An automated data verification system in carbon asset trading

Abstract

The present invention provides an automated data verification system in carbon asset trading, which relates to the technical field of carbon asset management. The system includes a data acquisition layer, a data integration layer, a verification model layer, and an interaction and security layer. The sliding window technology of the LSTM time series analysis unit increases the number of training samples. The SVM classification verification unit processes non-linear data through kernel function techniques and optimizes parameters through cross-validation. The two work together, making both the trend prediction and anomaly judgment of carbon data more reliable, greatly reducing the cases of misjudgment and missed judgment, thereby improving the accuracy and reliability of the entire automated data verification system, providing a strong guarantee for the fairness and justice of carbon asset trading, being able to effectively collect and integrate carbon asset trading data, conduct accurate verification using advanced algorithm models, and ensure the security and credibility of data interaction through blockchain and smart contracts, providing strong technical support for the healthy and stable development of the carbon asset trading market.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon asset management, and particularly to an automated data verification system in carbon asset trading. Background Art

[0002] According to a complex organizational structure coding system and coding method for calculating carbon asset verification disclosed in "CN116578289A" with a Chinese publication number, which includes an organizational structure coding system. A business management module and a collection module are provided on the organizational structure coding system. A main control module is provided between the business management module and the collection module. A data file storage unit is provided at one connection end of the main control module. In this coding system, the first digit of the code starts with C or E. C represents the organizational or company level, and E represents the project level. Coding is carried out according to the sorted parent-child node information. For the coding of the organizational and company levels, when calculating carbon assets for carbon data verification, first, the organizational structure can be quickly located according to this coding system. Then, according to the located company or project, the system can directly read the code of this project and then find the corresponding carbon asset data, realizing the fast and accurate search for carbon asset data at any level, any organization, and any project, with fast search speed and high accuracy.

[0003] According to a carbon asset full-life cycle management platform based on a new power system disclosed in "CN118429135A" with a Chinese publication number, the platform includes a production module, a processing module, a verification module, a circulation and trading module, and a post-evaluation module. Among them, the production module can collect and store carbon emission data generated in the new power system to uniformly manage carbon emission data from different emission sources; the processing module can process carbon emission data into carbon assets, so that the conversion of carbon assets can be realized using the same standard; the verification module can verify the data authenticity of carbon assets to ensure the authenticity and effectiveness of carbon assets; the circulation and trading module can connect each scenario node and share the circulation data of carbon assets in each scenario node, thus ensuring the transparency and legality of the circulation data; the post-evaluation module can comprehensively evaluate carbon emission data and each carbon asset, thereby reducing adverse factors.

[0004] The above patent documents and the prior art have the following technical problems when in use:

[0005] Problem 1: In traditional carbon asset trading data processing, when integrating multi-source data and model training, it is often necessary to centralize the original data of each data holder, which makes the data face a great risk of privacy leakage, and it is difficult to comprehensively converge the data, affecting the accuracy and comprehensiveness of carbon asset trading data verification;

[0006] Problem 2: The previous carbon data acquisition system lacks an effective preprocessing mechanism for massive and diverse data. A large amount of data is directly transmitted without optimization, occupying a large amount of network bandwidth and having a slow transmission speed. Moreover, the carbon data verification model mostly uses relatively simple algorithms or single algorithms, making it difficult to handle the complex temporal and non-linear characteristics of carbon asset transaction data, prone to misjudgment or missed judgment, resulting in inaccurate abnormal judgment of carbon data;

[0007] Problem 3: The carbon trading data integration method often simply piles up the data together, without establishing deep connections at the semantic level. Data from different sources lack a unified semantic framework, making it difficult to conduct effective correlation analysis. Moreover, the complex relationships between trading entities in carbon trading and between trading and the external environment are not deeply explored, unable to comprehensively and accurately reflect the true situation of carbon assets, and not conducive to the accurate accounting and verification of carbon asset transactions;

[0008] Problem 4: The previous carbon trading smart contracts have relatively single functions, tightly coupling functions such as carbon data submission, verification, and trading settlement. When carbon trading rules or verification processes change, such as adjustments to carbon emission standards or the introduction of new trading varieties, large-scale modifications need to be made to the entire smart contract code. This not only has a high development cost but also easily introduces new errors, making it difficult for smart contracts to adapt to the dynamic changes in the carbon trading market. At the same time, the execution process of trading smart contracts lacks a transparent supervision mechanism, and it is difficult for external regulatory agencies and third-party audits to deeply understand the contract execution details, unable to timely discover loopholes or illegal operations in contract execution. Summary of the Invention

[0009] Technical Problems to be Solved

[0010] In view of the deficiencies of the prior art, the present invention provides an automated data verification system in carbon asset trading, which solves the following problems:

[0011] 1. The problem of insufficient data privacy protection in the carbon trading process;

[0012] 2. The problem of poor data processing efficiency and accuracy in carbon trading;

[0013] 3. The problem of lack of deep semantic association and difficulty in mining hidden information in complex trading relationships during carbon trading data integration;

[0014] 4. The problem of single functions and difficult maintenance of smart contracts in the carbon trading process.

[0015] Technical Solution

[0016] To achieve the above objectives, the present invention is realized through the following technical solutions: An automated data verification system in carbon asset trading, the system includes a data acquisition layer, a data integration layer, a verification model layer, an interaction and security layer, wherein:

[0017] The data acquisition layer is used to obtain carbon asset trading-related data from multiple sources and perform preliminary processing, including a federated learning and edge computing module, which is used to realize collaborative training of verification models by multiple data holders through the federated learning algorithm and preprocess data at the data source end using edge computing. The federated learning and edge computing module includes a federated learning subunit and an edge computing subunit. The federated learning subunit is used to establish a federated learning coordination server to communicate and coordinate with local nodes of each data provider. Each data provider trains a local model based on local carbon data and encrypts and uploads the model parameters to the coordination server. After the server aggregates the parameters, it issues updates to the local model to iteratively train the global model. The edge computing subunit is used to deploy edge computing devices near the enterprise internal data source to perform real-time preprocessing on the original data, including data compression, format conversion, and preliminary screening. The screened suspicious and abnormal data is transmitted to the data integration layer;

[0018] The data integration layer is used to integrate, correlate, store, and manage the collected data based on semantic web technology, including a semantic web data integration module. The semantic web data integration module includes an ontology construction unit, a semantic mapping and association unit, and a data storage and management unit. The ontology construction unit is used to construct an ontology model in the carbon trading field to define core concepts and their relationships. The semantic mapping and association unit is used to map and match the collected data with the ontology model and perform correlation integration. The data storage and management unit is used to classify and store the integrated data and manage access permissions;

[0019] The verification model layer is used to verify the integrated data using multiple algorithm models, including a fusion LSTM-SVM verification module and a GNN relationship verification module. The fusion LSTM-SVM verification module includes an LSTM time series analysis unit and an SVM classification verification unit. The LSTM time series analysis unit is used to obtain historical carbon data and related time series data to train an LSTM network model to output a time series feature vector and a predicted trend value. The SVM classification verification unit is used to receive the feature vector and judge whether the data is normal based on the trained SVM model. The GNN relationship verification module includes a carbon trading relationship graph construction unit and a GNN analysis and verification unit. The carbon trading relationship graph construction unit is used to construct a carbon trading relationship graph. The GNN analysis and verification unit is used to mine abnormal patterns based on the relationship graph using the GNN algorithm;

[0020] The interaction and security layer is used to ensure the security and trustworthiness of data interaction and automatically execute transaction and verification processes, including a blockchain and smart contract module. The blockchain and smart contract module includes a blockchain data storage and interaction unit and a smart contract execution unit. The blockchain data storage and interaction unit is used to establish a blockchain network to store key data and provide a query interface. The smart contract execution unit is used to write smart contract code and automatically execute carbon asset transaction and verification related operations according to conditions.

[0021] Preferably, in the federated learning sub-unit, each local node of the data provider encrypts the local model parameters using homomorphic encryption technology to ensure that data privacy is not leaked during the parameter upload process, and the coordination server uses a multi-party computing protocol to aggregate the encrypted parameters to achieve secure global model parameter updates.

[0022] Preferably, during the data compression process, the edge computing sub-unit adopts an adaptive compression algorithm according to the time series characteristics and data types of the data. A higher compression ratio is used for the basic data with slow changes, and a lower compression ratio is used for the real-time monitoring data with large fluctuations, minimizing the data transmission volume to the greatest extent while ensuring data validity.

[0023] Preferably, when constructing the ontology model in the carbon trading field, the ontology construction unit adopts a construction method combining top-down and bottom-up based on international common carbon trading standards and industry specifications. First, determine the core concept framework, and then gradually refine the sub-concepts and attributes, and regularly update and maintain the ontology model according to policy and regulation changes and industry development trends.

[0024] Preferably, when performing semantic mapping, the semantic mapping and association unit adopts a mapping method combining rules and machine learning. For common data formats and standard terms, predefined rules are used for rapid mapping, and for fuzzy data expressions, a machine learning model is used for automatic learning and mapping matching.

[0025] Preferably, when training the LSTM network model, the LSTM time series analysis unit uses a sliding window technique to segment historical carbon data to increase the number of training samples, combines the early stopping method to prevent overfitting, and introduces an attention mechanism to enable the model to focus on the time series feature parts that have an important impact on carbon verification.

[0026] Preferably, when constructing the classification hyperplane, the SVM classification and verification unit uses a kernel function technique to map non-linearly separable data, selects the radial basis function as the kernel function, and optimizes the kernel function parameters and penalty factors through a cross-validation method to improve the accuracy of the SVM model in classifying complex carbon data.

[0027] Preferably, when constructing the carbon trading relationship graph, the carbon trading relationship graph construction unit adds basic trading entities, trading relationships, changes in policies and regulations related to carbon trading, and market dynamic information as virtual nodes and edges to the relationship graph to more comprehensively reflect the impact of the carbon trading environment on trading relationships.

[0028] Preferably, when performing information propagation and node feature update, the GNN analysis and verification unit adopts a multi-layer GNN structure. Each layer of GNN uses different message passing functions and aggregation functions, and dynamically adjusts the scope and intensity of information propagation according to the node type and edge weight to more accurately detect abnormal patterns in complex relationships.

[0029] Preferably, when writing the smart contract code, the smart contract execution unit adopts a modular design, encapsulating functions such as carbon data submission, verification process, transaction settlement, and penalty for violations into independent modules respectively, which is convenient for contract maintenance and upgrade. At the same time, an audit mechanism for smart contracts is set up to allow regulatory agencies and third-party audit agencies to supervise and audit the contract execution process.

[0030] Beneficial effects

[0031] The present invention provides an automated data verification system in carbon asset trading. It has the following beneficial effects:

[0032] 1. The system of the present invention adopts the homomorphic encryption technology of the local nodes of the data providers in the federated learning subunit and the multi-party computing protocol of the coordination server. During the entire data collection and model training process, the data privacy of each participating party is strongly protected. Data holders such as enterprises do not need to worry about the leakage of sensitive carbon data information and can rest assured to participate in the construction of the global verification model, ensuring that even in the complex environment of multi-source data collaborative training, the confidentiality and integrity of the data are still maintained, meeting the characteristics of extremely high data security requirements in the carbon trading field, effectively avoiding risks such as the exposure of trade secrets and malicious tampering of carbon data caused by data leakage, and providing a solid support for the stability and trust foundation of the carbon asset trading market.

[0033] 2. The system of the present invention utilizes the edge computing sub-unit in the data acquisition layer to adopt an adaptive compression algorithm, which flexibly adjusts the compression ratio according to the time series characteristics and types of data. For a large amount of basic data with slow changes, a high compression ratio reduces the transmission bandwidth requirements and storage pressure. For real-time monitoring data with large fluctuations, a lower compression ratio ensures that the key details of the data are retained. This makes data transmission more efficient and the data quality undamaged, providing a timely and accurate data basis for the subsequent verification process, accelerating the data flow speed of the entire system. The sliding window technique of the LSTM time series analysis unit increases the number of training samples, combines the early stopping method to prevent overfitting, and the attention mechanism to focus on key features, making the learning of the temporal characteristics of carbon data more accurate. The SVM classification verification unit processes non-linear data through kernel function techniques and optimizes parameters through cross-validation, effectively improving the accuracy of classifying complex carbon data. The synergistic effect of the two makes both the trend prediction and anomaly judgment of carbon data more reliable, greatly reducing the cases of misjudgment and missed judgment, thereby improving the accuracy and reliability of the entire automated data verification system, and providing a strong guarantee for the fairness and justice of carbon asset trading.

[0034] 3. The system of the present invention adopts the ontology construction unit in the data integration layer to construct an ontology model in the carbon trading field by combining the top-down and bottom-up methods, and updates and maintains it according to policies, regulations and industry development dynamics. This enables the system to effectively integrate the multi-source data collected based on a comprehensive and up-to-date conceptual framework, clarify the internal relationships between the data, break the data silo phenomenon, and achieve deep integration at the data semantic level, laying a good foundation for accurate carbon asset accounting and trading verification. The GNN relationship verification module incorporates various key elements when constructing the carbon trading relationship graph, and analyzes through a multi-layer GNN structure and a dynamically adjusted information propagation mechanism. It can not only uncover the potential associations and abnormal patterns between trading entities, but also comprehensively consider the impact of external environmental factors on carbon trading relationships, helping regulatory agencies to grasp the overall health of the carbon trading market from a macro perspective, timely discover systemic risks and illegal trading behaviors, and promote the standardization and sustainable development of the carbon trading market.

[0035] 4. The system of the present invention uses a modular design for the intelligent contract execution unit, and encapsulates the core functions related to carbon data into independent modules. This design method makes the structure of the contract clearer and enhances the independence of the functional modules. When the carbon trading rules or verification processes change, only the corresponding modules need to be modified or upgraded accordingly, without the need for large-scale changes to the entire intelligent contract, greatly reducing the complexity and cost of contract maintenance. At the same time, the set audit mechanism allows regulatory agencies and third-party audit agencies to intervene and supervise and audit the contract execution process, further enhancing the transparency and credibility of the intelligent contract, ensuring that the carbon asset trading and verification processes strictly follow the preset rules, effectively preventing the exploitation of loopholes in the contract execution process, improving the security and credibility of the entire carbon trading system, and promoting the healthy and orderly operation of the carbon asset trading market. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is the system architecture diagram of the present invention;

[0037] Figure 2 is the system data flow diagram of the present invention;

[0038] Figure 3 is the system carbon asset trading step diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to 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 of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Specific Embodiment 1:

[0041] As Figures 1-3 shown, an automated data verification system in carbon asset trading, the system includes a data acquisition layer, a data integration layer, a verification model layer, an interaction and security layer, wherein:

[0042] The data acquisition layer is used to obtain carbon asset trading-related data from multiple sources and perform preliminary processing, including a federated learning and edge computing module, which is used to implement collaborative training of the verification model by multiple data holders through the federated learning algorithm and preprocess data at the data source end using edge computing. The federated learning and edge computing module includes a federated learning subunit and an edge computing subunit;

[0043] The data integration layer is used to integrate, correlate, store and manage the collected data based on semantic web technology, including a semantic web data integration module, and the semantic web data integration module includes an ontology construction unit, a semantic mapping and association unit, and a data storage and management unit;

[0044] The verification model layer is used to verify the integrated data using multiple algorithm models, including a fused LSTM-SVM verification module and a GNN relationship verification module. The fused LSTM-SVM verification module includes an LSTM time series analysis unit and an SVM classification verification unit. The LSTM time series analysis unit is used to obtain historical carbon data and related time series data to train the LSTM network model to output a time series feature vector and a predicted trend value. The SVM classification verification unit is used to receive the feature vector and determine whether the data is normal based on the trained SVM model. The GNN relationship verification module includes a carbon trading relationship graph construction unit and a GNN analysis and verification unit. The carbon trading relationship graph construction unit is used to construct a carbon trading relationship graph. The GNN analysis and verification unit is used to mine abnormal patterns based on the relationship graph using the GNN algorithm;

[0045] The interaction and security layer is used to ensure the security and trustworthiness of data interaction and automate the execution of trading and verification processes, including a blockchain and smart contract module. The blockchain and smart contract module includes a blockchain data storage and interaction unit and a smart contract execution unit. The blockchain data storage and interaction unit is used to establish a blockchain network to store key data and provide a query interface. The smart contract execution unit is used to write smart contract code and automatically execute carbon asset trading and verification related operations according to conditions.

[0046] This automated data verification system can effectively collect and integrate carbon asset trading data, perform accurate verification using advanced algorithm models, and ensure the security and trustworthiness of data interaction through blockchain and smart contracts, providing strong technical support for the healthy and stable development of the carbon asset trading market. Specific Embodiment 2:

[0048] As Figures 1-3 shown, based on the content in the above specific embodiment, the following content is further disclosed:

[0049] For each unit module in each layer of the system, the following further includes:

[0050] The federated learning sub-unit is used to establish a federated learning coordination server to communicate and coordinate with local nodes of each data provider. Each data provider trains a local model based on local carbon data and encrypts and uploads the model parameters to the coordination server. After aggregating the parameters, the server distributes them to update the local model for iterative training of the global model. In the federated learning sub-unit, local nodes of each data provider use homomorphic encryption technology to encrypt local model parameters, ensuring that data privacy is not leaked during the parameter upload process. Moreover, the coordination server uses a multi-party computing protocol to perform aggregation operations on the encrypted parameters to achieve secure global model parameter updates. Through the homomorphic encryption technology of local nodes of data providers and the multi-party computing protocol of the coordination server in the federated learning sub-unit, during the entire data collection and model training process, the data privacy of each participant is strongly protected. Data holders such as enterprises do not need to worry about the leakage of sensitive carbon data information and can rest assured to participate in the construction of the global verification model. This encryption and secure aggregation mechanism ensures that even in a complex environment of multi-source data collaborative training, the confidentiality and integrity of data are still maintained, meeting the extremely high requirements for data security in the carbon trading field, effectively avoiding risks such as the exposure of trade secrets and malicious tampering of carbon data that may be caused by data leakage, and providing a solid support for the stability and trust foundation of the carbon asset trading market.

[0051] The edge computing sub-unit is used to deploy edge computing devices near the internal data sources of enterprises to perform real-time preprocessing on the original data, including data compression, format conversion, and preliminary screening. The screened suspicious and abnormal data is transmitted to the data integration layer. During the data compression process in the edge computing sub-unit, an adaptive compression algorithm is adopted according to the time series characteristics and data types of the data. A higher compression ratio is used for the basic data with slow changes, and a lower compression ratio is used for the real-time monitoring data with large fluctuations. While ensuring the data validity, the data transmission volume is minimized to the greatest extent. The edge computing sub-unit adopts an adaptive compression algorithm and flexibly adjusts the compression ratio according to the time series characteristics and types of the data. For a large amount of basic data with slow changes, the high compression ratio reduces the transmission bandwidth requirements and storage pressure; while for the real-time monitoring data with large fluctuations, the lower compression ratio ensures that the key details of the data are retained, making the data transmission more efficient and the data quality undamaged, providing a timely and accurate data basis for the subsequent verification process and accelerating the data flow speed of the entire system.

[0052] The ontology construction unit is used to construct the core concepts and their interrelationships in the ontology model of the carbon trading field. When constructing the ontology model of the carbon trading field, the ontology construction unit adopts a construction method that combines top-down and bottom-up. First, it determines the core concept framework, then gradually refines the sub-concepts and attributes, and regularly updates and maintains the ontology model according to policy and regulatory changes and industry development trends. The semantic mapping and association unit is used to map, match, and integrate the collected data with the ontology model. When performing semantic mapping, the semantic mapping and association unit adopts a mapping method that combines rules and machine learning. For common data formats and standard terms, predefined rules are used for rapid mapping, while fuzzy data expressions are automatically learned and mapped and matched using a machine learning model. The data storage and management unit is used to classify and store the integrated data and manage access permissions. The ontology construction unit constructs the ontology model of the carbon trading field by combining top-down and bottom-up methods and updates and maintains it according to policies, regulations, and industry development trends. This enables the system to effectively integrate multi-source data collected based on a comprehensive and up-to-date concept framework, clarify the internal relationships between the data, break the data silo phenomenon, and achieve deep integration at the data semantic level, laying a good foundation for accurate carbon asset accounting and trading verification.

[0053] When training the LSTM network model, the LSTM time series analysis unit uses the sliding window technique to segment historical carbon data to increase the number of training samples, combines the early stopping method to prevent overfitting, and introduces an attention mechanism to enable the model to focus on the time series feature parts that have an important impact on carbon verification. When constructing the classification hyperplane, the SVM classification verification unit uses the kernel function technique to map non-linearly separable data, selects the radial basis function as the kernel function, and optimizes the kernel function parameters and penalty factors through the cross-validation method to improve the accuracy of the SVM model in classifying complex carbon data. The sliding window technique of the LSTM time series analysis unit increases the number of training samples, combines the early stopping method to prevent overfitting, and the attention mechanism focuses on key features, making the learning of the temporal features of carbon data more accurate. The SVM classification verification unit processes non-linear data through the kernel function technique and optimizes the parameters through cross-validation, effectively improving the accuracy of classifying complex carbon data. The synergistic effect of the two makes both the trend prediction and anomaly judgment of carbon data more reliable, greatly reducing the cases of misjudgment and missed judgment, thereby improving the accuracy and reliability of the entire automated data verification system and providing a strong guarantee for the fairness and justice of carbon asset trading.

[0054] When constructing a carbon trading relationship graph, the carbon trading relationship graph construction unit adds basic trading entities, trading relationships, changes in policies and regulations related to carbon trading, and market dynamic information as virtual nodes and edges to the relationship graph to more comprehensively reflect the impact of the carbon trading environment on trading relationships. When performing information dissemination and node feature update, the GNN analysis and verification unit adopts a multi-layer GNN structure. Each layer of GNN uses different message passing functions and aggregation functions, and dynamically adjusts the scope and intensity of information dissemination according to node types and edge weights to more accurately detect abnormal patterns in complex relationships. The GNN relationship verification module incorporates various key elements when constructing a carbon trading relationship graph, including trading entities, relationships, changes in policies and regulations, and market dynamic information, etc., and analyzes them through a multi-layer GNN structure and a dynamic information dissemination mechanism. This can not only detect potential associations and abnormal patterns between trading entities, but also comprehensively consider the impact of external environmental factors on carbon trading relationships, helping regulatory agencies to grasp the overall health of the carbon trading market from a macro perspective, timely discover systemic risks and illegal trading behaviors, and promote the standardization and sustainable development of the carbon trading market.

[0055] When writing smart contract code, the smart contract execution unit adopts a modular design, encapsulating functions such as carbon data submission, verification process, transaction settlement, and penalty for violations into independent modules respectively, which is convenient for contract maintenance and upgrade. At the same time, an audit mechanism for smart contracts is set up to allow regulatory agencies and third-party audit agencies to supervise and audit the contract execution process. The smart contract execution unit adopts a modular design and independently encapsulates the core functions related to carbon data into modules. This design method makes the contract structure clearer and the independence of functional modules stronger. When carbon trading rules or verification processes change, only the corresponding modules need to be modified or upgraded targeted, without large-scale modification of the entire smart contract, greatly reducing the complexity and cost of contract maintenance. At the same time, the set audit mechanism allows regulatory agencies and third-party audit agencies to intervene and supervise and audit the contract execution process, further enhancing the transparency and credibility of smart contracts, ensuring that carbon asset transactions and verification processes strictly follow preset rules, effectively preventing the exploitation of loopholes during contract execution, improving the security and credibility of the entire carbon trading system, and promoting the healthy and orderly operation of the carbon asset trading market. Specific Embodiment Three:

[0057] As Figures 1-3 shown, based on the content in the above specific embodiments, the following content is further disclosed:

[0058] In the entire system, the algorithm and model content at each level further includes the following content:

[0059] The formula content of the data collection layer - federated learning and edge computing module is as follows:

[0060] Federated Learning Subunit - Related to Homomorphic Encryption, Principle of Encryption Process: Assume that the local model parameters to be encrypted by the local node of the data provider are \(x\), the homomorphic encryption function is \(E(\cdot)\), and the encrypted parameters are represented as \(E(x)\). The characteristic of homomorphic encryption here is that for certain specific operations, such as addition, multiplication, etc., which are related to model parameter aggregation operations in this scenario, operations can be directly performed on the encrypted data, and the result obtained after decryption is the same as the result of performing the same operation on the plaintext data. For example, if there are two encrypted parameters \(E(x_1)\) and \(E(x_2)\), performing an addition operation gives \(E(x_1)+E(x_2)\), where the addition here is defined in the homomorphic encryption operation system, and the decrypted result should be equal to the result of the plaintext \(x_1 + x_2\).

[0061] Implementation Steps: After the local node of the data provider trains the local model, it obtains the local model parameters \(x\), encrypts \(x\) using the homomorphic encryption function \(E(\cdot)\) to get \(E(x)\), and uploads the encrypted parameters \(E(x)\) to the coordination server.

[0062] Operation Logic: Each data provider independently trains the local model, generates local model parameters, and then uploads the encrypted parameters through homomorphic encryption. After receiving the encrypted parameters, the coordination server performs an aggregation operation using the multi - party computing protocol. Throughout the process, the data is always transmitted and processed in encrypted form until the local node needs to update the local model according to the updates sent by the coordination server, at which point decryption operations are performed. Decryption operations are not elaborated here and can be understood as being carried out in the reverse process of homomorphic encryption.

[0063] Beneficial Effects: Ensure that data privacy is not leaked during the parameter upload process, enabling each data provider to participate in the construction of the global verification model with confidence. Even in a complex environment of multi - source data collaborative training, it can maintain the confidentiality and integrity of the data, effectively avoiding risks such as the exposure of trade secrets and malicious tampering of carbon data that may be caused by data leakage.

[0064] Edge Computing Subunit - Adaptive Compression Algorithm includes the following:

[0065] Let the original data volume be \(D\), the compression ratio be \(r\), and the compressed data volume be \(D'\), then \(D'=D\times r\). For the basic data with slow changes, assume its corresponding compression ratio is \(r_1\), taking a lower value, such as \(r_1 = 0.2\), indicating compression to 20% of the original. For the real - time monitoring data with large fluctuations, assume its corresponding compression ratio is \(r_2\), taking a higher value, such as \(r_2 = 0.5\), indicating compression to 50% of the original;

[0066] Among them:

[0067] D: Represents the size of the original data, that is, the amount of information of the uncompressed original carbon data obtained near the enterprise internal data source. The unit can be bytes, etc.;

[0068] r: Compression ratio, which represents the degree to which the original data is compressed. Its value range is between [0, 1]. The closer it is to 0, the higher the compression degree and the smaller the amount of compressed data;

[0069] D′: Represents the size of the data after compression, which is obtained by multiplying the original data volume by the compression ratio;

[0070] r1 and r2: Respectively, the compression ratios set for different types of data, namely the slowly changing basic data and the real-time monitoring data with large fluctuations. Different compression ratios are selected according to the characteristics of the data to minimize the data transmission volume while ensuring the data validity.

[0071] Implementation steps: First, determine the type of data, whether it is slowly changing basic data or real-time monitoring data with large fluctuations. If it is slowly changing basic data, determine the compression ratio as r1, and then calculate the compressed data volume according to the formula D′ = D × r1. If it is real-time monitoring data with large fluctuations, determine the compression ratio as r2, and then calculate the compressed data volume according to the formula D′ = D × r2, and transmit the compressed data to the data integration layer.

[0072] Operation logic: The edge computing device deployed near the enterprise internal data source will judge the type of the obtained original data in real time, select the corresponding compression ratio according to different types, and then calculate the compressed data volume according to the compression formula and transmit it. In this way, the compression degree can be flexibly adjusted according to the actual situation of the data, which not only reduces the transmission bandwidth requirement and storage pressure (for slowly changing basic data), but also ensures that the key details of the data are retained (for real-time monitoring data with large fluctuations).

[0073] Beneficial effects: Make data transmission more efficient and the data quality undamaged, provide a timely and accurate data basis for the subsequent verification process, and accelerate the data flow speed of the entire system.

[0074] The algorithms and models in the semantic web data integration module of the data integration layer further include the following:

[0075] Ontology construction unit - Ontology model construction:

[0076] Let the set of core concepts in the carbon trading field be C = {c1, c2,..., c3}, and the set of relationships between core concepts be R = {r1, c1 → c2, r2, c2 → c3,...}. Here, the relationship r i represents from concept c ito another concept c i+1 Regarding the relationship, the ontology model O=(C, R) in the carbon trading field;

[0077] Among them:

[0078] C: represents the set of core concepts in the carbon trading field. These core concepts are abstractions and generalizations of things and phenomena related to carbon trading, such as carbon trading entities, carbon emissions, carbon quotas, etc., and are the basic elements for constructing the ontology model;

[0079] R: represents the set of relationships between core concepts. It describes how different core concepts are interrelated. For example, there may be an ownership relationship between a carbon trading entity and its carbon emissions. The relationship here is represented in a similar form, indicating a certain connection from one concept to another concept.

[0080] O: is the ontology model in the carbon trading field. It is an overall structure composed of the set of core concepts C and the set of relationships R. By clarifying these core concepts and their interrelationships, it provides a clear framework for subsequent data integration and semantic association.

[0081] Implementation steps: Based on internationally recognized carbon trading standards and industry norms, first determine the set of core concepts C in the carbon trading field. By analyzing relevant standards and norms, extract key concept elements. Then determine the set of relationships R between core concepts. By analyzing the actual carbon trading business processes and scenarios, find various connections existing between different concepts. Finally, construct the ontology model O=(C, R) in the carbon trading field, and regularly update and maintain the ontology model according to changes in policies and regulations and industry development trends, that is, re-examine and adjust the set of core concepts and the set of relationships.

[0082] Operation logic: First, obtain information from established standards and norms and actual business scenarios to determine core concepts and the relationships between them. Then combine these elements into an ontology model. Over time, due to changes in policies, regulations, and industry development, this ontology model needs to be dynamically updated to ensure that it can accurately reflect the current situation in the carbon trading field, thereby providing an effective framework for subsequent operations such as data integration.

[0083] Beneficial effects: Enable the system to effectively integrate multi-source data collected based on a comprehensive and up-to-date conceptual framework, clarify the internal connections between data, break the data silo phenomenon, achieve deep integration at the data semantic level, and lay a good foundation for accurate carbon asset accounting and trading verification.

[0084] Semantic mapping and association unit - a mapping method combining rules and machine learning (not an atypical mathematical formula, mainly based on method description)

[0085] Let the collected data set be D = {d1, d2,..., d m}, the ontology model be O = (C, R) (defined previously), and the mapping result set be M = {m1, m2,..., m m}. For common data formats and standard terms, use predefined rules for mapping. Let the predefined rule function be f rule (d i ). Then for this type of data, the mapping result g ml = f rule (d i ) (i = 1, 2,..., m). For fuzzy data representations, let the machine learning model be g ml (d i ). Then the mapping result m i = g ml (d i ) (i = 1, 2,..., m);

[0086] Where:

[0087] D: Represents the set of carbon trading-related data collected from the data acquisition layer. These data may have different formats and expressions and need to be mapped and matched with the ontology model;

[0088] O: Ontology model in the carbon trading field, which is the previously constructed model for providing a conceptual framework and relationship description and serves as the target framework for mapping;

[0089] M: Mapping result set, which records the results of mapping each data element in the collected data set D to the ontology model O, obtained through different mapping methods (rule-based or machine learning-based);

[0090] f rule (d i ): Predefined rule function. For common data formats and standard terms, map the data according to the predefined rules, match it with the corresponding elements in the ontology model, and obtain the mapping result m i ;

[0091] g ml (d i ): Machine learning model. For fuzzy data representations, use the machine learning model to automatically learn and map-match the data d i and obtain the mapping result m i .

[0092] Implementation steps: For each data element d i in the collected data set D, first determine whether it belongs to common data formats and standard terms. If so, use the predefined rule function f rule (di ) Perform mapping to obtain the mapping result m i , and add it to the mapping result set M. If not, use the machine learning model g ml (d i ) Perform mapping to obtain the mapping result m i , and add it to the mapping result set M. Finally, obtain the complete mapping result set M, completing the mapping matching and correlation integration of the collected data and the ontology model.

[0093] Operation logic: For each data element collected, according to its data format and expression characteristics, select an appropriate mapping method (rule-based or machine learning-based) to perform mapping operations, match it with the corresponding elements in the ontology model, so as to achieve effective mapping and correlation integration of the data and the ontology model for subsequent data storage and management operations.

[0094] Beneficial effects: It can accurately map and correlate the collected data with the ontology model and integrate them. Whether it is common data formats and standard terms, or newly emerging or ambiguous data expressions, better mapping effects can be obtained, thus providing an accurate basis for subsequent data storage and management, and further promoting the in-depth integration at the data semantic level.

[0095] Verification model layer - Fusion LSTM-SVM verification module

[0096] The LSTM time series analysis unit - The further training of the LSTM network model includes the following:

[0097] The core formula of LSTM mainly involves several gating mechanisms inside it, including the input gate i t , the forget gate f t , the output gate o t and the cell state C t ;

[0098] Forget gate: f t =σ(W f ·[h t-1 , x t +b f );

[0099] Among them, σ is the sigmoid function, W f is the weight matrix of the forget gate, h t-1 is the hidden state of the previous moment, x t is the input data of the current moment, b f is the bias vector of the forget gate;

[0100] Input gate: i t =σ(W i ·[ht-1 , x t + b i );

[0101] Among them, the meanings of the parameters are similar to those of the forget gate. W i is the weight matrix of the input gate, and b i is the bias vector of the input gate;

[0102] Cell state update: C t = f t · C t-1 + i t · tanh(W c · [h t-1 , x t + b c );

[0103] Among them, tanh is the hyperbolic tangent function. W c is the weight matrix for updating the cell state, and b c is its bias vector;

[0104] Output gate: o t = σ(W o · [h t-1 , x t + b o );

[0105] Among them, the meanings of the parameters are similar to those of the forget gate. W o is the weight matrix of the output gate, and b o is the bias vector of the output gate;

[0106] Final hidden state: h t = o t · tanh(C t );

[0107] Among them:

[0108] σ: sigmoid function, and its formula is It maps the input value to the interval (0, 1), controls the flow and selection of information, and plays a key role in each gating mechanism of the LSTM, such as determining which information should be forgotten (forget gate), which information should be input (input gate), etc.;

[0109] W f , W i , W c , W o: They are the weight matrices of the forget gate, input gate, cell state update, and output gate respectively. These weight matrices determine the influence degree of the input data and the previous hidden state on each gating mechanism and cell state update. During the training process, the values of these weight matrices are continuously adjusted to enable the LSTM model to better learn the features of the data;

[0110] b f and b i and b c and b o : They are the bias vectors of the forget gate, input gate, cell state update, and output gate respectively. Together with the weight matrices, they affect the calculation results of each gating mechanism. Similarly, their values will be continuously adjusted during the training process;

[0111] h t-1 : The previous hidden state, which reflects the learning and processing results of the model on the data at the previous moment, is an important reference for calculating each gating mechanism and cell state update at the current moment;

[0112] x t : The input data at the current moment, that is, the data point at the current moment obtained from the historical carbon data and related time series data, is the object for the LSTM model to learn and process;

[0113] C t : The cell state, which is an important state variable inside the LSTM model, records some key information of the model at different moments, and is continuously updated through mechanisms such as the forget gate and input gate to adapt to the changes in the data and learn the features of the data;

[0114] h t : The final hidden state, which is the output result of the model's processing of the data at the current moment, and is also a reference for calculating each gating mechanism and cell state update at the next moment. By continuously iteratively calculating the hidden states at each moment, the LSTM model can learn the time series features of the data.

[0115] Implementation steps:

[0116] First, obtain the historical carbon data and related time series data as the input data x t , and initialize the weight matrices W f , W i , W c , W o , and the bias vectors b f , b i , b c , b o , as well as the initial hidden state h t-1 . At each moment t, calculate the forget gate f in sequence according to the above formulat , input gate i t , cell state update C t , output gate o t and final hidden state h t , during the calculation process, the historical carbon data is segmented through the sliding window technique to increase the number of training samples. For example, assuming the original historical carbon data sequence is x1, x2, …, x N , with the sliding window size of w, multiple training samples can be obtained, such as [x1, x2, …, x w , [x2, x3, …, x w+1 , etc. Combining with the early stopping method to prevent overfitting, that is, during the training process, when the loss function on the validation set no longer decreases or the decrease amplitude is very small (lower than a certain set threshold), stop the training.

[0117] During the training process, an attention mechanism is introduced, and the calculation of the attention mechanism is as follows:

[0118] Let the hidden state sequence be H = [h1, h2, …, h T (T is the sequence length), and the attention weights are calculated as

[0119]

[0120] where, e t = v T tanh(W h h t + b h ), v, W h , b h are trainable parameters. Through the attention mechanism, the model can focus on the time series feature part that has an important impact on carbon verification, and obtain the weighted hidden state representation:

[0121]

[0122] Train the LSTM network model until the training stop condition is met (such as reaching the set number of training epochs or the early stopping method is triggered), and obtain the trained LSTM model, which can output the time series feature vector and the predicted trend value.

[0123] Operating logic: At each time step, the LSTM, based on the input data x t and the hidden state h t-1 at the previous moment, updates the cell state C t through the gating mechanism, and calculates the hidden state h t at the current moment. The sliding window technique enables the model to utilize more historical data information for learning, increasing the diversity and quantity of training samples. The attention mechanism calculates the attention weights αt weighted summation is performed on the hidden states h at different times, highlighting the characteristic information important for carbon verification, enabling the model to better capture the key patterns and trends in the data. Finally, the trained LSTM model can output a vector reflecting the time series characteristics of the data and a predicted value of the future carbon data trend based on the input historical carbon data sequence, providing valuable input for subsequent SVM classification verification. t Beneficial effects: The sliding window technique of the LSTM time series analysis unit increases the number of training samples. Combining the early stopping method to prevent overfitting and the attention mechanism to focus on key features makes the learning of the time series characteristics of carbon data more accurate. This helps to more accurately grasp the variation laws and trends of carbon data in the time dimension, providing a reliable basis for subsequent judgment of whether the data is abnormal, and improving the accuracy and reliability of the entire automated data verification system for the analysis of the time series characteristics of carbon data.

[0124] SVM Classification Verification Unit - The construction and classification of the SVM model are as follows:

[0125] For the linearly separable case, the decision function of SVM is

[0126] f(x) = sign(w

[0127] x + b) T

[0128] where w is the weight vector, x is the input feature vector (here it is the time series feature vector output by LSTM), and b is the bias term;

[0129] The goal is to find the optimal w and b such that, under the constraint conditions, is minimized, and the constraint condition is that for the training samples (x i , y i ), y i is the class label of the sample, such as normal or abnormal:

[0130] y i (w T x i + b) ≥ 1 (i = 1, 2,..., n)

[0131] For the non-linearly separable case, the kernel function technique is adopted to map the input data to a high-dimensional space. Let the kernel function be

[0132]

[0133] where is the mapping function;

[0134] The commonly used radial basis function RBF kernel function is

[0135] K(x​i , x j ) = exp(-γ||x i - x j || 2 )

[0136] where γ is the kernel parameter, and the decision function becomes

[0137]

[0138] where α i is the Lagrange multiplier, obtained by solving a quadratic programming problem;

[0139] Explanation:

[0140] w: The weight vector, which determines the direction of the decision boundary in the feature space. Its value is adjusted during the training process so that the decision boundary can correctly separate samples of different classes. In the case of non-linearly separable data, although the data cannot be linearly divided in the original space, after mapping the data to a high-dimensional space through the kernel function, w plays a similar separation role in the high-dimensional space;

[0141] x: The input feature vector, which is the temporal feature vector output by the LSTM time series analysis unit here, containing the time series feature information of the carbon data. The SVM judges the class to which the data belongs based on these feature vectors;

[0142] b: The bias term, which together with the weight vector determines the position of the decision boundary in the feature space and has an important impact on the classification result. It is also optimized and adjusted during the training process;

[0143] y i : The class label of the sample, used to indicate whether the carbon data sample is normal or abnormal, which is the supervision information for SVM model training. The model parameters are optimized by making the classification result match these labels as much as possible.

[0144] K(x i , x j ) : The kernel function. For non-linearly separable data, it maps the data points in the original space to a high-dimensional feature space, making the data potentially linearly separable in the high-dimensional space. Different kernel functions have different mapping methods. The RBF kernel function is a commonly used kernel function. The greater the complexity of the feature space after mapping controlled by the parameter, the stronger the model's fitting ability to the training data, but it may also lead to overfitting, and optimization is required through methods such as cross-validation;

[0145] α i:Lagrange multiplier, a variable introduced when solving the SVM optimization problem (whether linear or non - linear cases). Its value is determined by solving the quadratic programming problem, and then the values of the weight vector and the bias term are determined.

[0146] Implementation steps: Receive the feature vector x output by the LSTM time - series analysis unit. For linearly separable data, directly solve for the weight vector w and the bias term b according to the optimization objective and constraint conditions of the linear SVM, and construct the decision function f(x) = sign(w T x + b). For non - linearly separable data, select the radial basis function as the kernel function, and optimize the kernel function parameter γ and the penalty factor (related to the constraint conditions in solving the quadratic programming problem) through the cross - validation method. The specific approach is to divide the training data into a training set and a validation set, train the SVM model under different combinations of γ and penalty factor values, and evaluate the performance of the model (such as accuracy, recall, etc.) on the validation set. Select the parameter combination with the best performance, and then construct the decision function Classify the input feature vector according to the constructed decision function to determine whether the data is normal.

[0147] Operating logic: The SVM classification verification unit takes the feature vector output by the LSTM as input. First, it determines whether the data is linearly separable. If so, it solves for the optimal weight vector and bias term according to the principle of the linear SVM and constructs a decision function for classification. If the data is non - linearly separable, it uses the kernel function to map the data to a high - dimensional space, optimizes the kernel function parameter and the penalty factor through cross - validation, and constructs a decision function based on the kernel function for classification. In the whole process, the SVM model divides the feature vector into normal and abnormal categories by finding the optimal decision boundary, thus realizing the judgment of whether the carbon data is normal.

[0148] Beneficial effects: The SVM classification verification unit processes non - linear data through the kernel function technique and optimizes the parameters through cross - validation, effectively improving the accuracy of classifying complex carbon data. It can accurately determine whether there are abnormalities in the carbon data based on the time - series feature vector extracted by the LSTM. Acting in cooperation with the LSTM time - series analysis unit, it is more reliable for both trend prediction and abnormality judgment of carbon data, greatly reducing the cases of misjudgment and missed judgment, further improving the accuracy and reliability of the entire automated data verification system, and providing a strong guarantee for the fairness and justice of carbon asset trading.

[0149] Verification model layer - GNN relationship verification module

[0150] Carbon trading relationship graph construction unit - Carbon trading relationship graph construction

[0151] Let the carbon trading relationship graph be G=(V, E),

[0152] Among them, the set of trading entities is V s ={v s1 , v s2 , …, v sn};

[0153] The set of trading relationships is E r ={e r1 , e r2 , …, e rm};

[0154] The set of policy and regulation changes and market dynamic information is V d ={v d1 , v d2 , …, v dk}, then the node set V = V s ∪V d ;

[0155] The edge set E includes the trading relationship edges between trading entities and the association edges between trading entities and policy and regulation changes, market dynamic information. For example, if the trading entity v s1 is affected by the policy and regulation change v d1 , then there is an edge e = (v s1 , v d1 ) ∈ E);

[0156] Wherein:

[0157] G: Carbon trading relationship graph, which is a graph structure used to represent the relationships between various elements in carbon trading, and describes the complex environment of carbon trading through nodes and edges;

[0158] V s : Set of trading entities, which includes various entities participating in carbon trading, such as enterprises, emission trading institutions, etc. These entities are the core participants in carbon trading and are important node types for constructing the relationship graph;

[0159] E r : Set of trading relationships, which describes various trading relationships between trading entities, such as the buying and selling relationships of carbon quotas, the transfer relationships of carbon credits, etc. These relationships connect the corresponding trading entity nodes through edges;

[0160] V d : Set of policy and regulation changes and market dynamic information, which also includes policy and regulation changes and market dynamic information related to carbon trading as nodes in the relationship graph. For example, newly introduced carbon emission standard policies, carbon trading market price fluctuation information, etc. These nodes can reflect the impact of external environmental factors on carbon trading relationships;

[0161] V: The set of nodes, which is formed by merging the set of trading entity nodes and the set of nodes for policy and regulation changes and market dynamics information, covering all node types in the carbon trading relationship graph;

[0162] E: The set of edges, which not only includes the trading relationship edges between trading entities but also the association edges between trading entities and policy and regulation changes and market dynamics information. Through these edges, a complex relationship network among various elements in carbon trading is completely constructed.

[0163] Implementation steps:

[0164] Collect the information of trading entities in carbon trading and construct the set of trading entities V s , determine the trading relationships between trading entities and construct the set of trading relationships E r , collect the policy and regulation changes and market dynamics information related to carbon trading and construct the set of policy and regulation changes and market dynamics information V d , construct the set of nodes V = V s ∪V d and the set of edges E, add the trading relationship edges between trading entities and the association edges between trading entities and policy and regulation changes and market dynamics information to the relationship graph to form the carbon trading relationship graph G = (V, E).

[0165] Operation logic: First, collect and organize the trading entities, trading relationships, and external policy and regulation changes and market dynamics information in carbon trading respectively. Then, regard the trading entities as one type of nodes and the policy and regulation changes and market dynamics information as another type of nodes, and construct edges according to the trading relationships and their association relationships, thus forming a complete carbon trading relationship graph. This relationship graph can intuitively display the mutual relationships among various elements in carbon trading and provides a data structure basis for subsequent mining of abnormal patterns using the GNN algorithm.

[0166] Beneficial effects: Adding the basic trading entities, trading relationships, policy and regulation changes related to carbon trading, and market dynamics information as virtual nodes and edges to the relationship graph can more comprehensively reflect the impact of the carbon trading environment on trading relationships, enabling not only the internal relationships between trading entities to be considered but also external environmental factors to be comprehensively considered during the mining of abnormal patterns. This helps regulatory agencies grasp the overall health status of the carbon trading market from a macro level, promptly discover systemic risks and illegal trading behaviors, and promote the standardization and sustainable development of the carbon trading market.

[0167] GNN Analysis and Verification Unit - Application of the GNN Algorithm in Abnormal Pattern Mining

[0168] Let the node feature matrix of the l-th layer GNN be H (l) (each row represents the feature vector of a node), and the message passing function be M(l) (·), the aggregation function is A (l) (·), then the node feature update formula is:

[0169]

[0170] Among them, N(v i ) represents the set of neighbor nodes of node v i , σ is the activation function, such as the ReLU function, etc.;

[0171] During the information propagation process, the message passing function can be expressed as:

[0172]

[0173] Among them, W (l) is the trainable weight matrix of the l-th layer, are the feature vectors of nodes v i , v j at the l-th layer respectively, e ij is the feature vector of the edge connecting nodes v i and v j ;

[0174] The aggregation function can adopt methods such as sum aggregation, average aggregation, etc. For example, when using sum aggregation:

[0175]

[0176] Among them:

[0177] H (l) : The node feature matrix of the l-th layer GNN. In each layer of GNN, the features of nodes are updated according to the information of their neighbor nodes, and this matrix records the feature states of nodes in each layer;

[0178] M (l) (·): The message passing function, which generates the message passed to the node according to the node's own features, the features of neighbor nodes, and the features of the edges connecting them. Through the trainable weight matrix W (l) combines and transforms these features, enabling the node to obtain relevant information of neighbor nodes and update features based on this information;

[0179] A (l) (·): The aggregation function, which is used to aggregate multiple messages received from neighbor nodes. Different aggregation methods (such as sum, average, etc.) will affect the comprehensive processing method of node neighbor information. Through the aggregation function, the node integrates the information of neighbor nodes into a single information representation for subsequent feature update.

[0180] N(vi ):The set of neighbor nodes of node v i , which defines the nodes from which node v i will receive information during the information propagation process, reflecting the local connection relationship between nodes in the graph structure;

[0181] σ: Activation function, such as the ReLU function ReLU(x) = max(0, x), which performs a non-linear transformation on the aggregated and updated node features, increasing the expressive power of the model and enabling the GNN to learn more complex relationship patterns;

[0182] W (l) : The trainable weight matrix of the l-th layer. During the training process of the GNN, by adjusting the values of these weight matrices, the message passing function can better capture the feature information of nodes and edges, thereby learning the abnormal patterns in the carbon trading relationship graph.

[0183] are the feature vectors of nodes v i and v j at the l-th layer respectively. They are updated according to the message passing and aggregation operations in each layer of the GNN, reflecting the changes in the feature representations of nodes at different layers;

[0184] e ij : The feature vector of the edge connecting node v i and v j . It participates in the message passing process together with the node features, enabling the information of the edge to also affect the update of the node features. In the carbon trading relationship graph, the edge may contain information such as the type and intensity of the trading relationship, which may play an important role in mining abnormal patterns.

[0185] Implementation steps: Initialize the node feature matrix H (0) of each layer of the GNN, which can be set according to the initial attribute information of the nodes in the carbon trading relationship graph;

[0186] For each layer l = 0, 1,..., L - 1, where L is the set number of layers of the GNN, for each node v i , according to its set of neighbor nodes N(v i ), use the message passing function M (l) (·) to calculate the message received from the neighbor nodes

[0187] Use the aggregation function A (l) (·) to aggregate the messages to obtain the aggregated information According to the node feature update formula Update the node feature matrix H(l+1) ;

[0188] After L layers of information propagation and node feature update, based on the final node feature matrix H (L) judge the abnormal pattern. For example, some anomaly detection rules can be set, such as the node feature value exceeding a certain threshold range, the relationship pattern between nodes being significantly different from the normal pattern, etc. According to these rules, judge whether there is an abnormal pattern in the carbon trading relationship graph.

[0189] Operation logic: In the GNN analysis and verification unit, starting from the initial node feature matrix of the carbon trading relationship graph, in each layer of the GNN, each node collects information from its neighbor nodes through the message passing function and uses the aggregation function to integrate this information. Then, after the non-linear transformation of the activation function, the features of the node itself are updated. As the number of GNN layers increases, the node can gradually obtain the information of multi-hop neighbors, thereby learning more extensive graph structure information and potential relationship patterns. Through the multi-layer GNN structure and different message passing functions and aggregation functions are used in different layers, and the scope and intensity of information propagation are dynamically adjusted according to the node type and the weight of the edge, the model can more accurately mine the abnormal patterns in complex relationships. For example, for trading entity nodes and policy and regulation change nodes, different message passing and aggregation strategies may be adopted to adapt to their different roles and importance in the carbon trading relationship graph. At the same time, the weight of the edge can reflect the tightness of the trading relationship or the magnitude of the impact of policies and regulations on trading entities, thereby affecting the effect of information propagation.

[0190] Beneficial effects: Adopting a multi-layer GNN structure, different message passing functions and aggregation functions are used in each layer of the GNN, and the scope and intensity of information propagation are dynamically adjusted according to the node type and the weight of the edge, which can more accurately mine the abnormal patterns in complex relationships. It can not only discover the potential associations and abnormal patterns between trading entities, but also comprehensively consider the impact of external environmental factors (such as policy and regulation changes, market dynamic information) on carbon trading relationships. This helps regulatory agencies grasp the overall health status of the carbon trading market from a macro level, timely discover systemic risks and illegal trading behaviors, and promote the standardization and sustainable development of the carbon trading market, providing strong technical support for the supervision and risk prevention and control of carbon asset trading.

[0191] Interaction and Security Layer - Blockchain and Smart Contract Module

[0192] The blockchain data storage and interaction unit - the blockchain network data processing further includes the following content:

[0193] T = {t1, t2,..., t n} and the blocks in the blockchain are B = {b1, b2,..., b m}, each block contains a block header H = {h1, h2,..., h m} and a block body C = {c1, c2,..., c m};

[0194] The block body stores transaction data, that is indicating that the data in the i-th block body is a subset of the transaction data set T;

[0195] The block header contains information such as the hash value of the previous block, h i = Hash(b i-1 ), and this chained structure ensures the immutability and integrity of the data;

[0196] In terms of data interaction, let the query request be Q and the query result be R, then there exists a query function R = Query(Q, B), indicating that the result R is obtained by querying in the blockchain B according to the query request Q;

[0197] Where:

[0198] T: The set of carbon asset transaction data, which contains various carbon trading-related information, such as the information of the trading parties, the trading amount, the carbon quota quantity, etc. These data are the core data that need to be stored on the blockchain to ensure their security and traceability.

[0199] B: The set of blocks in the blockchain. The blockchain is a chained structure composed of a series of blocks, and each block contains a certain amount of transaction data and related block header information.

[0200] H: The set of block headers. Some key information is stored in the block headers, such as the hash value of the previous block. The hash function is one-way and collision-resistant. The hash value obtained by performing a hash operation on the information of the previous block is stored in the current block header, making the blocks in the blockchain form a chained structure. Once the data in a certain block is tampered with, the hash values of the subsequent blocks will not match, thus ensuring the immutability and integrity of the data.

[0201] C: The set of block bodies, which is used to store transaction data. The data in each block body is a part of the transaction data set. Storing the transaction data in different block bodies facilitates data management and query, and also ensures data security.

[0202] Hash(·): The hash function, which converts the input data (such as block data) into a hash value of a fixed length. In the blockchain, the hash function plays a key role, such as generating the hash value of the previous block in the block header and performing a hash process on the transaction data to verify data integrity.

[0203] Q and R: respectively represent the query request and the query result. In the blockchain data storage and interaction unit, a query interface needs to be provided so that users or other systems can query the carbon trading data stored on the blockchain. The query function R = Query(Q, B) means that according to the given query request Q, data retrieval and processing are performed in the blockchain B, and finally the query result R is obtained.

[0204] Implementation steps:

[0205] When new carbon asset trading data is generated, it is organized into a trading data element t in the trading data set T i , the trading data is packaged into a block body C according to certain rules, and at the same time, the information in the block header H is calculated, such as the hash value of the previous block, etc. The generated block is added to the blockchain B to form a new blockchain state. When a query request Q is received, according to the query function R = Query(Q, B), data retrieval and processing are performed in the blockchain, the trading data related to the query request is found, and it is organized into the query result R and returned to the querying party.

[0206] Operation logic: The new carbon trading data is first organized into a trading data set, and then packaged into a block together with the block header information and added to the blockchain. During the data storage process, the hash function is used to ensure the chained structure between blocks and the immutability of data. During data interaction, according to the query request, each block in the blockchain is traversed to find the required trading data and return the result. Throughout the process, the blockchain network, as a distributed ledger, ensures the security, traceability, and immutability of carbon trading data, providing a reliable data storage and interaction foundation for carbon asset trading.

[0207] Beneficial effects: Establishing a blockchain network to store key data and providing a query interface ensures the security and traceability of carbon asset trading data. The immutability of data effectively prevents trading data from being maliciously tampered with or forged, providing a guarantee for the fairness and justice of carbon trading. At the same time, the traceability enables the trading history to be clearly queried and audited, which helps regulatory agencies to supervise the carbon trading market and conduct compliance inspections, promoting the healthy and orderly development of the carbon trading market.

[0208] Smart contract execution unit - Realization of smart contract functions

[0209] Let the carbon data submission function be SubmitData(d), where d is the submitted carbon data;

[0210] The verification process function is VerifyProcess(d);

[0211] The transaction settlement function is SettleTransaction(t), where t is the transaction information;

[0212] The violation punishment function is PunishViolation(v), where v is the violation information;

[0213] The smart contract can be expressed as:

[0214] Cointract

[0215] ={SubmitData, VerifyProcess, SettleTransaction, PunishViolation}

[0216] If Condition(d, t, v) holds, the corresponding function is executed. For example, if Condition(d, t, v) indicates that the carbon data d passes the verification and the transaction information t is legal, then the transaction settlement function SettleTransaction(t) is executed.

[0217] Where:

[0218] SubmitData(d): The carbon data submission function, which is used to receive the carbon data d submitted externally. These data may come from the data acquisition layer or other relevant systems and are the basic data sources for the subsequent processing of the smart contract;

[0219] VerifyProcess(d): The verification process function, which performs verification operations on the submitted carbon data d. It may involve the invocation of various verification algorithms and models in the aforementioned verification model layer to determine whether the carbon data meets the requirements, such as data authenticity, accuracy, integrity, etc.

[0220] SettleTransaction(t): The transaction settlement function, which is executed when the carbon data passes the verification and the transaction information t is legal. It is used to handle the settlement operations of carbon asset transactions, such as fund transfer, carbon quota transfer, etc., to ensure the smooth completion of the transaction;

[0221] PunishViolation(v): The violation punishment function, which is executed when violation information v is detected, such as carbon data fraud, illegal transactions, etc. It imposes corresponding penalties on the violator according to the preset violation punishment rules, such as fines, trading restrictions, etc., to maintain the order of the carbon trading market;

[0222] Contract: The smart contract, which is a collection composed of the above several core functional functions and defines the main operation logic in the carbon asset transaction and verification process.

[0223] Condition(d, t, v): A conditional function used to determine whether the conditions for executing a certain smart contract function are met. These conditions are usually related to carbon data, transaction information, and violation information. By judging these information, the execution of the corresponding smart contract function is triggered to ensure that the execution of the smart contract conforms to the preset rules and logic.

[0224] Implementation steps:

[0225] The external system calls the carbon data submission SubmitData(d) to submit the carbon data d to the smart contract. The smart contract internally calls the verification process function VerifyProcess(d) to verify the submitted carbon data. According to the verification results, as well as the transaction information t and violation information v, it is judged whether the conditional function Condition(d, t, v) holds. If the conditions are met, for example, for the transaction settlement function, when the carbon data is verified and the transaction information is legal, the transaction settlement function settleTransaction(t) is executed; for the violation penalty function, when violation information is detected, the violation penalty function PunishViolation(v) is executed.

[0226] Operation logic: The smart contract runs according to the preset rules and logic. First, it receives the carbon data submitted externally, and then conducts verification. According to the verification results, as well as the situation of transaction information and violation information, it is judged whether the conditions for executing a specific function are met. If the conditions are met, the corresponding function is executed to complete operations such as carbon data processing, transaction settlement, or violation penalty. The whole process is automatically executed without manual intervention, ensuring the efficiency and accuracy of the carbon asset trading and verification process.

[0227] Beneficial effects: Write smart contract code and automatically execute carbon asset trading and verification related operations according to conditions. Adopt modular design to encapsulate functions such as carbon data submission, verification process, transaction settlement, and violation penalty into independent modules respectively, which is convenient for contract maintenance and upgrade. When the carbon trading rules or verification process change, only the corresponding modules need to be modified or upgraded targeted, rather than making large-scale changes to the entire smart contract, greatly reducing the complexity and cost of contract maintenance. At the same time, set up an audit mechanism for the smart contract, allowing regulatory agencies and third-party audit agencies to supervise and audit the contract execution process, further enhancing the transparency and credibility of the smart contract, ensuring that the carbon asset trading and verification process strictly follows the preset rules, effectively preventing the exploitation of loopholes in the contract execution process, improving the security and credibility of the entire carbon trading system, and promoting the healthy and orderly operation of the carbon asset trading market. Specific embodiment four:

[0229] As Figures 1-3 shown, based on the content in the above specific embodiments, the following content is further disclosed:

[0230] The operating steps in the whole system are as follows:

[0231] The data transfer trends in each layer of the whole system are as follows:

[0232] Sp1: Data collection: The data collection layer operates. The federated learning sub-unit prompts each data provider to train a local model and encrypt and upload the parameters to the coordination server, and the server aggregates and updates to train the global model; the edge computing sub-unit performs preprocessing such as adaptive compression on the original data at the data source end and transmits the suspicious abnormal data to the data integration layer;

[0233] Sp2: Data integration: After receiving the data, the ontology construction unit in the data integration layer constructs a carbon trading ontology model; the semantic mapping and association unit integrates the data according to rules and machine learning methods; the data storage and management unit stores and manages data permissions;

[0234] Sp3: Data verification: The verification model layer works. In the integrated LSTM-SVM verification module, the LSTM unit analyzes historical data to obtain feature vectors, and the SVM unit judges whether the data is normal based on this; in the GNN relationship verification module, first construct a carbon trading relationship graph, and then the GNN unit mines abnormal patterns;

[0235] Sp4: Interaction and execution: The interaction and security layer acts. The blockchain data storage and interaction unit stores the verification key data and provides queries; the smart contract execution unit writes code according to the data and automatically executes trading and verification operations and outputs the results.

[0236] Through these four steps, the system completes the entire process of carbon asset trading data processing, ensures the accuracy and security of trading data verification, and promotes the orderly progress of transactions.

[0237] Sp1: Data collection layer: Multi-source data acquisition and preliminary processing:

[0238] The data collection layer of the system first obtains carbon asset trading-related data from multiple sources. These data sources are diverse and cover all aspects related to carbon asset trading;

[0239] The federated learning and edge computing modules in this layer play important roles. Among them, the federated learning sub-unit establishes a federated learning coordination server, communicates and coordinates with the local nodes of each data provider. Each data provider trains a local model based on local carbon data, encrypts the local model parameters using homomorphic encryption technology, and then uploads the encrypted model parameters to the coordination server. The coordination server then performs an aggregation operation on the encrypted parameters using a multi-party computing protocol to update the local model to iteratively train the global model.

[0240] Meanwhile, the edge computing sub-unit deploys edge computing devices near the enterprise's internal data sources to perform real-time preprocessing on the raw data. During the preprocessing, an adaptive compression algorithm is adopted according to the time series characteristics and data types of the data. For example, a higher compression ratio is used for the slowly changing basic data, and a lower compression ratio is used for the real-time monitoring data with large fluctuations. After preprocessing, the data undergoes data compression, format conversion, and preliminary screening, and the screened suspicious abnormal data is transmitted to the data integration layer.

[0241] Sp2: Data integration layer: Integration, association, storage, and management based on semantic web technology:

[0242] After receiving the suspicious abnormal data screened by the data acquisition layer, the data integration layer processes it through the semantic web data integration module;

[0243] First, the ontology construction unit constructs an ontology model in the carbon trading field based on internationally common carbon trading standards and industry norms, adopting a construction method that combines top-down and bottom-up. First, determine the core concept framework, and then gradually refine the sub-concepts and attributes, and regularly update and maintain the ontology model according to policy and regulatory changes and industry development trends;

[0244] Next, the semantic mapping and association unit maps and matches the collected data with the ontology model and associates and integrates them. When performing semantic mapping, a mapping method that combines rules and machine learning is adopted. For common data formats and standard terms, predefined rules are used for rapid mapping, and fuzzy data expressions are automatically learned and mapped and matched using machine learning models;

[0245] Finally, the data storage and management unit stores the integrated data in categories and manages access permissions, providing an orderly and manageable data basis for subsequent verification processes.

[0246] Sp3: Verification model layer: Conduct verification using multiple algorithm models:

[0247] The verification model layer conducts verification work on the integrated data, mainly through the fusion of the LSTM-SVM verification module and the GNN relationship verification module;

[0248] In the LSTM-SVM verification module, the LSTM time series analysis unit first obtains historical carbon data and related time series data, and uses the sliding window technology to segment the historical carbon data to increase the number of training samples. The early stopping method is combined to prevent overfitting, and the attention mechanism is introduced to enable the model to focus on the time series feature parts that have an important impact on carbon verification. Then, the LSTM network model is trained to output the time series feature vector and the predicted trend value. Then, the SVM classification verification unit receives the feature vector, uses the kernel function technique, selects the radial basis function as the kernel function, maps the nonlinearly separable data, and optimizes the kernel function parameters and penalty factors through the cross-validation method. The trained SVM model is used to determine whether the data is normal.

[0249] In terms of the GNN relationship verification module, the carbon trading relationship graph construction unit adds basic trading entities, trading relationships, carbon trading-related policy and regulatory changes, and market dynamic information as virtual nodes and edges into the relationship graph to construct a carbon trading relationship graph. Afterwards, the GNN analysis and verification unit uses a multi-layer GNN structure based on this relationship graph. Each layer of GNN uses different message passing functions and aggregation functions, and dynamically adjusts the scope and intensity of information dissemination according to the node type and edge weight, so as to more accurately mine abnormal patterns in complex relationships.

[0250] Sp4: Interaction and security layer: Ensure data interaction is secure and reliable and execute transaction and verification processes:

[0251] The blockchain and smart contract modules in the interaction and security layer are responsible for ensuring the data interaction security of the entire system and automating the transaction and verification process;

[0252] The blockchain data storage and interaction unit establishes a blockchain network to store key data and provides a query interface to ensure data storage security and queryability;

[0253] The smart contract execution unit adopts a modular design, encapsulating functions such as carbon data submission, verification process, transaction settlement, and violation penalties into independent modules. When the carbon trading rules or verification process change, only the corresponding modules need to be modified or upgraded in a targeted manner. At the same time, an audit mechanism for smart contracts is set up to allow regulators and third-party auditing agencies to supervise and audit the contract execution process, ensure that the carbon asset trading and verification process strictly follows the preset rules, ensure the security and credibility of the entire carbon asset trading system, and promote the healthy and orderly operation of the carbon asset trading market.

[0254] Through the collaborative operation of the above layers, the system realizes a complete data processing route from data collection, integration, verification to ensuring the security of data interaction and executing transaction verification processes, effectively improving the accuracy and reliability of data verification in carbon asset transactions and the security and credibility of the entire trading system. Specific Embodiment Five:

[0256] As Figures 1-3 shown, based on the content in the above specific embodiments, the following content is further disclosed:

[0257] When the entire system is actually in use, the hardware architectures corresponding to the modules and units at each level are as follows:

[0258] Hardware-related description of the federated learning and edge computing module in the data acquisition layer:

[0259] Hardware requirements for the federated learning subunit:

[0260] Server device: A server with relatively strong performance needs to be configured to act as the federated learning coordination server, which is used to communicate and coordinate with the local nodes of each data provider and handle operations such as aggregated parameters. This server should have high processing capabilities, such as a multi-core processor, such as the Intel Xeon series, etc., to handle the operation of a large amount of data and the communication and coordination between multiple nodes. At the same time, it is equipped with a large-capacity memory, such as 32GB or more, to ensure the smoothness during the data processing process, and a high-speed network interface, such as a 10 Gigabit Ethernet interface, etc., to ensure fast and stable data transmission with the nodes of each data provider;

[0261] Local node device of the data provider: The local nodes of each data provider may be various computer devices, such as desktop computers or servers, etc. For desktop computers, generally, a processor with medium performance should be equipped, such as the Intel Core i5 series and above, an appropriate amount of memory, such as 8GB or more, and a stable network connection, such as a Gigabit Ethernet interface, to ensure that model training can be carried out based on local carbon data and communication with the coordination server can be achieved, and operations such as uploading encrypted model parameters can be realized. If a server is used as the local node, its hardware configuration can refer to the low-end version of the federated learning coordination server, such as a multi-core processor, a lower configuration model of the Intel Xeon series, 16GB or more of memory, etc., and also needs to have a high-speed network interface to ensure data transmission;

[0262] Hardware requirements for the edge computing subunit:

[0263] Edge computing device: The edge computing device deployed near the enterprise internal data source is the key hardware for real-time preprocessing of raw data. Such devices can be dedicated edge computing servers or industrial gateways with edge computing capabilities, etc. Taking the edge computing server as an example, it needs to have a certain processing capacity. Usually, a low-power but moderately performing processor can be adopted, such as the Intel Atom series or a processor with an ARM architecture, etc., and be equipped with an appropriate amount of memory, such as 4GB and above, to meet the running requirements of operations such as data compression, format conversion, and preliminary screening. At the same time, it should have a variety of data interfaces, such as common Ethernet interfaces for connecting to the enterprise internal network, and other possible interfaces such as serial ports for connecting different types of data source devices, such as sensors, etc., in order to obtain raw data for preprocessing. In addition, considering that it may be deployed in some industrial environments, the device should also have certain protection performances such as dust-proof, moisture-proof, and earthquake-proof.

[0264] Hardware-related description of the semantic web data integration module in the data integration layer: This module mainly focuses on software-level processing of data, such as constructing ontology models, semantic mapping and association, and data storage and management, etc. Relatively speaking, its dependence on specific hardware devices is not as direct as that of the data acquisition layer, but it still needs to rely on a certain hardware foundation during actual operation;

[0265] Server device: Generally, one or more servers are required to run the software programs related to the semantic web data integration module. General-purpose servers with moderate performance can be selected for the servers, such as those equipped with Intel Core i7 series processors, 16GB and above of memory, and a large-capacity hard disk, such as 1TB and above, for storing the integrated data and running related software programs. The server should have a high-speed network interface, such as a gigabit Ethernet interface, to ensure the data transmission speed between it and other layers such as the data acquisition layer and the verification model layer.

[0266] Hardware-related description of the fusion LSTM-SVM verification module and the GNN relationship verification module in the verification model layer:

[0267] High-performance computing device: Since these two verification modules involve relatively complex algorithm operations, such as LSTM network model training, SVM model processing, and GNN algorithm mining, etc., high-performance computing devices are required to ensure the operation speed and accuracy. Usually, dedicated high-performance servers or workstations can be adopted. For example, they are equipped with multi-core high-performance processors, such as high-performance models of the Intel Xeon series, a large amount of memory, such as 64GB and above, and high-speed storage devices, such as solid-state drives, the capacity of which can be determined according to actual needs, and generally 500GB and above are recommended, for storing training data, model files, etc. At the same time, these devices should have high-speed network interfaces, such as 10 Gigabit Ethernet interfaces, to ensure the fast transmission of data between different modules and ensure the efficient progress of the verification process.

[0268] Hardware-related descriptions of the blockchain and smart contract modules in the interaction and security layer:

[0269] Blockchain node devices: For the blockchain data storage and interaction unit, if you want to establish a blockchain network to store key data and provide query interfaces, you need to deploy blockchain node devices. These devices can be dedicated blockchain servers or general-purpose servers with blockchain functions. Taking blockchain servers as an example, their hardware configurations can be similar to high-performance computing devices, that is, equipped with multi-core high-performance processors, such as high-performance models of the Intel Xeon series, large-capacity memory, such as 64GB and above, and high-speed storage devices, such as solid-state drives, the capacity of which can be determined according to actual needs, and generally 500GB and above are recommended. At the same time, in order to ensure the stable operation of the blockchain network and the security of data transmission, it is necessary to have high-speed and secure network interfaces, such as 10 Gigabit Ethernet interfaces and cooperate with encryption transmission protocols, etc.

[0270] Smart contract execution devices: When the smart contract execution unit performs related operations, it also relies on certain hardware devices. Generally, the relevant software programs of the smart contract can be run on the above-mentioned servers or workstations and other devices. Its hardware configuration requirements are similar to those of the devices running the software programs in the verification model layer, that is, equipped with multi-core high-performance processors, such as high-performance models of the Intel Xeon series, large-capacity memory, such as 64GB and above, and high-speed storage devices, such as solid-state drives, the capacity of which can be determined according to actual needs, and generally 500GB and above are recommended. This can ensure the efficiency and accuracy of the smart contract when performing functions such as carbon data submission, verification process, transaction settlement, and violation punishment.

[0271] The above hardware is only a rough reference based on the system function requirements. In actual application scenarios, the hardware devices can be further optimized and adjusted according to specific business scales, data volumes, performance requirements, and other factors. Specific Example Six:

[0273] As Figures 1-3 shown, based on the content in the above specific examples, the following content is further disclosed:

[0274] When the entire system is actually in use and facing a large amount of carbon asset transaction data, to ensure the real-time and integrity of data collection, it further includes the following content:

[0275] Data collection layer

[0276] Federated learning and edge computing module:

[0277] Federated Learning Sub-unit: It establishes communication and coordination between the federated learning coordination server and the local nodes of each data provider. Each data provider trains a local model based on local carbon data and encrypts and uploads the model parameters to the coordination server. After aggregating the parameters, the server distributes and updates the local model to iteratively train the global model. This distributed learning method can make full use of the data resources of all parties, and without directly sharing the original data, it can achieve collaborative training of the model, ensure the integrity of data collection, and at the same time ensure the timeliness of data update through an efficient communication coordination mechanism;

[0278] Edge Computing Sub-unit: Edge computing devices are deployed near the internal data sources of the enterprise to perform real-time preprocessing on the original data, including data compression, format conversion, and preliminary screening. During the data compression process, an adaptive compression algorithm is adopted according to the time series characteristics and data types of the data. A higher compression ratio is used for the basic data with slow changes, and a lower compression ratio is used for the real-time monitoring data with large fluctuations. While ensuring the validity of the data, it minimizes the data transmission volume to the greatest extent, thereby improving the real-time performance of data transmission, and transmitting the screened suspicious abnormal data to the data integration layer to ensure the integrity of the data;

[0279] Adopting technical means such as RPA robots: For example, the carbon asset management RPA robot independently developed by Jiangsu Petroleum can read the basic information and related data of gas stations, match the basic data with the filling items one by one, automatically calculate and fill the results into the carbon asset system form template, realizing the automation of operations such as data reading, filling item matching, and calculation and deduction. It can complete a large amount of data processing in a short time, effectively avoiding problems such as incorrect filling, missing filling, and calculation errors, providing guarantee for the integrity of data collection, and its high efficiency helps to improve the real-time performance of data collection.

[0280] Data Integration Layer

[0281] Semantic Web Data Integration Module:

[0282] Ontology Construction Unit: It constructs an ontology model in the field of carbon trading, defines the core concepts and their interrelationships. Based on international common carbon trading standards and industry norms, it adopts a construction method combining top-down and bottom-up, and regularly updates and maintains the ontology model according to policy and regulation changes and industry development trends. This enables the collected data to be integrated according to unified standards and specifications, ensuring the consistency and integrity of the data at the semantic level, and laying a foundation for subsequent accurate verification and analysis;

[0283] Semantic Mapping and Association Unit: Adopt a mapping method that combines rule-based and machine learning-based approaches to map and match the collected data with the ontology model and associate and integrate them. For common data formats and standard terms, use predefined rules for rapid mapping, and for fuzzy data expressions, use a machine learning model for automatic learning and mapping matching, so as to comprehensively and accurately integrate various types of data and ensure the integrity and accuracy of data integration;

[0284] Data Storage and Management Unit: Classify and store the integrated data and manage access permissions. Through reasonable data classification and strict access control, ensure the orderly storage and secure access of data, prevent data loss or tampering, and further guarantee the integrity of data.

[0285] Interaction and Security Layer

[0286] Blockchain and Smart Contract Module:

[0287] Blockchain Data Storage and Interaction Unit: Establish a blockchain network to store key data and provide query interfaces. The characteristics of multi-party participation, immutability, and traceability of blockchain technology ensure the authenticity and integrity of data. Once the data is on the chain, it can be shared and queried in real time, ensuring the real-time nature and transparency of data, and providing a reliable basis for the collection and management of carbon asset trading data;

[0288] Smart Contract Execution Unit: Write smart contract code and automatically execute carbon asset trading and verification-related operations according to conditions. Adopt a modular design to encapsulate functions such as carbon data submission, verification process, transaction settlement, and violation penalties into independent modules respectively, which is convenient for contract maintenance and upgrade. At the same time, set up an audit mechanism for smart contracts to allow regulatory agencies and third-party audit agencies to supervise and audit the contract execution process, ensure that the data collection and trading process strictly follows the preset rules, and guarantee the integrity and compliance of data.

[0289] 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 "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a reference structure" does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0290] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An automated data verification system for carbon asset trading, characterized in that: The system includes a data collection layer, a data integration layer, a verification model layer, and an interaction and security layer, wherein: The data acquisition layer is used to obtain carbon asset trading related data from multiple sources and perform preliminary processing, including a federated learning and edge computing module, which is used to realize the collaborative training and verification model of multiple data holders through a federated learning algorithm and pre-process data at the data source end using edge computing. The federated learning and edge computing module includes a federated learning subunit and an edge computing subunit. The federated learning subunit is used to establish a federated learning coordination server to communicate and coordinate with the local nodes of each data provider. Each data provider trains a local model based on local carbon data and encrypts and uploads the model parameters to the coordination server. After the server aggregates the parameters, it sends down an updated local model to iteratively train the global model. The edge computing subunit is used to deploy edge computing equipment near the internal data source of the enterprise to perform real-time pre-processing on the raw data, including data compression, format conversion and preliminary screening. The screened suspicious abnormal data is transmitted to the data integration layer; The data integration layer is used to integrate, associate, store and manage the collected data based on semantic web technology, including a semantic web data integration module, which includes an ontology construction unit, a semantic mapping and association unit, and a data storage and management unit. The ontology construction unit is used to construct an ontology model in the field of carbon trading to define core concepts and their interrelationships. The semantic mapping and association unit is used to map and match the collected data with the ontology model and associate and integrate them. The data storage and management unit is used to classify and store the integrated data and manage access rights. The verification model layer is used to verify the integrated data using multiple algorithm models, including a fused LSTM-SVM verification module and a GNN relationship verification module. The fused LSTM-SVM verification module includes an LSTM time series analysis unit and an SVM classification verification unit. The LSTM time series analysis unit is used to obtain historical carbon data and related time series data to train the LSTM network model to output a time series feature vector and a predicted trend value. The SVM classification verification unit is used to receive the feature vector and determine whether the data is normal based on the trained SVM model. The GNN relationship verification module includes a carbon trading relationship graph construction unit and a GNN analysis and verification unit. The carbon trading relationship graph construction unit is used to construct a carbon trading relationship graph. The GNN analysis and verification unit is used to mine abnormal patterns based on the relationship graph using the GNN algorithm; The interaction and security layer is used to ensure the security and reliability of data interaction and automatically execute transactions and verification processes, including a blockchain and smart contract module. The blockchain and smart contract module includes a blockchain data storage and interaction unit and a smart contract execution unit. The blockchain data storage and interaction unit is used to establish a blockchain network to store key data and provide a query interface. The smart contract execution unit is used to write smart contract code and automatically execute carbon asset trading and verification related operations based on conditions.

2. The automated data verification system in carbon asset trading according to claim 1, characterized in that: In the federated learning subunit, each data provider's local node uses homomorphic encryption technology to encrypt local model parameters to ensure that data privacy is not leaked during the parameter upload process, and the coordination server uses a multi-party computing protocol to aggregate the encrypted parameters to achieve secure global model parameter updates.

3. The automated data verification system in carbon asset trading according to claim 1, characterized in that: During the data compression process, the edge computing subunit adopts an adaptive compression algorithm according to the time series characteristics and data type of the data, and uses a higher compression ratio for basic data that changes slowly and a lower compression ratio for real-time monitoring data that fluctuates greatly, thereby minimizing the amount of data transmission while ensuring data validity.

4. The automated data verification system in carbon asset trading according to claim 1, characterized in that: When constructing the ontology model in the field of carbon trading, the ontology construction unit adopts a top-down and bottom-up construction method based on internationally accepted carbon trading standards and industry specifications, first determining the core concept framework, then gradually refining the sub-concepts and attributes, and regularly updating and maintaining the ontology model according to changes in policies and regulations and industry development trends.

5. The automated data verification system in carbon asset trading according to claim 1, characterized in that: The semantic mapping and association unit adopts a mapping method that combines rule-based and machine learning-based mapping when performing semantic mapping. Predefined rules are used to quickly map common data formats and standard terms, and machine learning models are used to automatically learn and map fuzzy data expressions.

6. The automated data verification system in carbon asset trading according to claim 1, characterized in that: When training the LSTM network model, the LSTM time series analysis unit uses sliding window technology to segment the historical carbon data to increase the number of training samples, and combines the early stopping method to prevent overfitting. At the same time, an attention mechanism is introduced to enable the model to focus on the time series feature parts that have an important impact on carbon verification.

7. The automated data verification system in carbon asset trading according to claim 1, characterized in that: When constructing the classification hyperplane, the SVM classification verification unit uses the kernel function technique to map the nonlinearly separable data, selects the radial basis function as the kernel function, and optimizes the kernel function parameters and penalty factors through the cross-validation method to improve the accuracy of the SVM model in classifying complex carbon data.

8. The automated data verification system in carbon asset trading according to claim 1, characterized in that: When constructing the carbon trading relationship graph, the carbon trading relationship graph construction unit adds basic trading entities, trading relationships, changes in carbon trading-related policies and regulations, and market dynamic information into the relationship graph as virtual nodes and edges to more comprehensively reflect the impact of the carbon trading environment on trading relationships.

9. The automated data verification system in carbon asset trading according to claim 1, characterized in that: The GNN analysis and verification unit adopts a multi-layer GNN structure when performing information propagation and node feature update. Each layer of GNN adopts different message passing functions and aggregation functions, and dynamically adjusts the scope and intensity of information propagation according to the node type and edge weight, so as to more accurately mine abnormal patterns in complex relationships.

10. The automated data verification system in carbon asset trading according to claim 1, characterized in that: When writing the smart contract code, the smart contract execution unit adopts a modular design, encapsulating the functions of carbon data submission, verification process, transaction settlement, and violation penalty into independent modules to facilitate contract maintenance and upgrades. At the same time, an audit mechanism for smart contracts is set up to allow regulators and third-party audit agencies to supervise and audit the contract execution process.

Images