prior carbon emission distribution.

[0113] S344. Repeat the iterative process from S341 to S343 until the difference between the prior carbon emission distributions after two consecutive updates is less than the preset convergence threshold. Then terminate the iteration and output the carbon emission accounting results.

[0114] S4. Based on the carbon emission accounting results, predict the supply and demand trend of carbon quotas, calculate the supply and demand gap under different scenarios, compare the supply and demand gap with the preset threshold, and trigger the adaptive adjustment mechanism to dynamically adjust the carbon quotas.

[0115] It should be further explained that, based on the carbon emission accounting results, the supply and demand trends of carbon allowances are predicted, the supply and demand gap under different scenarios is calculated, and the supply and demand gap is compared with a preset threshold to trigger an adaptive adjustment mechanism to dynamically adjust carbon allowances. Specifically, this includes:

[0116] Based on carbon emission accounting results, a carbon quota demand forecasting model is constructed. An LSTM neural network is used to perform time series analysis on historical emission data. Input features include industry activity levels, energy consumption intensity, and clean technology penetration rates. This yields future carbon emission forecasts for various industries or regions at different time steps, such as annual or quarterly. These forecasts include not only static averages but also probability distributions to reflect emission uncertainties. For example, a Bayesian structured time series model can be used to predict the 95% confidence interval for future carbon emissions. Simultaneously, by combining policy settings, carbon market rules, and reserve release plans, a corresponding future carbon quota total supply curve is constructed. Supply forecasts can be achieved using linear programming models or rule-based models based on policy simulations. The supply and demand trend represents the relative relationship between carbon quota demand and supply over time.

[0117] Several scenarios are constructed for each time step, such as the baseline emission scenario, the technology-accelerated emission reduction scenario, and the high carbon price shock scenario. A weight is assigned to each scenario, based on historical similarity or expert knowledge, for example, 0.4, 0.35, and 0.25. Under each scenario, the carbon quota supply and demand gap ΔQ = expected demand - predicted supply is calculated, and its expected value and variance are recorded. For example, if ΔQ > 0, it indicates a shortage of quotas in the market, which will lead to price increases and increased pressure on enterprises to reduce emissions. Conversely, it indicates a loose market, leading to a surplus of quotas and a price drop. At this time, a dynamic supply and demand tolerance threshold ε is introduced. This value can be a fixed value, such as 10% of the total quota, or a dynamic value, such as being linked to the historical volatility of the carbon market. The ΔQ under each current scenario is compared with ε. If the ΔQ of a certain scenario exceeds ε and the probability of the scenario occurring is greater than a certain critical probability, such as 0.3, the system judges that there is a possibility of risk imbalance and enters the next adjustment process.

[0118] Once the triggering conditions are met, an adaptive adjustment mechanism is activated. This mechanism consists of three parts: a rapid adjustment path, which adjusts short-term quota release strategies, such as releasing some quotas from carbon reserves or reducing auction quotas for future trading periods; the rate of this adjustment can be automatically calculated by a PID controller; a medium-term structural adjustment, which modifies quota allocation weighting coefficients, such as adjusting them according to industry α coefficients, i.e., adjusting the benchmark emission reference values ​​or baseline slopes for different industries, so that resources are shifted from inefficient industries to industries with high emission reduction potential; and a long-term adjustment strategy, such as modifying the offsetting mechanism, such as the CCER ratio cap, adjusting the pace of new industry inclusion, or revising carbon market regulations, to achieve a rebalancing of the system's supply and demand structure at the institutional level.

[0119] like Figure 2 As shown, according to another embodiment of the present invention, a carbon quota management system for carbon asset trading is also provided, the system comprising:

[0120] Industry classification module 1 is used to collect multimodal data of companies participating in carbon asset trading, construct a knowledge graph linking companies to the environment, and use a cross-modal attention mechanism to fuse multimodal data to classify companies by industry.

[0121] The carbon quota determination module 2 is used to determine the carbon emission risk threshold of each industry based on the industry classification results and using the data envelopment analysis algorithm. It also constructs a two-layer game framework and combines a multi-head self-attention mechanism to generate a sequence of technological progress coefficients to determine the carbon quota range.

[0122] The carbon emission accounting module 3 is used to combine historical carbon emission data of the industry with carbon quota ranges to construct dynamic feature vectors, and simulate carbon emission evolution paths under different scenarios through Monte Carlo algorithm to output carbon emission accounting results.

[0123] The carbon quota adjustment module 4 is used to predict the supply and demand trend of carbon quotas based on the carbon emission accounting results, calculate the supply and demand gap under different scenarios, compare the supply and demand gap with the preset threshold, and trigger the adaptive adjustment mechanism to dynamically adjust the carbon quotas.

[0124] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for managing carbon quotas in carbon asset trading, characterized in that, The method includes: S1. Collect multimodal data of companies participating in carbon asset trading, construct a knowledge graph linking companies and the environment, and use cross-modal attention mechanisms to fuse multimodal data to classify companies by industry. S2. Based on the industry classification results, the carbon emission risk threshold of each industry is determined by the data envelopment analysis algorithm, and a two-layer game framework is constructed. The technology progress coefficient sequence is generated by combining the multi-head self-attention mechanism to determine the carbon quota range. S3. Combine historical carbon emission data and carbon quota ranges in the industry to construct a dynamic feature vector, and use the Monte Carlo algorithm to simulate the carbon emission evolution path under different scenarios to output carbon emission accounting results. S4. Based on the carbon emission accounting results, predict the supply and demand trend of carbon quotas, calculate the supply and demand gap under different scenarios, compare the supply and demand gap with the preset threshold, and trigger the adaptive adjustment mechanism to dynamically adjust the carbon quotas. S2 includes: S21. Based on the industry classification results, collect input and output data for each industry and construct a data envelopment analysis model. S22. Calculate the technical efficiency value of each industry using the data envelopment analysis model, and combine the collected historical carbon emission data of the industry with the preset conditional value at risk model to quantify carbon emission risk at different confidence levels and determine the carbon emission risk threshold of each industry. S23. Based on the determined carbon emission risk threshold, construct a two-layer game framework, combine a multi-head self-attention mechanism to output the optimal technological progress coefficient, and input the optimal technological progress coefficient into a preset Bayesian dynamic linear model to generate a technological progress coefficient sequence. S24. Use digital twin technology to simulate the sequence of technological progress coefficients, simulate the carbon emission probability distribution under different scenarios, and calculate the corresponding carbon emission risk value. S25. Based on the carbon emission probability distribution and carbon emission risk value, use fuzzy mathematics theory to define fuzzy sets, establish a bi-objective optimization model, and combine it with the preset risk preference to output the carbon quota range.

2. The carbon quota management method for carbon asset trading according to claim 1, characterized in that, The process of collecting multimodal data from companies participating in carbon asset trading, constructing a knowledge graph linking companies to the environment, and using a cross-modal attention mechanism to fuse the multimodal data to classify companies by industry includes: S11. Collect multimodal data from companies participating in carbon asset trading, preprocess the multimodal data, and design the pattern layer of the knowledge graph based on the interaction between companies and the environment. S12. Based on the designed pattern layer, the preprocessed multimodal data is mapped into structured triples to construct a knowledge graph relating enterprises and the environment. The knowledge graph embedding algorithm is then used to map the entities and relationships in the knowledge graph to the target vector space to generate structured embedding vectors. S13. Based on the constructed knowledge graph, a heterogeneous graph neural network is established, and the structure embedding vector is used as the initial node. Multimodal data is fused using a cross-modal attention mechanism to generate node embedding vectors. S14. Input the generated node embedding vector into the preset classifier to predict the industry category of the enterprise, and output the industry classification probability corresponding to each enterprise. Select the industry category with the highest probability as the industry classification result.

3. A carbon quota management method for carbon asset trading according to claim 1, characterized in that, The process involves constructing a two-layer game framework based on a determined carbon emission risk threshold, combining a multi-head self-attention mechanism to output the optimal technological progress coefficient, and inputting the optimal technological progress coefficient into a preset Bayesian dynamic linear model to generate a technological progress coefficient sequence, including: S231. Using the carbon emission risk threshold as a dynamic risk constraint, construct a two-layer game framework consisting of a regulatory meta-agent and several industry agents. S232. Embed a multi-head self-attention mechanism in a two-layer game framework to obtain the dynamic correlation between industries, and combine it with the risk constraint information in the regulatory meta-agent to generate a cross-industry fusion vector. S233. Based on the generated cross-industry fusion vector, the strategy network in the two-layer game framework is optimized using a centralized training and decentralized execution mechanism. The collected historical time series data is input into the optimized two-layer game framework, and the optimal technological progress coefficient of each industry agent at different time steps is output. S234. The output optimal technological progress coefficient is used as the observation value and input into the preset Bayesian dynamic linear model. The state is estimated by recursive filtering algorithm, and a technological progress coefficient sequence is generated based on the evaluation results.

4. A carbon quota management method for carbon asset trading according to claim 3, characterized in that, The method of embedding a multi-head self-attention mechanism into a two-layer game framework to obtain dynamic correlations between industries, and combining risk constraint information from the regulatory meta-agent to generate a cross-industry fusion vector includes: S2321. Based on the constructed two-layer game framework, a multi-head self-attention mechanism is embedded in the policy network of each industry agent, a multi-dimensional feature vector is defined, and the dynamic correlation between industries is obtained by calculating the attention weights between the industry agents. S2322. Based on the dynamic correlation between industries, each industry agent uses a query, key-value and value mapping mechanism to calculate the influence weight of each industry agent on the remaining industry agents. S2323: Combine the influence weights between industry intelligent agents with the dynamic risk constraints in the regulatory meta-intelligent agent and input them into the gating fusion mechanism to generate a cross-industry fusion vector.

5. A carbon quota management method for carbon asset trading according to claim 4, characterized in that, The method of jointly inputting the influence weights among industry intelligent agents and the dynamic risk constraints in the regulatory meta-intelligent agent into the gating fusion mechanism to generate a cross-industry fusion vector includes: S23231. The influence weights among industry intelligent agents are used as target features, and the dynamic risk constraints of regulatory meta-intelligent agents are used as supplementary features. A fusion input vector is generated by concatenating the features. S23231. Combining the fusion input vector with the gating fusion mechanism, dynamically adjust the fusion ratio between the target feature and the supplementary feature to generate a technology synergy matrix and a risk constraint matrix; S23231. Based on the repetitive features in the technology collaboration matrix and risk constraint matrix, a selection gating mechanism is used for screening, and a cross-industry integration vector is generated based on the screening results.

6. A carbon quota management method for carbon asset trading according to claim 5, characterized in that, The process involves defining fuzzy sets using fuzzy mathematics theory based on the carbon emission probability distribution and carbon emission risk value, establishing a dual-objective optimization model, and combining this with a preset risk preference to output carbon quota ranges, including: S251. Combining the carbon emission probability distribution and the corresponding carbon emission risk value under various scenarios, the carbon quota is defined as a fuzzy set using fuzzy mathematics theory, and the fuzzy membership degree of different carbon quotas is quantified by a preset membership function. S252. Based on the fuzzy membership quantification results of carbon quotas, and combined with the preset economic constraints and emission reduction targets, the membership function is dynamically adjusted to establish a dual-objective optimization model. S253. Solve the bi-objective optimization model using a multi-objective evolutionary algorithm to generate a Pareto optimal solution set. Combine this with a preset risk preference to determine the weights of the two objectives. Use a weighted sum method to select the optimal solution from the Pareto frontier to output the carbon quota range.

7. A carbon quota management method for carbon asset trading according to claim 1, characterized in that, The process involves combining historical carbon emission data from the industry with carbon quota ranges to construct a dynamic feature vector. Then, using the Monte Carlo algorithm, it simulates the carbon emission evolution paths under different scenarios, outputting carbon emission accounting results including: S31. Combining historical carbon emission data and carbon quota ranges in the industry, using time-series causal extraction algorithms, identify the dynamic causal relationships and evolution patterns among carbon emission behaviors, and construct dynamic feature vectors. S32. Calculate the matching degree between the dynamic feature vector and the carbon quota range, dynamically adjust the carbon quota in combination with the preset carbon quota adjustment rules, and use the hash algorithm to encode the adjusted carbon quota to generate a digital fingerprint. S33. Write the generated digital fingerprint into the distributed ledger of the consortium blockchain, generate the corresponding carbon quota digital certificate through smart contract, and combine it with dynamic feature vector to form a dual blockchain evidence storage. S34. Based on the formed dual blockchain evidence storage, the Monte Carlo algorithm is used to simulate the carbon emission evolution path under different scenarios, and the simulation results are aggregated by the federated learning algorithm to output the carbon emission accounting results.

8. A carbon quota management method for carbon asset trading according to claim 7, characterized in that, The dual blockchain-based notarization uses the Monte Carlo algorithm to simulate carbon emission evolution paths under different scenarios, and combines the simulation results with a federated learning algorithm to output carbon emission accounting results, including: S341. Using the carbon quota digital certificate and dynamic feature vector in the dual blockchain notarization as input parameters, the carbon emission simulation task is divided into layers according to the time scale and spatial scale. S342. Based on the hierarchical partitioning results, set differentiated sampling densities, use the Monte Carlo algorithm for random sampling, simulate the carbon emission evolution path under different scenarios, and generate the posterior probability distribution of carbon emissions at each level. S343. Using the federated learning algorithm, calculate the statistical characteristics of the posterior distribution of carbon emissions at each level, and aggregate the global statistical characteristics through the federated averaging mechanism. Using the aggregated results as observations, use the Bayesian update rule to iteratively correct the global prior carbon emission distribution. S344. Repeat the iterative process from S341 to S343 until the difference between the prior carbon emission distributions after two consecutive updates is less than the preset convergence threshold. Then terminate the iteration and output the carbon emission accounting results.

9. A carbon quota management system for carbon asset trading, used to implement the carbon quota management method for carbon asset trading as described in any one of claims 1-8, characterized in that, The system includes: The industry classification module is used to collect multimodal data from companies participating in carbon asset trading, construct a knowledge graph linking companies to the environment, and use a cross-modal attention mechanism to fuse multimodal data to classify companies by industry. The carbon quota determination module is used to determine the carbon emission risk threshold of each industry based on the industry classification results and using the data envelopment analysis algorithm. It also constructs a two-layer game framework and combines a multi-head self-attention mechanism to generate a sequence of technological progress coefficients to determine the carbon quota range. The carbon emission accounting module is used to combine historical carbon emission data of the industry with carbon quota ranges to construct dynamic feature vectors, and simulate the carbon emission evolution path under different scenarios through Monte Carlo algorithm to output carbon emission accounting results. The carbon quota adjustment module is used to predict the supply and demand trend of carbon quotas based on carbon emission accounting results, calculate the supply and demand gap under different scenarios, compare the supply and demand gap with a preset threshold, and trigger an adaptive adjustment mechanism to dynamically adjust the carbon quotas. The carbon quota determination module includes: collecting input and output data for each industry based on industry classification results, and constructing a data envelopment analysis (DEA) model; calculating the technical efficiency value of each industry using the DEA model, and quantifying carbon emission risk at different confidence levels by combining the collected historical carbon emission data of the industry with a preset conditional value at risk (VaR) model, and determining the carbon emission risk threshold for each industry; constructing a two-layer game framework based on the determined carbon emission risk threshold, outputting the optimal technical progress coefficient by combining a multi-head self-attention mechanism, and inputting the optimal technical progress coefficient into a preset Bayesian dynamic linear model to generate a technical progress coefficient sequence; simulating the technical progress coefficient sequence using digital twin technology to simulate the carbon emission probability distribution under different scenarios, and calculating the corresponding carbon emission risk value; and defining a fuzzy set using fuzzy mathematics theory based on the carbon emission probability distribution and the carbon emission risk value, establishing a bi-objective optimization model, and outputting the carbon quota range by combining a preset risk preference.

Images