Carbon asset transaction intelligent risk control method

Abstract

The invention relates to the technical field of data processing, and provides an intelligent risk control method for carbon asset transaction, which is used for solving the problems of data island, response lag, low efficiency, safety risk and the like in traditional carbon transaction. According to the invention, standardized access of a carbon emission permit trading system, a registration system and enterprise internal data is realized by constructing a carbon asset data middle platform, integrating a block chain and an RPA technology, a price fluctuation rate and a quota gap risk are calculated in real time based on Apache Flink, a carbon price trend is predicted through an LSTM + Attention model, and an intelligent contract is triggered to generate a hedging instruction; a three-level risk prevention and control system is constructed by adopting an ABAC model and biological characteristic verification, a quota regulation path is optimized through a graph neural network, automatic performance is driven, strategy self-adaption is realized in combination with PPO reinforcement learning and BERT policy analysis, and the problems of data islands, response lag, low efficiency and security risks in traditional carbon transactions are systematically solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an intelligent risk control method for carbon asset trading. Background Art

[0002] Intelligent risk control for carbon asset trading refers to a management system that uses technologies such as big data, artificial intelligence, and blockchain to dynamically monitor and warn of risks such as market fluctuations, data authenticity, and policy changes in the process of carbon emission rights trading. Existing intelligent risk control collects multi-dimensional data such as energy consumption, corporate emissions, and climate policies in real time, and uses machine learning algorithms to build prediction models. It can identify risks such as abnormal price fluctuations, imbalances in quota supply and demand, or green certification fraud in advance, and automatically generate hedging strategies or trigger fuse mechanisms. For example, when a sudden drop in renewable energy power generation in a certain area is detected, the system will predict a surge in demand for carbon quotas and automatically recommend that companies adjust their trading strategies; blockchain technology can ensure the traceability of carbon assets and prevent duplicate calculations or data tampering, thereby improving the transparency and stability of the carbon trading market.

[0003] The core pain points of existing carbon trading, such as low efficiency, data silos, delayed response and security risks, mainly stem from the decentralized manual operation mode and the lack of a unified data integration platform, which leads to the inability to communicate data across systems, reliance on post-manual review rather than real-time dynamic monitoring, and loose authority management. In addition, the sources of carbon emission data are complex, business processes are fragmented, information is delayed, and operational loopholes occur frequently, making it difficult to cope with the large-scale and high-frequency carbon asset management needs. Summary of the invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an intelligent risk control method for carbon asset trading, which is used to solve the problems of low efficiency, data silos, delayed response and security risks in traditional carbon trading.

[0005] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows: The carbon asset trading intelligent risk control method provided by the present invention comprises: Step 1: Cross-system data integration and trusted circulation: Build a carbon asset data middle platform, integrate blockchain distributed ledgers and API gateways, and connect to the carbon emission trading system, registration system, thermal power company fuel-side carbon emission data automatic collection system and smart lock system through robotic process automation technology (RPA) to achieve standardized access to multi-source data and hardware linkage; Step 2: Dynamic intelligent risk control: Based on the Apache Flink streaming computing framework, price volatility, transaction volume deviation, and quota supply and demand gap risk indicators are calculated in real time, and the LSTM+Attention model is used to predict carbon price trends and generate risk warning signals; Step 3, Automated transaction execution and permission control: Drive transaction plan decomposition and approval process through smart contracts, perform dynamic permission management based on the ABAC model, key operations require fingerprint and face recognition verification, and link with the Ukey smart lock system to control device access permissions; Step 4: Intelligent quota management optimization: Use the graph neural network model to build a multi-layer graph structure of enterprise-unit-region, output the optimal quota adjustment path that prioritizes the same type of units in the same region, output the optimal quota adjustment path based on graph neural network modeling, and drive automatic fulfillment operations through smart contracts; Step 5, adaptive strategy adjustment, uses the proximal policy optimization (PPO) reinforcement learning algorithm to optimize the risk control strategy, and combines the policy semantic analysis of the BERT model to generate adjustment suggestions. The adjustment suggestions are updated in real time through smart contracts for the dynamic risk threshold of step 2, the authority strategy of step 3, and the quota adjustment weight of step 4.

[0006] Furthermore, in the carbon asset trading intelligent risk control method of the present invention, in step 1: Use ETL tools to extract and clean multi-source data from carbon emission data collection systems, trading systems, and registration systems in real time, and establish a unified data model; Quota allocation records, transaction instructions and performance certificates are used as key data and written into the blockchain distributed ledger through smart contracts.

[0007] Furthermore, in the intelligent risk control method for carbon asset trading described in the present invention, the blockchain distributed ledger adopts a cross-chain interoperability protocol based on the Polkadot framework to establish a trusted channel between the carbon market main chain and the carbon market sub-chain, and realizes the cross-chain mapping and circulation of quota assets through smart contracts, wherein the carbon market main chain is a core blockchain with unified carbon trading rules, and the carbon market sub-chain is a dedicated blockchain for local pilot markets or specific industries.

[0008] Furthermore, in the carbon asset trading intelligent risk control method of the present invention, in step 2: The streaming computing framework is Apache Flink, which accesses market data, corporate carbon emission data streams and policy change text data in real time, and inputs the processed data into the LSTM+Attention model to predict carbon price trends and evaluate the probability of a quota supply and demand gap.

[0009] Furthermore, in the carbon asset trading intelligent risk control method of the present invention, the dynamic intelligent risk control includes: Generate quota gap warning signals through Monte Carlo simulation, which generates random scenarios based on historical quota transaction data in the past three years; When it is detected that the quota gap exceeds the dynamic threshold described in step 2, the smart contract is automatically called to generate a hedging instruction.

[0010] Furthermore, in the carbon asset trading intelligent risk control method of the present invention, in step 3: The Robotic Process Automation (RPA) program logs into the carbon emission trading system through optical character recognition (OCR) technology, simulates user operations to automatically execute trading instructions, and drives the trading plan decomposition and approval process through smart contracts; dynamic permission management is based on the ABAC model, dynamically generates the minimum permission set according to role attributes, transaction subject attributes and operation type attributes, and key operations require fingerprint and face recognition verification. The access to physical equipment is controlled by the Ukey smart lock and the on-chain authorization linkage, realizing three-level risk prevention and control (prior authorization, in-process monitoring, and post-audit); Dynamic permission management is based on the ABAC model. It dynamically generates the minimum permission set according to role attributes, transaction subject attributes, and operation type attributes, and verifies permissions through smart contracts.

[0011] Furthermore, in the carbon asset trading intelligent risk control method of the present invention, the authority management includes: Key operations require fingerprint and face recognition verification. The fingerprint and face recognition verification uses a zero-knowledge proof algorithm. The biometric hash value is stored in the blockchain distributed node. After the verification is passed, it is authorized to execute by the smart contract; Access to physical devices is controlled through the Ukey smart lock and on-chain authorization linkage.

[0012] Furthermore, in the carbon asset trading intelligent risk control method of the present invention, in step 4: The graph neural network model constructs a multi-layer graph structure of enterprises, units and regions, inputs unit types and regional policy factors, and outputs the optimal internal quota adjustment path based on unit type matching and regional policy constraints; After the smart contract monitors the completion of the enterprise's emissions verification, it automatically calls the registration system API to transfer quotas.

[0013] Furthermore, in the intelligent risk control method for carbon asset trading described in the present invention, the quota adjustment path preferentially matches the surplus quotas of the same type of units in the same region.

[0014] Furthermore, in the carbon asset trading intelligent risk control method of the present invention, in step 5: The reinforcement learning algorithm is the PPO algorithm, which abstracts the carbon market into a Markov decision process, and the state space includes prices, policy text vectors and corporate emission data; The policy semantic analysis uses the BERT model to extract entities and classify intents of policy documents. The extracted entities include quota carry-over ratios, and the intent classification includes tightening or loosening policy orientations.

[0015] Beneficial effects of the present invention: The present invention integrates multi-source heterogeneous data (such as emissions, transactions, and registration data) by building a carbon asset data middle platform, and uses blockchain to store key transaction instructions and quota records, so as to achieve cross-system data standardization and non-tamperability, eliminate the inefficiency and errors of manual integration, and provide a high-quality data foundation for dynamic risk control. The data middle platform solves the problem of inconsistent formats, and the blockchain ensures that the data is credible. The two work together to break down information barriers, so that risk control decisions are based on comprehensive and real data support, avoiding the risk of misjudgment caused by data loss or tampering.

[0016] Based on Apache Flink, we can calculate price volatility (e.g., the sliding window is 10 minutes, and the volatility calculation uses the historical standard deviation method), transaction volume deviation and other risk indicators in real time, and use the LSTM+Attention model to predict carbon price trends. We can quantify the probability of quota gaps through Monte Carlo simulation and generate graded warning signals (e.g., yellow / red warnings). Compared with traditional manual analysis, we can improve the risk identification time from hours to seconds. Stream computing captures instantaneous market changes, machine learning models integrate multi-dimensional data to predict risks, and probabilistic evaluation makes warnings more accurate, ensuring that companies can take hedging measures before abnormal price fluctuations or quota shortages to avoid huge losses.

[0017] The Robotic Process Automation (RPA) program automatically logs into the trading system through optical character recognition (OCR) technology, encrypts API calls to execute instructions, and reduces the time taken for a single transaction from 15 minutes to within 3 seconds; the zero-trust architecture (ABAC model) dynamically generates the minimum permission set, superimposes biometric verification (fingerprint + face) and links with the Ukey smart lock to prevent unauthorized operations and theft of physical equipment. Automation replaces manual operations to reduce delays and errors, and dynamic permissions and biometric verification build multiple security lines to solve the risk of leakage of traditional password management and static permissions, ensuring efficient and compliant transactions.

[0018] Graph Neural Network (GNN) constructs a multi-layer graph of enterprise-unit-region, generates the optimal adjustment path based on the unit type matching and regional policy constraints (giving priority to units of the same type in the same region), and reduces cross-regional transaction friction costs; smart contracts automatically transfer quotas after emission verification is completed to avoid manual operation delays or violations. GNN quantifies technical differences and policy costs, and smart contracts have embedded rules to ensure real-time compliance. The two work together to solve the inefficiency and subjective bias of manual adjustment, and ensure that enterprises achieve optimal resource allocation within the policy framework.

[0019] Dynamic strategy adjustment and policy response: The proximal policy optimization (PPO) reinforcement learning algorithm models the carbon market as a Markov decision process, dynamically optimizing risk control thresholds and trading strategies; the BERT model analyzes the entities (such as the quota carryover ratio) and intentions (tightening / loosening) of the policy text, and provides real-time feedback to the risk control engine, allowing the system to adapt to policy changes and market evolution. Reinforcement learning continuously evolves strategies through trial and error, and policy analysis converts text semantics into quantifiable parameters. The combination of the two makes the risk control system forward-looking and flexible, avoiding the rigidity of strategies caused by static rules, and maintaining the effectiveness of risk control in the long term.

[0020] Data-Risk Control-Execution-Audit Closed Loop: From data integration, real-time risk control, automated execution to blockchain evidence storage, a complete "monitoring-decision-making-execution-traceability" link is formed. All operation logs and quota transfer records are stored on the chain to support rapid audits by regulatory agencies and meet compliance requirements. The closed-loop design ensures that data in each link is traceable and operations are verifiable. The tamper-proof nature of blockchain enhances regulatory trust and systematically reduces corporate compliance costs and legal risks.

[0021] This invention systematically solves the problems of integration delay, insufficient risk response time, low operation efficiency and poor strategy adaptability in traditional carbon trading due to data dispersion through cross-system data integration, real-time intelligent risk control, automated execution, dynamic strategy optimization and full-link closed-loop design. The synergy between its technical means not only improves the accuracy and response speed of risk control, but also provides an efficient, safe and compliant integrated solution for carbon asset trading through data credibility, automated operation and strategy adaptation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0023] Figure 1 A schematic flow chart of the intelligent risk control method for carbon asset trading provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.

[0025] In order to better understand the purpose of the present invention, the present invention is described in further detail below.

[0026] like Figure 1 As shown, Figure 1 The figure is a flow chart of the intelligent risk control method for carbon asset trading according to an embodiment of the present invention, which specifically includes the closed-loop control logic of steps 1 to 5. The intelligent risk control method for carbon asset trading provided by the present invention includes: Step 1: Cross-system data integration and trusted circulation: Build a carbon asset data middle platform, integrate blockchain distributed ledgers and API gateways, and connect to the carbon emission trading system, registration system, thermal power company fuel-side carbon emission data automatic collection system and smart lock system through robotic process automation technology to achieve standardized access to multi-source data and hardware linkage; Step 1: Cross-system data integration and trusted flow include the following: Construction of data middle platform: Data extraction and cleaning: Use ETL tools (such as Apache NiFi) to extract heterogeneous data (such as JSON and XML formats) from carbon emission data collection systems, trading systems, and registration systems in real time, clean redundant data (such as duplicate records and null values), and then perform structured storage according to a unified data model (such as a carbon emission measurement model based on the ISO 14064 standard).

[0027] Standardized conversion: Key fields such as fuel consumption are converted into unified units (such as tons of standard coal) according to ISO standards, and the data flow path is recorded through data lineage tracking technology.

[0028] Fuel consumption standardization conversion rules: Raw coal consumption: The discount factor based on low calorific value is 0.7143 kgce / kg; Natural gas consumption: converted into tons of standard coal (1 ton of standard coal = 29.3076 GJ) based on the average lower calorific value of 38.93 MJ / m³; Data verification: Grubbs outlier test (significance level α=0.05) was performed on the transformation results, and data deviating from the mean ±3σ were eliminated.

[0029] Blockchain evidence storage: Smart contract triggering: When a transaction instruction or performance certificate is generated, the smart contract is triggered to write the data hash value (such as SHA-256) into the Hyperledger Fabric consortium chain, and the nodes on the chain (regulatory agencies, exchanges) simultaneously verify the data integrity.

[0030] Cross-chain interoperability: Polkadot relay chain technology is used to establish a cross-chain bridge between the carbon market main chain and the local pilot chain. Cross-chain transactions must verify asset ownership through a notary mechanism to ensure that the mapping and transfer of quota assets comply with preset rules (such as regional quotas cannot be over-adjusted across provinces).

[0031] By integrating and standardizing multi-source data through the data middle platform, the problem of data silos is solved; blockchain evidence storage ensures that data cannot be tampered with, and cross-chain protocols enable multi-market data interoperability, providing a trusted data foundation for dynamic risk control.

[0032] Step 2: Dynamic intelligent risk control: Based on the Apache Flink streaming computing framework, price volatility, transaction volume deviation, and quota supply and demand gap risk indicators are calculated in real time, and the LSTM+Attention model is used to predict carbon price trends and generate risk warning signals; Step 2: Dynamic intelligent risk control includes the following: Real-time data processing: The Apache Flink stream processing engine receives market information (such as price and transaction volume), corporate emission data streams (such as hourly carbon emissions), and policy texts (through NLP to parse the Ministry of Ecology and Environment announcements) in real time to form a unified data stream.

[0033] Sliding window statistics: Calculate price volatility and trading volume deviation from the historical mean (such as the 3σ threshold) by time window (such as 10 minutes), and dynamically update the risk threshold.

[0034] Risk prediction and early warning: LSTM+Attention model: Input the historical carbon price series, capture key time nodes (such as policy release date) through the attention mechanism, and predict the price trend in the next hour.

[0035] Monte Carlo simulation: Generate random scenarios based on historical data on quota supply and demand, calculate the probability of a gap (if the probability > 80% triggers an early warning), and generate hedging recommendations (such as buying futures contracts).

[0036] The streaming computing framework captures market changes in real time, and combines machine learning model predictions with simulation evaluations to identify potential risks in advance and avoid losses caused by delayed manual responses.

[0037] Step 3, Automated transaction execution and permission control: Drive transaction plan decomposition and approval process through smart contracts, perform dynamic permission management based on the ABAC model, key operations require fingerprint and face recognition verification, and link with the Ukey smart lock system to control device access permissions; Step 3: Automated transaction execution and permission control include the following: RPA Automation Process: Login verification: The RPA robot uses OCR to identify the transaction system verification code, decrypts the AES-GCM encrypted account password (the key is stored in the hardware security module), and simulates user login.

[0038] The RPA automated login process includes: Verification code preprocessing: Perform Gaussian filtering (σ=1.5) on the verification code image to reduce noise, and extract character outlines using the Canny edge detection algorithm (threshold 50-150); OCR recognition: CRNN+Attention model is used to recognize distorted text, with a recognition rate of ≥98.5%; Password decryption: The account password is encrypted using AES-GCM-256 (IV length 12 bytes, tag length 16 bytes), the key is stored in the HSM hardware security module (compliant with FIPS 140-2 Level 3 standards), and the decryption response time is <10ms.

[0039] Instruction hierarchical execution rules: Automatic execution conditions: When the single transaction volume is ≤ 20% of the enterprise's average daily transaction volume and the dynamic threshold alarm is not triggered, the RPA robot will automatically call the API for execution; Manual review conditions: Risk control personnel are required to review online when any of the following conditions are met: Single transaction volume> 20% of the enterprise's average daily transaction volume; The transaction subject includes cross-region quotas (cross-chain mapping compliance needs to be verified); The price deviates from the current market average price by ±10%; Audit requirements: Automatic execution of instructions must record operation logs (including timestamps and transaction hashes) and write them into the blockchain. Manual review must simultaneously retain electronic signatures and approval opinions.

[0040] Zero Trust Permission Management: ABAC model: Dynamically generate the minimum permission set based on the user role (such as trader), transaction object attributes (such as 2023 quota label), and operation type (such as order withdrawal). For example, a trader can only operate the quota in his jurisdiction.

[0041] Biometric linkage: Over-limit transactions require fingerprint + face recognition. After verification, the smart contract will authorize the execution and trigger the Ukey smart lock (on-chain authorization is required) to unlock the physical device.

[0042] Transactions exceeding the limit require fingerprint + face double-factor verification: Fingerprint recognition: Use FIPS 201-2 standard to extract fingerprint feature points (bifurcation points, endpoints), generate SHA-3-256 hash values, and encrypt and store them in IPFS distributed storage nodes; Face recognition: Capture facial depth information through a 3D structured light camera (wavelength 850nm, accuracy ±0.1mm), extract 72 biometric feature points (eye distance, nose bridge curvature, etc.), and use a zero-knowledge proof algorithm to compare with the pre-stored template on the chain; After verification, a one-time ED25519 signature token is generated by the smart contract, triggering the Ukey smart lock (on-chain authorization is required) to unlock the physical device.

[0043] RPA replaces manual operations to improve efficiency, and the zero-trust architecture ensures operational security through dynamic permissions and biometric verification to prevent unauthorized access and internal fraud.

[0044] Step 4: Intelligent quota management optimization: Use the graph neural network model to build a multi-layer graph structure of enterprise-unit-region, output the optimal quota adjustment path that prioritizes the same type of units in the same region, output the optimal quota adjustment path based on graph neural network modeling, and drive automatic fulfillment operations through smart contracts; Step 4: Intelligent quota management optimization includes the following: Optimization of quota adjustment path: Optimization of quota adjustment path: Based on the types of units within the group (such as supercritical coal-fired units) and regional policy constraints, priority will be given to matching surplus quotas of units of the same type in the same region; Combined with carbon emission data on the fuel side (such as the amount of coal entering the furnace and the coal consumption for power supply), the adjustment path is dynamically adjusted, and the quota transfer is automatically executed through smart contracts.

[0045] Smart Contract Execution: Verification trigger: When the enterprise's emissions verification is completed (data comes from the registration system), the smart contract automatically calls the API to transfer quotas and generates a blockchain electronic performance certificate (including timestamp and hash value) in compliance with regulatory requirements.

[0046] Graph neural networks take into account regional policies and unit characteristics to optimize quota allocation; smart contracts automatically execute fulfillment and reduce delays and errors caused by human intervention.

[0047] Step 5, adaptive strategy adjustment, uses the proximal policy optimization (PPO) reinforcement learning algorithm to optimize the risk control strategy, and combines the policy semantic analysis of the BERT model to generate adjustment suggestions. The adjustment suggestions are updated in real time through smart contracts for the dynamic risk threshold of step 2, the authority strategy of step 3, and the quota adjustment weight of step 4.

[0048] Step 5: Adaptive strategy adjustment includes the following: Reinforcement Learning Dynamic Optimization: Strategy Optimization: According to the group's carbon asset trading strategy (such as the annual quota matching plan), the reinforcement learning algorithm is used to optimize the decomposition parameters of the trading plan; Combined with the performance warning results (such as quota gap probability), the risk control indicators are dynamically adjusted and synchronized to the transaction execution module through smart contracts.

[0049] Policy semantic analysis: BERT model application: perform entity recognition on policy documents (such as extracting "quota carry-over ratio" as a numerical value) and intent classification (such as "tightening" requires reducing quota positions), generate strategy adjustment suggestions (such as increasing the proportion of forward contracts), and use the BERT-base Chinese pre-training model for BERT model parameters, which are fine-tuned through policy corpus.

[0050] Reinforcement learning enables the system to adapt to market changes, and policy analysis ensures that strategies are in line with regulatory guidelines, forming a "monitoring-adjustment-execution" closed loop and enhancing long-term risk control capabilities.

[0051] Specifically, in the carbon asset trading intelligent risk control method of the present invention, in step 1: Use ETL tools to extract and clean multi-source data from carbon emission data collection systems, trading systems, and registration systems in real time, and establish a unified data model; Quota allocation records, transaction instructions and performance certificates are used as key data and written into the blockchain distributed ledger through smart contracts.

[0052] Step 1: Cross-system data integration and trusted flow include the following: The construction logic of the data center: Multi-source data access: Integrate carbon emission data collection systems (such as enterprise energy consumption monitoring equipment), trading systems (such as carbon exchange order book interfaces) and registration systems (such as government quota management platforms) through API gateways to solve the problem of heterogeneous data sources.

[0053] ETL tool features: Real-time extraction: Use Kafka or Flink to capture data change events (such as quota transaction record updates) in real time to avoid the lag of traditional T+1 data synchronization.

[0054] Data cleaning: deduplication of raw data (such as duplicated fuel consumption), normalization (such as unified measurement units of tons of CO2 equivalent), and outlier filtering (such as carbon emission values ​​that exceed the historical fluctuation range).

[0055] Unified data model: Design a data model based on industry standards, map fields such as fuel type and unit parameters into a standardized format, and ensure that data across systems can be aligned.

[0056] The blockchain evidence storage mechanism includes the following: Key data screening: Automatically identify the data that needs to be stored through the rule engine, for example: Quota allocation records: initial quota quantities allocated by the government and records of adjustments; Trading instructions: pending order price, transaction volume and buying and selling direction; Compliance certificate: the company's annual carbon emissions verification report and quota settlement record.

[0057] Smart contract writing process: Data hash calculation: Generate a unique hash value (such as SHA-256) for the original data; On-chain evidence storage: calling the smart contract to write the hash value and metadata (timestamp, data source) into the channel ledger of Hyperledger Fabric; Node consensus: Regulators, exchanges, and other participating nodes verify the legitimacy of the data and reach a consensus to ensure that the data cannot be tampered with.

[0058] Cross-chain interoperability implementation includes the following: Polkadot Protocol Applications: Relay chain bridging: Deploy the relay chain between the carbon market chain (main chain) and the local pilot chain (parallel chain), and realize cross-chain communication through XCMP (cross-chain messaging protocol).

[0059] Asset mapping rules: Smart contracts define the cross-chain transfer rules for quota assets, and the mapping ratio is dynamically calculated based on regional carbon emission intensity differences: Ratio formula: Mapping ratio = 1-(regional carbon emission intensity / benchmark carbon emission intensity), where the benchmark value is the average carbon emission intensity of the region over the past three years; Example: If the carbon emission intensity of a pilot area is 1.2 tCO2 / 10,000 yuan GDP, and the benchmark value is 1.0 tCO2 / 10,000 yuan GDP, then the mapping ratio = 1 - (1.2 / 1.0) = 0.8, that is, 1 unit of pilot quota = 0.8 unit of main chain quota; Dynamic update: The benchmark value is updated on an annual basis and is confirmed by the ecological environment regulatory agency node through on-chain governance voting.

[0060] Cross-chain transaction verification: Light client verification technology is used to ensure that cross-chain transactions comply with the preset rules on both chains (for example, local chain quotas cannot be transferred to the main chain in excess).

[0061] Necessity of data integration: Root cause of the problem: In traditional carbon trading, corporate emission data, transaction records, and government quota data are scattered in independent systems, and the data formats are not unified (such as Excel, API, database), resulting in low efficiency and error-prone manual aggregation.

[0062] Technical solution path: Data center: Through ETL tools and standardized models, multi-source data is unified into computable structured data to provide high-quality input for subsequent risk control; Blockchain evidence storage: solves the problem of data credibility and prevents companies from tampering with historical emission data or falsifying transaction records.

[0063] Synergy between blockchain and cross-chain design: Trusted flow logic: Blockchain evidence storage ensures the authenticity of a single piece of data, but cross-market transactions require solving the problem of data interoperability between chains; The cross-chain interoperability protocol based on the Polkadot framework realizes asset mapping between heterogeneous chains through the relay chain. For example, the quota of the local pilot market can be exchanged for the general market quota through cross-chain, avoiding the repeated development of independent interfaces.

[0064] Compliance guarantee: Cross-chain rules are enforced by smart contracts to ensure that quota transfers comply with regional restrictions.

[0065] Specifically, the carbon asset trading intelligent risk control method described in the present invention, the blockchain distributed ledger adopts a cross-chain interoperability protocol based on the Polkadot framework, establishes a trusted channel between the carbon market main chain and the carbon market sub-chain, and realizes the cross-chain mapping and circulation of quota assets through smart contracts, wherein the carbon market main chain is the core blockchain of unified carbon trading rules, and the carbon market sub-chain is a dedicated blockchain for local pilot markets or specific industries. The proportion of the cross-chain mapping is dynamically calculated based on the regional carbon emission intensity, and the calculation formula is: mapping ratio = 1 - (regional carbon emission intensity / benchmark value), where the benchmark value is the average carbon emission intensity over the past three years.

[0066] Definition of main chain and sub-chain: Carbon market main chain: deploy a core blockchain with unified carbon trading rules to store quota allocation, transaction settlement and regulatory audit data; Carbon market subchain: A dedicated blockchain for local pilot markets or specific industries (such as electricity and steel) that stores regional quotas, transaction records and compliance data in accordance with local policies.

[0067] Polkadot Relay Chain: As a cross-chain hub, it accesses the main chain and sub-chain through the parachain slot and is responsible for the routing and verification of cross-chain messages. The relay chain has a built-in governance module, and the cross-chain rules are jointly maintained by regulatory agencies, exchanges and other nodes.

[0068] Trusted channel establishment mechanism: Validator election: Validator nodes are elected through the NPoS (Nominee Proof of Stake) mechanism to perform multi-signature verification on cross-chain transactions.

[0069] Cross-chain Message Passing (XCMP): The main chain and the sub-chain transmit cross-chain transaction requests through the XCMP protocol. For example, sub-chain A initiates a quota cross-chain transfer application to the main chain. The message contains the transaction hash, asset type, quantity and target chain address, and is forwarded to the target chain after being verified for legitimacy by the relay chain validator.

[0070] Smart contract driven cross-chain mapping: Mapping rule definition: Smart contracts are deployed simultaneously on the main chain and the sub-chain to define the asset mapping ratio (e.g. 1 unit of quota on the pilot chain = 0.95 units on the main chain, taking into account the differences in regional emission reduction difficulties); The rules are embedded with policy constraints. For example, the quotas of pilot provinces cannot be circulated across provinces and can only be exchanged for the general quotas of the main chain in proportion.

[0071] Asset Locking and Minting: When a subchain initiates a cross-chain request, the smart contract locks the local quota and generates a mainchain quota casting certificate; After receiving the certificate, the main chain contract will cast an equivalent general quota in proportion and record the mapping relationship in the on-chain database.

[0072] Cross-chain transfer process: Transaction initiation: The enterprise submits a cross-chain transaction application through the sub-chain DApp, specifying the target chain (main chain) and the quota quantity; Compliance verification: The subchain smart contract verifies whether the application complies with local policies (such as the validity of the quota year and the company's contract performance status); Cross-chain execution: After verification, the sub-chain contract locks the assets and generates a cross-chain message. The relay chain verifier verifies the authenticity of the message. The main chain contract completes the quota casting and updates the account balance.

[0073] Compliance assurance mechanism: Participation of regulatory nodes: Relay chain validators include regulatory agency nodes, which conduct secondary audits of cross-chain transactions (such as preventing quota washing); On-chain rule synchronization: The main chain and sub-chain contracts regularly synchronize the latest policies (such as the quota carry-over rules of the Ministry of Ecology and Environment), and update the mapping ratio and constraints through governance voting.

[0074] Necessity of cross-chain interoperability: Root of the problem: In the traditional carbon market, the main chain and local chains operate independently, and quota assets cannot be circulated across markets, resulting in regional market fragmentation and insufficient liquidity.

[0075] Technical solution path: Polkadot relay chain: solves the communication problem of heterogeneous chains and avoids repeated development of private interfaces; Dynamic mapping rules: flexibly adapt to policy changes (such as regional quota limit adjustments) through smart contracts.

[0076] Synergy of technical means: Main chain-subchain division of labor: The main chain undertakes global clearing and supervision functions, and the sub-chain serves local specific needs, forming a "global unification + local flexibility" architecture; Cross-chain mapping enables interoperability of asset values ​​and improves the overall efficiency of the carbon market.

[0077] The core functions of smart contracts: It not only defines cross-chain rules, but also automatically executes asset locking and minting, reducing the risk of manual intervention; On-chain rules and policies are synchronized in real time to ensure that cross-chain operations are legal and compliant.

[0078] Compliance design logic: Double verification mechanism: initial review of subchain contracts (local policies) + final review by relay chain validators (global rules) to prevent illegal cross-chain operations; Regulatory penetration: All cross-chain transaction records are permanently stored on the main chain, supporting retrospective audits by regulatory agencies.

[0079] Summary of technical effects: Improve liquidity: Through cross-chain mapping, local pilot quotas can be converted into general quotas to expand the scope of asset circulation.

[0080] Policy adaptability: Smart contracts dynamically adjust mapping rules to quickly respond to policy changes (such as adding new regional transaction restrictions).

[0081] Controllable risks: The validator mechanism and double verification ensure the legality of cross-chain transactions and prevent quota abuse or illegal transfer.

[0082] Specifically, in the carbon asset trading intelligent risk control method of the present invention, in step 2: The streaming computing framework is Apache Flink, which accesses market data, corporate carbon emission data streams and policy change text data in real time, and inputs the processed data into the LSTM+Attention model to predict carbon price trends and evaluate the probability of a quota supply and demand gap.

[0083] Streaming data processing architecture: Real-time access mechanism: Market data: Obtain data such as carbon price, trading volume, buy and sell order depth in real time through the exchange API, align timestamps of different data sources with event time, and solve the problem of data disorder caused by network delay.

[0084] Enterprise carbon emission data stream: collect data such as carbon emissions and energy consumption types (such as coal and gas) in real time from enterprise edge devices (such as IoT sensors) or ERP systems, and filter out abnormal values ​​(such as sudden increases in invalid zero values) through data cleaning.

[0085] Policy change text data: crawl public texts from ecological environment websites, and use NLP technology to extract key policy events (such as "the quota carry-over ratio is adjusted to 5%") and policy-oriented labels (such as "tightening" or "easing").

[0086] Key functions of the streaming computing framework: Apache Flink core functions: Window calculation: define a sliding time window (e.g. 10-minute window, 1-minute sliding interval), and calculate price volatility, transaction volume mean and standard deviation in real time; Status management: Maintain the rolling status of the company's historical emission data (such as the emission trend in the past 24 hours) to dynamically adjust the risk threshold; Data diversion: The cleaned data is diverted to different processing pipelines by type (such as price data → prediction model, policy text → event parsing engine).

[0087] The design logic of the LSTM+Attention model: Input data processing: Time series feature construction: align the carbon price series with the enterprise emission data along the time axis to generate a multi-dimensional time series feature matrix; Policy event embedding: Parse the policy text into a vector (e.g., “tightening policy” is encoded as [-1, 0.5]) and input the model as an external feature.

[0088] Attention mechanism application: The model automatically identifies key time nodes that affect carbon prices (such as policy release dates and compliance deadlines) and amplifies the characteristic impact of these nodes through attention weights; For example, policy tightening events may lead to higher carbon prices in the future, and the model strengthens the weight of such events through the attention mechanism.

[0089] Assessment of quota supply and demand gap: Monte Carlo simulation process: Historical data sampling: Time window: Extract the rolling historical data of the last three years (calculated by natural years, such as 2020-2022), and exclude outliers that are beyond the statistical error range (±2σ) of the ISO 14064:2018 standard; Regional matching: Prioritize the use of historical data of regions with similar economic development levels and industrial structures to the target region (if they are energy-intensive industrial regions); Data verification: Historical data must contain at least 80% of complete trading day records, and missing data must be interpolated using the ARIMA model; Random scenario generation: simulate possible future market scenarios (such as extreme weather causing a surge in emissions) and calculate the probability of quota gaps under each scenario; Warning trigger logic: When the gap probability exceeds the dynamic threshold (such as >70%), a warning signal is generated and the hedging strategy is triggered.

[0090] Root cause of the problem: Traditional risk control relies on offline batch processing of data, resulting in delayed risk response (such as T+1 analysis) and an inability to cope with high-frequency market fluctuations.

[0091] Technical solution path: Stream computing framework: Apache Flink is used to achieve millisecond-level data processing and capture market changes in real time; Timing model optimization: LSTM captures long-term price trends, and the attention mechanism strengthens the impact of short-term events and improves the timeliness of predictions.

[0092] Collaborative analysis of multi-source data: Market conditions reflect current supply and demand, corporate emission data predict future quota demand, and policy changes affect long-term market expectations. The combination of the three can comprehensively assess risks. For example, when a region experiences a sudden increase in carbon emissions, if coupled with a signal of policy tightening, the model can predict that the risk of quota shortage in that region will increase.

[0093] Dynamic threshold and early warning mechanism: Adaptive adjustment logic: The initial threshold is set based on historical data, and the streaming computing framework updates the threshold in real time (for example, the trading volume volatility threshold is dynamically adjusted according to market activity); Monte Carlo simulation provides probabilistic risk assessment and avoids false positives or negatives caused by a single threshold.

[0094] Synergistic Effects of Technological Means Streaming computing → model prediction → simulation evaluation: Flink real-time processing provides high-quality input for the model; Model output guides the direction of scenario generation for Monte Carlo simulation; The simulation results are fed back to the risk control strategy, forming a closed loop of “data → prediction → decision-making”.

[0095] Specifically, the carbon asset trading intelligent risk control method of the present invention, the dynamic intelligent risk control includes: Generate quota gap warning signals through Monte Carlo simulation, which generates random scenarios based on historical quota transaction data in the past three years; The quota gap warning signal is generated through Monte Carlo simulation. The specific process is as follows: Historical data sampling: extract the rolling historical quota transaction data of the past three years (calculated by natural year, such as 2020-2022), and exclude outliers that exceed the statistical error range of ±2σ; Dynamic threshold adjustment: The threshold is updated based on the real-time market volatility (the price standard deviation is calculated through Apache Flink). When the volatility increases, the threshold is lowered, and vice versa. Random scenario generation: Generate 1,000 sets of random scenarios based on historical data distribution, including extreme events such as carbon price fluctuations (±20%) and regional emissions surges (±30%); Gap probability calculation: The percentage of scenarios where the quota gap exceeds the dynamic threshold is calculated. When the probability is > 80%, a red warning is triggered.

[0096] When it is detected that the quota gap exceeds the dynamic threshold described in step 2, the smart contract is automatically called to generate a hedging instruction.

[0097] Quota gap warning mechanism based on Monte Carlo simulation: Input Data Source: Historical quota transaction data: extract historical quota prices, transaction volumes and regional allocation records from the blockchain distributed ledger; Real-time market dynamics: current carbon prices, real-time corporate emissions, and policy change event labels (such as "quota carryover ratio adjustment") output by the streaming computing framework; External environmental variables: macroeconomic indicators (such as industrial output growth rate), climate data (such as extreme weather warnings), etc.

[0098] Simulation process: Scenario generation: Based on the distribution of historical data, randomly generate N possible market scenarios in the future (such as carbon price fluctuation of ±20%, emission surge of 30% in a certain region); Gap calculation: For each scenario, calculate the difference between the enterprise's quota demand and the current holdings, and calculate the probability of the gap (e.g., scenarios with a gap > 10% account for 80%); Signal triggering: When the gap probability exceeds the dynamic threshold (such as probability > 75%), a graded warning signal (such as yellow warning, red warning) is generated.

[0099] Dynamic threshold adjustment logic: Initial threshold value: a benchmark threshold is set based on historical risk events (such as the quota shortfall event in the past year); Real-time correction: The streaming computing framework continuously monitors market volatility, policy sensitivity and other indicators, and updates thresholds through sliding windows (such as 24-hour windows): If market volatility increases (e.g., price standard deviation increases), the dynamic threshold is lowered (e.g., from 75% to 65%) to improve the sensitivity of early warning; If policy stability is high (e.g., no major policy adjustments in the near future), the threshold will be increased to reduce false positives.

[0100] Smart contract hedging instruction generation: Hedging strategy matching: Gap type identification: distinguish between short-term liquidity gaps (need to buy futures) and long-term structural gaps (need to buy CCER to offset); Strategy library call: Smart contract built-in strategy library, for example: Buy quota futures (to hedge against price upside risk); Selling call options (locking in a future carbon price cap); Cross-regional quota adjustment (using cross-chain protocols to obtain low-priced regional quotas).

[0101] Instruction execution guarantee: Permission verification: Hedging instructions must verify operation permissions through a zero-trust architecture (e.g. only risk control administrators can trigger large transactions); Compliance verification: The smart contract calls the policy parsing module to ensure that the hedging strategy complies with the latest regulatory requirements (such as prohibiting excessive cross-provincial transactions).

[0102] Applicability of Monte Carlo simulation: Root of the problem: Traditional risk control relies on static thresholds (such as fixed gap ratios) and cannot cope with the high uncertainty of the carbon market (such as sudden policy changes and climate anomalies).

[0103] Technical advantages: Probabilistic assessment: Monte Carlo simulation quantifies the possibility of risk occurrence, rather than relying solely on absolute values, which is more in line with actual decision-making needs; Multi-factor fusion: consider multi-dimensional variables such as market, policy, and environment simultaneously to avoid the limitations of a single data source.

[0104] Collaborative logic between dynamic threshold and smart contract: Real-time guarantee: The streaming computing framework dynamically adjusts the threshold to ensure that the warning signal matches the current market status; Automation closed loop: The early warning signal triggers the smart contract, which automatically matches the optimal hedging strategy; The strategy execution results (such as futures transaction records) are written back to the blockchain for evidence storage, forming a closed loop of "risk identification → hedging decision → execution evidence storage".

[0105] Compliance design of hedging strategies: Preset rules in the policy library: All hedging strategies are embedded with regulatory compliance conditions (e.g. Article 15 of the Interim Regulations on Carbon Emission Trading restricts the proportion of cross-provincial transactions); Automatically verify the legality of the strategy before the smart contract is executed to prevent illegal operations.

[0106] Risk diversification mechanism: Reduce the volatility risk of a single hedging tool through a combination of multiple strategies (such as futures + options); Cross-chain adjustments utilize regional price differences to optimize hedging costs.

[0107] Summary of technical effects Accurate early warning: Monte Carlo simulation quantifies risk probability, and dynamic thresholds reduce false alarms and missed alarms.

[0108] Smart hedging: The strategy library is combined with compliance verification to achieve automated and legalized risk hedging.

[0109] Closed-loop risk control: Automating the entire process of early warning, decision-making, execution, and evidence storage to improve the efficiency of carbon asset management.

[0110] Specifically, in the carbon asset trading intelligent risk control method of the present invention, in step 3: The Robotic Process Automation (RPA) program logs into the carbon emission trading system through optical character recognition (OCR) technology, simulates user operations to automatically execute trading instructions, and drives the trading plan decomposition and approval process through smart contracts; dynamic permission management is based on the ABAC model, dynamically generates the minimum permission set according to role attributes, transaction subject attributes and operation type attributes, and key operations require fingerprint and face recognition verification. The access to physical equipment is controlled by the Ukey smart lock and the on-chain authorization linkage, realizing three-level risk prevention and control (prior authorization, in-process monitoring, and post-audit); The execution of the trading instructions meets the grading rules: when the single transaction volume is ≤ 20% of the enterprise's average daily transaction volume, it will be automatically executed; if it exceeds this, manual review is required.

[0111] Dynamic permission management is based on the ABAC model. It dynamically generates the minimum permission set according to role attributes, transaction subject attributes, and operation type attributes, and verifies permissions through smart contracts.

[0112] RPA robot login and transaction execution process: OCR verification code recognition: Preprocessing and feature extraction: The RPA robot grayscales and performs noise reduction on the verification code image on the transaction system login interface, extracts character contour features through a convolutional neural network (CNN), and identifies distorted and contiguous verification code texts.

[0113] Dynamic decryption mechanism: The login password is encrypted and stored using the AES-GCM algorithm. The key is managed by the hardware security module (HSM) and is only dynamically decrypted and used during login to avoid the risk of plain text password leakage.

[0114] API command execution: Instruction encapsulation and signature: Trading instructions (such as order price and quantity) are encapsulated in JSON format, and digital signatures are generated through asymmetric encryption (RSA) to ensure instruction integrity and identity authenticity.

[0115] Abnormal retry mechanism: If the API call fails due to network fluctuations, RPA automatically retries according to the exponential backoff strategy (such as intervals of 1 second, 2 seconds, or 4 seconds) until it succeeds or triggers a manual intervention alarm.

[0116] ABAC dynamic rights management logic: Attribute definition and association: Role attributes: user roles (such as traders and risk controllers) are associated with basic permissions (such as traders can place orders and risk controllers can review orders); Transaction subject attributes: quota label (such as 2023 quota, pilot area quota), transaction type (such as spot, futures); Operation type attributes: operation classification (such as placing an order, cancelling an order, and inquiring), operation risk level (such as high-risk large-value cross-provincial transactions).

[0117] Minimum permission set generation: Context-aware: Dynamically adjust permissions based on real-time environment (such as trading hours, market price volatility). For example, pending orders are prohibited during non-trading hours; Strategy engine calculation: The strategy engine matches the preset rule base based on attribute combinations (such as "trader + 2023 quota + pending order") to generate a permission set that only contains necessary operations.

[0118] Smart contract verification process: On-chain strategy synchronization: permission strategies (such as "daily order volume ≤ 1,000 tons") are deployed to the blockchain through smart contracts, making the strategy transparent and tamper-proof; Real-time verification: When a user initiates an operation, the ABAC engine generates a permission request, and the smart contract verifies whether the request complies with the on-chain policy and returns an "allow / deny" result; Audit traceability: All permission verification records (including request attributes and verification results) are written into the blockchain to support post-event tracing.

[0119] Biometrics and physical devices linked: Multimodal verification: key operations (such as a single transaction ≥ 1,000 tons) require fingerprint + face recognition, and the biometric data is encrypted and compared with the pre-stored template on the chain to prevent forgery; Ukey smart lock control: The use of physical devices (such as transaction key storage) requires on-chain authorization. After the smart contract verifies the operation authority, the Ukey is unlocked through the IoT command to ensure the triple binding of "people-authority-device".

[0120] Necessity of RPA automation: Traditional pain points: Manual login to the trading system has problems such as low verification code recognition efficiency (depending on manual input), password management risks (plain text storage), and operation delays (manual response minutes). Technical advantages: OCR+API realizes fully automatic login and command execution, reducing operation time from minutes to seconds; Encrypted storage and dynamic decryption mechanisms eliminate the risk of password leakage and improve account security.

[0121] Adaptive design of ABAC model: Dynamic permission logic: Traditional RBAC (role-based access control) cannot adapt to complex scenarios (such as temporary policies restricting transactions in a certain area). ABAC achieves fine-grained control through dynamic combination of attributes; For example, when the policy restricts quota trading in the 2023 pilot area, ABAC automatically prohibits all operations containing the "pilot area + 2023" attribute without the need to manually adjust role permissions.

[0122] Compliance protection of smart contracts: On-chain strategies and regulatory rules are updated synchronously (such as new regulations from the Ministry of Ecology and Environment) to ensure real-time compliance of permission verification; Decentralized verification avoids the risk of single-point tampering and improves the system's ability to resist attacks.

[0123] Biometrics and on-chain authorization collaboration: Two-factor security enhancement: Biometric authentication solves the problem that traditional passwords are easy to crack, and on-chain authorization ensures that the authentication results cannot be denied; Significance of physical device linkage: Ukey smart lock binds digital permissions to physical devices to prevent unauthorized persons from accessing the system through illegal devices.

[0124] Specifically, in the carbon asset trading intelligent risk control method of the present invention, the authority management includes: Key operations require fingerprint and face recognition verification. The fingerprint and face recognition verification uses a zero-knowledge proof algorithm. The biometric hash value is stored in the blockchain distributed node. After the verification is passed, it is authorized to execute by the smart contract; Biometric verification uses a zero-knowledge proof algorithm, and the specific process is as follows: Hash generation: When a user registers, the hash value of the fingerprint feature points (bifurcation points, endpoints) and face depth information (72 biometric feature points) is generated through the SHA-3-256 algorithm, and then encrypted and stored in the IPFS distributed node; On-chain comparison: During verification, the local device generates a temporary biometric hash, and the smart contract calls the pre-stored hash on the chain for comparison, returning only the verification result (pass / reject) without transmitting the original biometric data; Dynamic authorization: After verification, the smart contract generates a one-time ED25519 signature token to authorize key operations (such as large transactions). The token is valid for ≤5 minutes.

[0125] Access to physical devices is controlled through the Ukey smart lock and on-chain authorization linkage.

[0126] Synergy mechanism of biometric verification: Multimodal feature acquisition: Fingerprint recognition: Use a capacitive sensor to obtain fingerprint pattern feature points (such as bifurcation points and endpoints) and generate a 256-bit hash value; Face recognition: Capture facial depth information through a 3D structured light camera to extract key features (such as eye distance and nose bridge outline) to avoid forgery of two-dimensional photos.

[0127] On-chain verification process: Encrypted storage: When a user registers, the fingerprint and facial feature hash values ​​are encrypted and written into the blockchain (such as IPFS distributed storage + symmetric encryption); Real-time comparison: During verification, the local device collects biometrics and generates a temporary hash, and the smart contract calls the pre-stored hash on the chain for comparison; Dynamic authorization: After two-factor authentication is passed, the smart contract generates a one-time token to authorize key operations (such as large transactions).

[0128] Linkage control of Ukey smart lock: Hardware device binding: Digital identity association: Each Ukey has a built-in unique digital certificate, which is bound to the blockchain account address to ensure that the device corresponds to the user identity; Physical operation audit: Ukey plug-in and unplugging, and usage records are uploaded to the chain in real time through the IoT module, forming a dual log of "physical operation-chain record".

[0129] On-chain authorization unlocking logic: Permission application: The user initiates a request to use the device, triggering the smart contract to verify its operation permissions (such as whether it has the quota transfer permission); Dynamic unlocking: After the contract verification is passed, an encrypted instruction (such as AES-256 encryption) is sent to Ukey to drive the electronic lock tongue to unlock; Timeout lock: After the device is accessed, if there is no operation for more than a preset time (such as 5 minutes), the smart contract will automatically trigger the lock instruction.

[0130] Compliance execution guarantee of smart contracts: Policy Preset and Update: Compliance rules built into the contract; Regulators can update policies (such as adding iris recognition requirements) through on-chain governance voting.

[0131] Immutability design: After the authorization record (such as the two-factor verification result of a transaction) is written into the blockchain, it can only be appended and cannot be modified, making the audit traceability credible.

[0132] Solve the defects of traditional permission management: Root of the problem: The traditional carbon trading system authority management relies on static passwords or single biometrics, which poses the following risks: Passwords are vulnerable to brute force or phishing attacks; A single biometric feature can be forged by a high-quality imitation model (such as a silicone fingerprint film); Physical devices (such as Ukey) are out of system control and there is a risk of misappropriation.

[0133] Technical advantages: Two-factor authentication: fingerprint + face dual biometrics greatly reduce the possibility of fraud (two biometrics must be forged at the same time); On-chain dynamic authorization: One-time tokens avoid the risk of leakage caused by long-term validity of permissions; Device-chain linkage: The use of Ukey requires real-time authorization on the chain to prevent illegal use of the device after it is lost.

[0134] Biodata security and privacy protection: Decentralized storage: Biometric hashes are distributed and stored in blockchain nodes rather than centralized databases, reducing the risk of single-point leakage; Zero-knowledge proof application: During verification, only hash values ​​are compared, and the original biometric data is not transmitted, which meets the privacy protection requirements.

[0135] A secure closed loop between the physical and digital worlds: Operation traces: Ukey usage records and on-chain authorization logs are stored as dual evidence, supporting the full traceability of "who-when-which device-performed what operation"; Defense in Depth: First layer: biometric verification ensures the operator’s identity is legitimate; Second layer: Smart contract verification ensures compliance of operation permissions; The third layer: physical equipment linkage prevents unauthorized terminal access.

[0136] Summary of technical effects: Improved security level: Two-factor biometric verification + on-chain dynamic authorization reduces the risk of fraudulent use of key operations to near zero; Compliance and auditability: Biodata storage and permission logs meet compliance requirements; Improve the defense system: realize the triple binding of "identity-authority-device" and build a secure closed loop that integrates physical and digital elements.

[0137] Specifically, in the carbon asset trading intelligent risk control method of the present invention, in step 4: The graph neural network model constructs a multi-layer graph structure of enterprises, units and regions, inputs unit types and regional policy factors, and outputs the optimal internal quota adjustment path based on unit type matching and regional policy constraints; After the smart contract monitors the completion of the enterprise's emissions verification, it automatically calls the registration system API to transfer quotas.

[0138] Technical details and explanation (Step 4: Intelligent quota management optimization) Graph Neural Network (GNN) modeling logic: Multi-layer graph structure definition: Enterprise layer: Nodes represent enterprises, and attributes include enterprise ID, industry, and historical performance records; Unit layer: Nodes represent power generating units / production equipment, and attributes include unit type (supercritical coal-fired unit, subcritical gas-fired unit), rated power, and emission intensity; Regional layer: Nodes represent administrative regions, and attributes include regional emission reduction targets, renewable energy consumption weights, and quota adjustment policies.

[0139] Edge relationship construction: Enterprise-generator edge: represents the generator equipment owned by the enterprise, and the weight is the enterprise's dependence on the generator (such as the proportion of emissions contributed by the generator); Unit-region edge: represents the region to which the unit belongs, and the weight is the constraint intensity of regional policies on unit emissions (for example, supercritical units must give priority to meeting regional low-carbon indicators).

[0140] Input data processing and feature fusion: Unit type matching calculation: Technical characteristic coding: Map the unit type (such as supercritical, subcritical) to a characteristic vector (such as [0.9, 0.1] represents a high-efficiency unit); Matching score: Calculate the technical similarity between units in the same region (e.g., the matching degree between supercritical units is higher than that between supercritical and subcritical units) as the weight basis for path optimization.

[0141] Regional policy factor embedding: The policy text is parsed into structured rules through NLP and converted into edge constraints of the graph neural network (such as setting the edge weight for prohibiting cross-provincial adjustments to infinity).

[0142] Optimal adjustment path generation mechanism: Path search algorithm: Based on the node embedding results of the graph neural network, the improved Dijkstra algorithm is used to search for the path with the lowest adjustment cost. The cost includes: Technology matching cost: Additional compensation is required for inter-unit transfers with low matching (e.g., the difference in technology upgrades is required for transfers from subcritical units to supercritical units); Policy compliance costs: Cross-regional adjustments are subject to policy restriction costs (if the adjustment ratio exceeds the regional quota, additional fees will be charged).

[0143] Dynamic adjustment logic: When regional policies are updated (such as a province temporarily relaxing transfer restrictions), the graph neural network recalculates edge weights and generates a new optimal path.

[0144] Smart contract driven quota transfer: Check trigger conditions: The enterprise emission verification data is stored on the blockchain (such as signature confirmation by a third-party verification agency), and the smart contract is automatically triggered after detecting the completion of the on-chain verification event; API call security guarantee: The smart contract generates quota transfer instructions and uses the OAuth 2.0 protocol to interact with the registration system. The token authority is limited to "write transfer records only" to prevent unauthorized operations. The transfer results are written back to the blockchain to generate an electronic certificate (including timestamp and transaction hash) for regulatory audit tracing.

[0145] Necessity of graph neural network modeling: Root cause of the problem: Traditional quota adjustment relies on manual experience and it is difficult to quantify the technical differences between units and the impact of regional policies, resulting in low adjustment efficiency and high compliance risks.

[0146] Technical advantages: Multi-layer graph structure: visualizes enterprise, unit, regional entities and their relationships, providing a structured data foundation for quantitative analysis; Dynamic weight adjustment: Update edge constraints in real time through policy factors to ensure that the path meets the latest regulatory requirements.

[0147] Synergy between path optimization and policy constraints: Matching priority logic: prioritize matching of units of the same type to reduce technical adaptation costs (no additional modification is required between units of the same type); Policy hard constraints: Through edge weight restrictions (such as prohibiting cross-provincial edges), policy requirements are enforced to avoid violation risks.

[0148] Trusted interaction between smart contracts and external systems: Data credibility guarantee: The verification results are stored on the blockchain to ensure that the triggering conditions are authentic and reliable; Minimized permission design: API calls only allow transfer operations, and token permissions are limited to prevent malicious tampering of other data.

[0149] Specifically, in the intelligent risk control method for carbon asset trading described in the present invention, the quota adjustment path gives priority to matching the surplus quotas of the same type of units in the same region.

[0150] Definition and screening mechanism of “units of the same type in the same region”: Unit type classification standards: Technical characteristics classification: Based on the energy efficiency and emission intensity of the units, the units are divided into supercritical, subcritical, gas combined cycle and other types (e.g., CO2 emission intensity of supercritical coal-fired units ≤ 800g / kWh); Regional affiliation identification: The region to which the unit belongs is uniquely identified by an administrative code (such as a provincial administrative division code) to ensure clear regional boundaries.

[0151] Surplus quota determination: Real-time data monitoring: Obtain real-time emission data of the unit from the blockchain, and calculate the surplus amount based on the total annual quota (surplus amount = total quota amount - actual emission amount); Dynamic threshold trigger: When the surplus of a unit exceeds 20% of the regional average surplus level, it is marked as an “adjustable surplus unit”.

[0152] Priority matching logic design: Technical matching priority: Synergistic advantages of similar units: The technical parameters (such as emission factor and thermal efficiency) of the same type of units are similar, and quota adjustment does not require additional technical transformation and adaptation, reducing friction costs; Case description: The surplus quota of supercritical unit A in a certain province is transferred to supercritical unit B in the same province, which can be directly used to offset the emission gap of B without calculating the technical difference compensation.

[0153] Regional policy constraints take precedence: Compliance guarantee: Regional policies often limit the proportion of cross-regional adjustments (for example, the "Carbon Emission Trading Management Measures" stipulates that cross-provincial adjustments are ≤5%). Prioritizing adjustments within regions can avoid policy violations. Utilization of policy incentives: Local governments provide subsidies for quota adjustments within the region (such as exemption from adjustment fees) to further reduce corporate costs.

[0154] Priority implementation in graph neural networks: Edge weight assignment rules: Edges of the same type in the same region: The weight is set to the lowest value (such as weight = 1), and they are selected by the algorithm first; Cross-regional or heterogeneous edges: The weight is calculated by superimposing the policy restriction ratio and the technical difference (e.g. weight = regional adjustment coefficient × technical difference coefficient); Path search optimization: In the embedding space output by the graph neural network, a greedy algorithm is used to traverse low-weight edges first to ensure that the path meets the "same area and same type first" rule; If there is not enough surplus quota in the region, the algorithm will automatically expand to heterogeneous units across regions, but additional costs will be added (such as policy fines + technical compensation).

[0155] Solve the inefficiency problem of traditional adjustment: Traditional pain points: Manual dispatching relies on experience and judgment, which makes it difficult to quickly locate the optimal unit and easily ignores regional policy restrictions, resulting in dispatch delays or violations; Technical advantages: Data-driven decision-making: quantify unit matching and policy costs through graph neural networks to generate the global optimal path; Embedded rule execution: The priority matching logic is enforced through algorithm weights to avoid subjective biases caused by human operations.

[0156] Synergy between technology matching and policy compliance: Technology adaptation and cost reduction: The technical parameters of the same type of units are consistent, reducing the additional transformation costs caused by technical differences (such as no need to adjust monitoring equipment); Regional policy compliance: Priority intra-regional adjustments naturally comply with policy restrictions (such as avoiding triggering cross-provincial transaction reviews), while using local incentive policies to reduce adjustment costs.

[0157] Dynamic expansion flexibility: Fault-tolerance mechanism: When there is no surplus of the same type of units in the region, the algorithm automatically searches for cross-regional or heterogeneous units in order of increasing cost to ensure the feasibility of adjustment; Cost transparency: The costs of all potential paths (technology + policy) are quantified in the graph structure to help companies make rational decisions.

[0158] Summary of technical effects: Improved efficiency: The priority matching mechanism reduces the search time for adjustment paths from hours to seconds, improving the turnover rate of carbon assets; Cost optimization: Reducing technical compensation and policy fines is estimated to reduce the total adjustment cost by more than 30%; Compliance guarantee: embedded regional policy rules to eliminate the risk of illegal adjustments.

[0159] Specifically, in the intelligent risk control method for carbon asset trading according to the present invention, in step 5: The reinforcement learning algorithm is the PPO algorithm, which abstracts the carbon market as a Markov decision process. The state space includes price, policy text vector, and enterprise emission data; The policy semantic parsing performs entity extraction and intention classification on the policy document through the BERT model. The extracted entities include the quota carry - over ratio, and the intention classification includes tightening or loosening policy orientations.

[0160] Carbon market modeling logic of the PPO algorithm: Definition of Markov decision process (MDP): State space (State): Price features: real - time carbon price, historical price volatility, market depth (buy and sell order volume); Policy text vector: semantic vector encoded by the BERT model for the policy document (e.g., "tightening policy" is encoded as [-0.8, 0.2]); Enterprise emission data: enterprise real - time emissions, quota holdings, historical compliance rate.

[0161] Action space (Action): risk control threshold adjustment (such as volatility tolerance ±5%), trading strategy selection (such as order price offset ±2%).

[0162] Reward function (Reward): comprehensively calculated based on risk control effect (such as early warning accuracy), trading income (such as profit and loss of hedging strategy), and compliance score (such as number of policy violations).

[0163] Policy optimization mechanism: Offline training: pre - train the agent using historical carbon market data to learn basic risk control rules; Online learning: input real - time market data into the PPO algorithm, and dynamically adjust policy parameters (such as threshold sensitivity) through trial - and - error to adapt to market mutations.

[0164] Technical implementation of policy semantic parsing: BERT model processing flow: Text pre - processing: tokenize the policy document, remove stop words (such as "de", "he"), and extract the core paragraphs (such as clauses related to quota management); Entity extraction: Quota carry - over ratio: identify numerical entities in the policy (such as "the carry - over ratio shall not exceed 5%" → 5%); Policy object: extract the affected entities (such as "key emission units", "pilot area enterprises").

[0165] Intention classification: Tightening policy orientation: Characteristic words include "strict restrictions", "reductions" and "prohibitions"; Loose policy orientation: characteristic words include “encouragement”, “relaxation” and “pilot”.

[0166] Semantic vector generation: The classification results (such as "tightening") and entities (such as "carry-over ratio 5%") are concatenated into a policy feature vector and input into the state space of the PPO algorithm.

[0167] Strategy feedback and closed-loop optimization: Dynamically adjust triggers: When policy semantic analysis detects the intention of "tightening", the PPO algorithm automatically reduces volatility tolerance (for example, from 10% to 8%) to reduce high-risk transactions; When an "increase in the carry-forward ratio" is detected, the algorithm increases the proportion of forward contract positions and locks in low-priced quotas in advance.

[0168] Cross-step collaboration: Real-time feedback of strategy adjustment suggestions to: Step 2 (Risk Control): Update dynamic thresholds; Step 3 (Trading): Adjust the order strategy; Step 4 (Quota Management): Optimize the adjustment path weights.

[0169] Scene adaptability of PPO algorithm: Root of the problem: Traditional risk control strategies rely on static rules and are unable to cope with high-frequency fluctuations in the carbon market and sudden policy changes (such as sudden quota reductions).

[0170] Technical advantages: Reinforcement learning characteristics: The PPO algorithm learns dynamic optimization strategies through trial and error, which is more adaptable to nonlinear market changes than fixed rule models; Multi-factor state space: Integrate price, policy, and enterprise data to avoid single-dimensional decision-making bias.

[0171] Necessity of policy semantic analysis: Manual analysis defects: Policy documents rely on manual interpretation, which may lead to subjective misjudgment and delayed response risks (e.g., adjusting strategies several days after the policy is released); Automated parsing value: The BERT model captures policy semantic changes in real time, improving parsing speed and consistency; The intent classification and entity extraction results provide the algorithm with quantifiable policy features to support data-driven decision-making.

[0172] The design logic of the closed-loop feedback mechanism: Real-time guarantee: Strategy adjustment suggestions are fed back to the previous steps within seconds, ensuring that the response speed of the entire process is synchronized with market changes; Risk diversification: Coordinated adjustment of multiple steps avoids overfitting of a single link (e.g., adjusting only the threshold may ignore the defects of the trading strategy).

[0173] Logical explanation of the technical solution of the present invention to solve the technical pain points: Correspondence between technical pain points and solutions: Pain point 1: Data silos lead to information fragmentation and lack of credibility: Traditional problems: Carbon emissions, trading, and registration data are scattered across different systems with inconsistent formats, and manual integration is inefficient and error-prone.

[0174] Solution: Cross-system data integration: Build a data middle platform through ETL tools and API gateway to standardize multi-source data (such as fuel consumption converted according to ISO); Blockchain evidence storage: key data (quota allocation, transaction instructions) is written into the Hyperledger Fabric consortium chain through smart contracts, making the data tamper-proof and traceable; Cross-chain interoperability protocol based on the Polkadot framework: realize the trusted flow of quotas between the main chain and local chains, and solve the problem of data interoperability between heterogeneous systems.

[0175] Pain point 2: Delayed response leads to failure of risk control: Traditional problem: Relying on manual review and offline analysis, it is difficult to capture market fluctuations and policy changes in a timely manner.

[0176] Solution: Stream computing framework (Apache Flink): real-time access to market, emission, and policy data, and sliding window statistics of dynamic thresholds (such as price volatility > 10%); LSTM+Attention model: predicts carbon price trends and evaluates gap probability, and generates graded warning signals in combination with Monte Carlo simulation; Smart contract driven hedging: When the gap exceeds the threshold, the hedging instruction (such as buying futures) is automatically triggered, and the response time is reduced from hours to seconds.

[0177] Pain point 3: Inefficiency and operational safety risks: Traditional problems: Manual operations are prone to errors, permission management is loose, and there is a risk of password leakage and unauthorized operation.

[0178] Solution: RPA Automation Process: Trading plan decomposition: According to the group's annual trading strategy, the monthly trading plan is imported into the system and automatically decomposed into daily trading instructions; Approval process trigger: After the transaction plan is submitted, an online approval work order is automatically generated, which is reviewed by the risk control officer and then pushed to the smart contract for execution; Device permission linkage: After the transaction instruction is approved, the Ukey smart lock automatically unlocks the corresponding account permissions to make the operation compliant; Zero Trust Architecture (ABAC Model): Dynamically generate minimum permission sets based on roles, targets, and operation attributes, and superimpose biometric verification (fingerprint + face) on key operations; Ukey smart lock linkage: The use of physical devices requires on-chain authorization, realizing the triple binding of "people-authority-device".

[0179] Pain point 4: Inefficient quota management and rigid policies: Traditional problems: Manual adjustments rely on experience, it is difficult to quantify technical differences and policy impacts, and strategy adjustments lag behind.

[0180] Solution: Graph Neural Network (GNN) modeling: construct a multi-layer graph of enterprise-unit-region, and output the optimal adjustment path based on unit matching and policy constraints (giving priority to units of the same type in the same region); Proximal Policy Optimization (PPO) reinforcement learning algorithm: abstracts the carbon market into a Markov decision process and dynamically adjusts risk control thresholds and trading strategies; BERT policy analysis: Identify policy entities (such as carry-over ratio) and intentions (tightening / easing), and provide real-time feedback to the policy engine.

[0181] Technical solution synergy and logical closed loop: Positive flow from data layer → risk control layer → execution layer; The data center integrates multi-source data to provide input for the risk control engine; The risk control model outputs early warning signals in real time, driving RPA and smart contracts to execute transactions or adjustments; The execution results are written back to the blockchain for evidence storage, forming an auditable closed loop.

[0182] Feedback optimization from strategy layer to execution layer; Reinforcement learning algorithms dynamically optimize strategies based on market feedback (e.g., adjusting volatility tolerance); The policy analysis results update the edge weights of the graph neural network in real time (such as limiting the cross-provincial adjustment ratio) to ensure path compliance.

[0183] Balanced design between safety and efficiency: Automated execution improves efficiency, while zero-trust architecture and biometric authentication ensure operational security; Blockchain evidence storage and smart contract verification double-ensure data and transaction compliance.

Claims

1. The intelligent risk control method for carbon asset trading is characterized by: include: Step 1: Build a carbon asset data middle platform, integrate blockchain distributed ledger and API gateway, and connect to the carbon emission trading system, registration system, thermal power enterprise fuel-side carbon emission data automatic collection system and smart lock system through robotic process automation technology; Step 2: Based on the Apache Flink streaming computing framework, the price volatility, transaction volume deviation and quota supply and demand gap risk indicators are calculated in real time, and the LSTM+Attention model is used to predict the carbon price trend and generate risk warning signals; Step 3: Drive the transaction plan decomposition and approval process through smart contracts, and perform dynamic permission management based on the ABAC model. Key operations require fingerprint and face recognition verification, and link with the Ukey smart lock system to control device access permissions; Step 4: Use the graph neural network model to build a multi-layer graph structure of enterprises, units and regions, output the optimal quota adjustment path that prioritizes the same type of units in the same region, output the optimal quota adjustment path based on graph neural network modeling, and drive automatic fulfillment operations through smart contracts; Step 5: Use the proximal strategy optimization reinforcement learning algorithm to optimize the risk control strategy, and generate adjustment suggestions based on the policy semantic analysis of the BERT model. The adjustment suggestions are updated in real time through smart contracts to update the dynamic risk threshold of step 2, the authority strategy of step 3, and the quota adjustment weight of step 4.

2. The intelligent risk control method for carbon asset trading according to claim 1 is characterized in that: The step 1 includes: Use ETL tools to extract and clean multi-source data from carbon emission data collection systems, trading systems, and registration systems in real time, and establish a unified data model; Quota allocation records, transaction instructions and performance certificates are used as key data and written into the blockchain distributed ledger through smart contracts.

3. The intelligent risk control method for carbon asset trading according to claim 2 is characterized in that: The blockchain distributed ledger adopts a cross-chain interoperability protocol based on the Polkadot framework to establish a trusted channel between the carbon market main chain and the carbon market sub-chain, and realizes the cross-chain mapping and circulation of quota assets through smart contracts. The carbon market main chain is the core blockchain with unified carbon trading rules, and the carbon market sub-chain is a dedicated blockchain for local pilot markets or specific industries.

4. The intelligent risk control method for carbon asset trading according to claim 1 is characterized in that: In step 2: The streaming computing framework is Apache Flink, which accesses market data, corporate carbon emission data streams and policy change text data in real time, and inputs the processed data into the LSTM+Attention model to predict carbon price trends and evaluate the probability of a quota supply and demand gap.

5. The intelligent risk control method for carbon asset trading according to claim 4 is characterized in that: The dynamic intelligent risk control includes: Generate quota gap warning signals through Monte Carlo simulation, which generates random scenarios based on historical quota transaction data in the past three years; When it is detected that the quota gap exceeds the dynamic threshold described in step 2, the smart contract is automatically called to generate a hedging instruction.

6. The intelligent risk control method for carbon asset trading according to claim 1 is characterized in that: In step 3: The robotic process automation program logs into the carbon emission trading system through optical character recognition technology, simulates user operations to automatically execute trading instructions, and drives the trading plan decomposition and approval process through smart contracts; dynamic permission management is based on the ABAC model, dynamically generates the minimum permission set according to role attributes, transaction subject attributes and operation type attributes, and key operations require fingerprint and face recognition verification. The access to physical equipment is controlled by the Ukey smart lock and the on-chain authorization linkage, which is used to achieve three-level risk prevention and control, including prior authorization, in-process monitoring and post-audit; Dynamic permission management is based on the ABAC model. It dynamically generates the minimum permission set according to role attributes, transaction subject attributes, and operation type attributes, and verifies permissions through smart contracts.

7. The intelligent risk control method for carbon asset trading according to claim 6 is characterized in that: The rights management includes: Key operations require fingerprint and face recognition verification. The fingerprint and face recognition verification uses a zero-knowledge proof algorithm. The biometric hash value is stored in the blockchain distributed node. After the verification is passed, it is authorized to execute by the smart contract; Access to physical devices is controlled through the Ukey smart lock and on-chain authorization linkage.

8. The intelligent risk control method for carbon asset trading according to claim 1 is characterized in that: In step 4: The graph neural network model constructs a multi-layer graph structure of enterprises, units and regions, inputs unit types and regional policy factors, and outputs the optimal internal quota adjustment path based on unit type matching and regional policy constraints; After monitoring the completion of the enterprise's emissions verification, the smart contract automatically calls the registration system API to transfer quotas.

9. The intelligent risk control method for carbon asset trading according to claim 8 is characterized in that: The quota adjustment path gives priority to matching the surplus quotas of the same type of units in the same region.

10. The intelligent risk control method for carbon asset trading according to claim 1, characterized in that: In step 5: The reinforcement learning algorithm is the PPO algorithm, which abstracts the carbon market into a Markov decision process, and the state space includes prices, policy text vectors and corporate emission data; The policy semantic analysis uses the BERT model to perform entity extraction and intent classification on the policy document. The extracted entities include the quota carry-over ratio, and the intent classification includes tightening or loosening policy orientation. The policy text vector is generated by semantically encoding the policy document through the BERT model.

Images