A digital renminbi-based payment risk assessment method

Abstract

The application belongs to the technical field of computer imaging, and particularly relates to a payment risk assessment method based on digital renminbi, which is characterized by the following technical scheme: a payment risk assessment method based on digital renminbi, comprising the following steps: S1: when a digital renminbi payment transaction is triggered, calling risk control rules embedded in a smart contract associated with the payment transaction to perform automatic risk checking before the transaction; the application significantly improves the security of double-offline payment. By deploying a lightweight gradient boosting decision tree model in the wallet security chip, transaction risk pre-evaluation can be completed in an offline state, and local risk preliminary blocking is realized; meanwhile, the transaction hash chain is used to lock the offline transaction sequence, and after networking, conflict detection is performed with the unspent transaction hash set of the main ledger, so that post-traceable verification of double-spend risk is realized, and the technical gap that the central risk control is completely ineffective in the double-offline scenario is filled.

Description

Technical Field

[0001] This invention belongs to the field of computer imaging technology, specifically referring to a payment risk assessment method based on digital RMB. Background Technology

[0002] As the pilot program for the digital yuan continues to expand, its payment scenarios have gradually extended from traditional online e-commerce and offline QR code payments to dual offline environments such as underground transportation, remote areas, and payments between IoT devices. The digital yuan combines the programmability of "digital currency" with the unconditional legal tender status of "legal tender," and its risk characteristics are fundamentally different from those of third-party payments, making it difficult for traditional risk control models to be directly adapted.

[0003] In existing technologies, most digital RMB payment risk assessment solutions directly transplant the risk control architecture of third-party payment platforms. Specifically, this involves collecting feature data after a transaction occurs, calculating a risk score through a centralized risk control engine, and then triggering corresponding actions based on the score results. This approach has at least five core flaws: First, the risk assessment for offline payments is severely lagging. Existing solutions rely on real-time calculations by a central server. When both the payer and payee are offline, transaction data cannot be uploaded in real time, rendering the risk control engine completely ineffective. Audits can only be conducted retrospectively after the network is connected, making it impossible to prevent risks before or at the moment of transaction activation.

[0004] Second, there is a lack of risk assessment for smart contracts. The digital yuan supports advanced functions such as conditional payments and targeted payments through smart contracts, but existing risk control models do not include the depth of smart contract calls, abnormal preconditions, and contract code vulnerabilities in the assessment dimensions. The execution of smart contracts lacks embedded risk control checks, resulting in significant risk exposure.

[0005] Third, it is difficult to balance controllable anonymity with the need for anti-fraud. The digital yuan follows the principle of "anonymity for small amounts and traceability for large amounts." Existing solutions mostly adopt a "one-size-fits-all" anonymity strategy, which cannot dynamically adjust the anonymity strength according to the real-time risk level of the transaction. This results in excessive exposure of user privacy in low-risk scenarios, or excessively high anonymity in high-risk scenarios, which is not conducive to tracking.

[0006] Fourth, the risk assessment model is static and simplistic. Existing solutions only focus on snapshot characteristics of transactions (amount, time, location), failing to utilize the unique identifier of the digital yuan and its full traceability to construct a profile of the currency's circulation behavior. This makes it impossible to identify structured risk patterns such as money laundering and splitting and refinancing within long chains.

[0007] Fifth, the risk management mechanism is rigid. The existing risk management measures are fixed, usually involving manual review or simple interception. They cannot automatically classify and implement differentiated measures according to the risk level, resulting in poor response time. Furthermore, the results of the measures are not fed back into the risk control model in a closed loop to drive continuous model optimization. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes a payment risk assessment method based on digital RMB.

[0009] The technical solution adopted by this invention is as follows: This invention provides a payment risk assessment method based on digital RMB, comprising the following steps: S1: When a digital RMB payment transaction is triggered, the risk control rules embedded in the smart contract associated with the payment transaction are invoked to perform automated risk checks before the transaction. S2: When the payment transaction is in a dual offline state, the lightweight risk assessment model built into the wallet security chip is used to perform a local pre-assessment of the payment transaction, and during the offline period, the transaction hash of the current transaction is generated by linking the hash values ​​of the previous transactions through a hash function, and a transaction hash chain for locking the order of offline transactions is constructed; after the wallet is connected to the network, the transaction hash chain is used to detect conflicts with the confirmed transaction hash tree of the main ledger, and a post-verification of the double-spending risk for offline transactions is performed. S3: Based on the real-name authentication level and real-time risk level of the payment transaction parties, the anonymity processing level of this transaction is dynamically determined by querying a preset hierarchical anonymity risk mapping table, and a transaction certificate matching the anonymity processing level is generated using zero-knowledge proof technology; the transaction certificate is used to prove to the regulatory node that the transaction complies with preset compliance requirements without disclosing the specific identity information of the counterparty, and enables the regulator to selectively disclose the specified information contained in the transaction certificate using the key it holds; S4: Integrate the multi-dimensional features of the payment transaction, combine them with the full lifecycle dynamic risk profile constructed based on the unique identifier of the digital RMB, and use an adaptive algorithm to calculate the real-time risk score; wherein, the full lifecycle dynamic risk profile includes a behavioral sequence feature vector formed based on the on-chain circulation path of the digital RMB, and is triggered to update after each transaction is completed; S5: Based on the real-time risk score, trigger the programmable risk handling mechanism and execute the handling action corresponding to the current risk level.

[0010] Furthermore, the risk control rules embedded in the smart contract include one or more of the following: transaction amount threshold check, counterparty wallet address blacklist verification, and contract execution precondition verification.

[0011] Furthermore, the step of using a lightweight risk assessment model built into the wallet security chip to perform local pre-assessment of the payment transaction, constructing a transaction hash chain during offline operation, and performing post-event verification of double-spending risk after connecting to the network further includes: The lightweight risk assessment model is a binary classification model based on gradient boosting decision trees. It extracts feature data of offline transactions and outputs a risk pre-score. If the risk pre-score exceeds a preset threshold, the offline transaction is directly rejected. If the risk pre-score does not exceed the preset threshold, the transaction is executed and the transaction hash of the i-th offline transaction is calculated as H_i = Hash(T_i || H_{i-1}) during the offline period, where T_i is the current transaction data and H_{i-1} is the hash value of the previous transaction, thus constructing a transaction hash chain that locks the order of offline transactions; After the wallet is connected to the network, each transaction hash in the transaction hash chain is compared with the set of unspent transaction hashes that have been confirmed on the digital RMB main ledger. If there is a matching hash value, the transaction is determined to be a double-spending transaction, and subsequent transactions involving double-spending are rejected or rolled back. If there is no conflict, the transaction batch in the transaction hash chain is submitted to the main ledger for confirmation.

[0012] Furthermore, the hierarchical anonymity risk mapping table stores a combination mapping relationship between wallet real-name level, real-time risk level and anonymity processing level; among them, transactions with low risk level and high real-name level are mapped to strong anonymity processing level, allowing the use of ring signature or group signature to hide the transaction party identifier; high risk level transactions are mapped to controllable anonymity processing level, which forces the use of decryptable identifiers encrypted based on the regulator's public key. The zero-knowledge proof technology is used to generate non-interactive zero-knowledge proofs, which prove that the transaction amount is within the compliant range and that the wallet addresses of both parties to the transaction belong to the set of certified addresses and are not blacklisted.

[0013] Furthermore, the multi-dimensional features include transaction amount, transaction time, transaction geographical location, device fingerprint, user's historical transaction behavior sequence, and smart contract call depth; The adaptive algorithm employs an online gradient boosting decision tree or an online deep neural network model, and continuously updates the model parameters online based on real-time feedback of the processing results.

[0014] Furthermore, the programmable risk management mechanism includes at least one of the following actions: freezing transaction funds, delaying transaction settlement, triggering an enhanced identity verification process for the transacting parties, and sending an early warning notification to the regulatory system; the actions are executed automatically through the smart contract.

[0015] Furthermore, in the case where the payment transaction is an offline transaction, after the wallet is connected to the internet, it also includes: The local pre-assessment records, transaction hash chains, and real-time risk scores of offline transactions are uploaded to the central risk control platform to update the full lifecycle dynamic risk profile and to serve as training data to optimize the lightweight risk assessment model.

[0016] A payment risk assessment system based on digital yuan, according to a payment risk assessment method based on digital yuan, includes: The smart contract risk control module is used to call the risk control rules embedded in the associated smart contract to perform automated risk checks before the transaction is triggered. The offline pre-evaluation module, deployed in the wallet's security chip, is used to perform local pre-evaluation of payment transactions using a built-in lightweight risk assessment model when both are offline. During the offline period, it generates a transaction hash chain by linking the hash values ​​of previous transactions using a hash function. After the wallet is connected to the network, it performs a post-event verification of double-spending risk for offline transactions by detecting conflicts between the transaction hash chain and the confirmed transaction hash tree in the main ledger. The tiered anonymity mapping module is used to query a preset tiered anonymity risk mapping relationship table based on the real-name level and real-time risk level of the transacting parties, dynamically determine the anonymity processing level of the transaction, and call the zero-knowledge proof module to generate transaction credentials that can be disclosed selectively. The dynamic risk assessment engine is used to integrate the multi-dimensional features of the payment transaction, combine them with the full lifecycle dynamic risk profile constructed based on the unique identifier of the digital RMB and recording the feature vector of the on-chain circulation behavior sequence, use an adaptive algorithm to calculate the real-time risk score, and trigger the update of the risk profile after each transaction is completed. The programmable response module is used to trigger and execute response actions corresponding to the current risk level based on the real-time risk score.

[0017] Furthermore, the offline pre-evaluation module is further used to: extract offline transaction feature data using a binary classification model based on gradient boosting decision trees and output a risk pre-score; reject transactions when the risk pre-score exceeds a threshold; during the offline period, calculate the transaction hash of the i-th transaction as H_i = Hash(T_i || H_{i-1}) to construct a transaction hash chain; after connecting to the network, compare each transaction hash in the transaction hash chain with the set of unspent transaction hashes confirmed in the main ledger; if there is a conflict, it is determined to be a double-spending transaction and subsequent transactions are rejected or rolled back; if there is no conflict, batch confirmation is submitted.

[0018] Furthermore, a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the method.

[0019] The beneficial effects achieved by the present invention using the above structure are as follows: (1) Significantly improve the security of dual offline payments. By deploying a lightweight gradient boosting decision tree model in the wallet security chip, transaction risk pre-assessment can be completed in the offline state, realizing localized initial risk blocking; at the same time, the offline transaction order is locked by constructing a transaction hash chain, and after connecting to the network, conflict detection is performed with the set of unspent transaction hashes in the main ledger, realizing the ex-post traceability verification of double-spending risk, filling the technical gap of complete failure of central risk control in dual offline scenarios.

[0020] (2) Fill the gaps in smart contract risk control. Risk control rules are directly embedded in the payment smart contract. Before the transaction is broadcast to the consensus network, the transaction amount threshold check, counterparty blacklist verification, and contract precondition verification check are automatically performed to eliminate the risk exposure during the execution of the smart contract.

[0021] (3) Achieve a dynamic balance between privacy protection and anti-fraud. Through a pre-set hierarchical anonymity risk mapping table, link the wallet real-name level with the real-time risk score. Automatically increase the anonymity protection strength for low-risk transactions, and forcibly reduce the anonymity level for high-risk transactions. Use the regulator's public key to encrypt the identifier. Combined with zero-knowledge proof technology, prove the compliance of the transaction without revealing the identity of the counterparty, so that the regulator can selectively disclose.

[0022] (4) Improve the accuracy and predictability of risk assessment. By using the unique identifier of digital RMB to construct a dynamic risk profile throughout the entire life cycle, the entire circulation link of the currency string from issuance to redemption is tracked, and behavioral sequence feature vectors are extracted. This upgrades risk assessment from a single-point snapshot to full-process behavioral analysis, which can effectively identify structured abnormal patterns that circumvent risk control through multi-hop transfers, decentralized convergence, and other methods. At the same time, by adopting adaptive algorithms such as online gradient boosting decision trees, the model parameters can be continuously updated online based on the feedback from the disposal, and the risk identification capability continues to improve with the running time.

[0023] (5) Achieve refined and closed-loop optimization of risk handling. The programmable handling module automatically performs differentiated handling actions such as normal settlement, delayed settlement, freezing funds, triggering enhanced verification, and sending early warning notifications based on real-time risk scores. The response is real-time and requires no manual intervention. The handling results are synchronously fed back to the dynamic risk assessment engine, forming a closed-loop learning link of "assessment → handling → feedback → optimization". The longer the system runs, the more accurate the risk control becomes. Detailed Implementation

[0024] The following are specific embodiments of the present invention, which further describe the technical solution of the present invention. However, the scope of protection of the present invention is not limited to these embodiments. All changes or equivalent substitutions that do not depart from the concept of the present invention are included within the scope of protection of the present invention.

[0025] Example 1: System Overall Architecture This embodiment provides a payment risk assessment system based on digital RMB. The system is deployed on top of the digital RMB payment infrastructure and includes the following functional modules: Smart contract risk control module: Deployed on the smart contract virtual machine, it works in conjunction with the digital RMB payment contract.

[0026] Offline pre-assessment module: embedded in the trusted execution environment of the wallet security chip (Secure Element, abbreviated as SE), it is responsible for local risk assessment of offline transactions and maintenance of the transaction hash chain.

[0027] Tiered Anonymity Mapping Module: Deployed in the middleware layer between the wallet client and the central bank's regulatory node, it is responsible for the dynamic scheduling of anonymity strategies.

[0028] Dynamic risk assessment engine: Running on a risk control server cluster of the central bank or authorized financial institutions, responsible for centralized real-time risk scoring.

[0029] Programmable handling module: It exists in the form of a smart contract sub-contract, receives risk assessment results and automatically triggers handling actions.

[0030] Example 2: Construction of a Dynamic Risk Profile Throughout the Life Cycle The dynamic risk assessment engine builds and maintains a dynamic risk profile for each digital yuan throughout its entire lifecycle.

[0031] The profile contains a behavioral sequence feature vector based on the on-chain circulation path of the digital RMB. The specific extraction method is as follows: track the complete circulation link of the coin string from minting, each transfer to the current holding, and record the following features of each hop: transaction amount, transaction timestamp, counterparty wallet type, counterparty risk level, and time interval between two adjacent hops.

[0032] The aforementioned feature vector is updated after each transaction is completed and confirmed on the blockchain. Newly added transactions are appended to the end of the vector, forming a dynamically growing sequence of behaviors. This profile serves as an important input dimension for the adaptive algorithm.

[0033] Example 3: Pre-trade Smart Contract Embedded Risk Control When a user initiates a digital yuan payment transaction, the payment request is first routed to the smart contract risk control module associated with the target payment contract.

[0034] This module embeds risk control rules, which specifically include the following three types in this embodiment: Transaction amount threshold check: Determines whether the current transaction amount exceeds the daily single transaction limit of the wallet. If it does, the transaction is rejected directly.

[0035] Counterparty wallet address blacklist verification: Query the set of addresses maintained by the blacklist contract. If the recipient's address exists in the blacklist, the transaction is rejected.

[0036] Contract execution precondition verification: If the payment transaction is associated with a conditional smart contract (such as a cash-on-delivery contract), then verify whether the preconditions have been met (such as whether the logistics status has been signed for). If not, the transaction is suspended and execution continues after the conditions are met.

[0037] The above checks are completed before the transaction is broadcast to the consensus network, thus achieving pre-emptive risk prevention.

[0038] Example 4: Local pre-evaluation of dual offline payments and transaction hash chain anti-double-spending verification When both parties to the payment are in a dual offline state with no network connection, the payment transaction is handled by the offline pre-evaluation module.

[0039] 4.1 Lightweight Risk Assessment Model The offline pre-evaluation module incorporates a binary classification model based on Gradient Boosting Decision Tree (GBDT). This model undergoes pruning and quantization compression to accommodate the resource constraints of the security chip, keeping the model file size under 50KB. During inference, it only requires traversing no more than 32 decision trees, with each tree having a depth of no more than 6 levels.

[0040] The model input features include: offline transaction amount, current number of consecutive offline transactions, time since last network connection, and the recipient's local credit score. The model output is a risk pre-score between 0 and 1. If the risk pre-score exceeds a preset threshold (set to 0.7 in this embodiment), the transaction is directly rejected within the security chip, and an offline rejection record is generated.

[0041] 4.2 Construction of Transaction Hash Chain If the risk pre-score does not exceed the threshold, transaction execution is allowed, and a transaction hash chain is built during the offline period to lock the offline transaction order. The specific method is as follows: Define the transaction hash of the i-th offline transaction as: H_i = Hash(T_i || H_{i-1}) Where T_i represents the current transaction data, H_{i-1} represents the hash value of the previous transaction, || indicates data concatenation, and the hash uses the national cryptographic SM3 hash algorithm. The H_0 of the first offline transaction is set to the latest transaction hash when the wallet last went online for synchronization.

[0042] Thus, all offline transactions are linked into an immutable chain of transaction hashes, with each transaction's hash value depending on all its preceding transactions.

[0043] 4.3 Double-spending conflict detection after network connection Once the wallet regains its network connection, the offline pre-assessment module uploads each transaction hash in the transaction hash chain to the dynamic risk assessment engine in batches.

[0044] The dynamic risk assessment engine compares each transaction hash in the transaction hash chain with the "set of confirmed but unspent transaction hashes" maintained on the digital yuan master ledger. The comparison rules are as follows: If a transaction hash already exists in the set, it means that the input was spent by another transaction path during the offline period, and is determined to be a double-spending transaction; For transactions identified as double-spending, and all offline transactions following those transactions, mark them as invalid and trigger a rollback process; If no conflict occurs among all transaction hashes, the entire transaction hash chain corresponding to the transaction batch is submitted to the main ledger for final confirmation.

[0045] The above mechanism enables post-transaction double-spending detection and batch confirmation for offline transactions, overcoming the inherent limitation that double-spending cannot be verified in real time in dual offline scenarios.

[0046] Example 5: Hierarchical Anonymity Risk Mapping and Zero-Knowledge Proof 5.1 Hierarchical Anonymity Risk Mapping Relationship Table The pre-defined hierarchical anonymity risk mapping table in this embodiment is as follows: 5.2 Zero-Knowledge Proof Generation After determining the anonymity level, the zero-knowledge proof module is invoked to generate a non-interactive zero-knowledge proof. This embodiment employs a combination of range proofs based on Bulletproofs and set membership proofs based on Merkle Trees to prove the following statement: The transaction amount is within the compliant range (i.e., does not exceed the single transaction limit specified for this wallet type). Both parties' wallet addresses belong to the set of addresses certified by the central bank and are not listed in the blacklist of contracts.

[0047] The generated zero-knowledge proof is broadcast along with the transaction as part of the transaction certificate. Each verification node can efficiently verify the validity of the proof without knowing the specific identity information. At the same time, the regulator holds the path decryption key of the Merkle Tree, which can selectively disclose the identity information hidden in the transaction certificate, achieving "controlled anonymity".

[0048] Example 6: Multi-dimensional Feature Fusion and Adaptive Risk Scoring The dynamic risk assessment engine integrates the following multi-dimensional features to calculate a real-time risk score: Transaction amount characteristics: absolute amount, deviation of amount from historical average; Transaction time characteristics: the time period in which the transaction occurs and its degree of matching with the user's active time window; Transaction geographic location characteristics: geographic coordinates based on device GPS or base station positioning, and distance from frequently used locations; Device fingerprint characteristics: unique device identifier, security chip status, and system integrity verification results; User's historical transaction behavior sequence: transaction frequency, counterparty concentration, and rapid inflow and outflow of funds in the past 30 days; Smart contract call depth: The number of nested call levels of contracts involved in the current transaction.

[0049] This embodiment employs Online Gradient Boosting Decision Tree (Online GBDT) as the adaptive algorithm, which supports incremental sample streaming training. Whenever the programmable processing module provides a processing result (such as a transaction being confirmed as fraud), the engine extracts the features and labels of that sample, updates the model parameters online, and continuously improves the scoring accuracy.

[0050] Example 7: Programmable Risk Management Mechanism The programmable response module automatically triggers the following actions based on the real-time risk score range: The aforementioned actions are executed automatically through smart contract code without human intervention, and the results are synchronously fed back to the dynamic risk assessment engine for online model updates.

[0051] Example 8: System Workflow Example Taking a typical offline payment transaction as an example, the complete workflow is as follows: Step 1: User A initiates a digital RMB payment to User B in an offline environment.

[0052] Step 2: The offline pre-evaluation module extracts transaction features, runs the GBDT lightweight model, and outputs a risk pre-score of 0.35, which does not exceed the threshold of 0.7, allowing the transaction.

[0053] Step 3: The security chip uses the previous transaction hash H_{i-1} to concatenate with the current transaction data T_i to calculate the SM3 hash, generate H_i, and write it into the local transaction hash chain.

[0054] Step 4: User A restores network connection, and the offline pre-assessment module uploads the transaction hash chain to the dynamic risk assessment engine.

[0055] Step 5: The dynamic risk assessment engine compares H_i with the set of unspent transaction hashes in the main ledger. No conflict is found, and it is determined to be a non-double-spending transaction.

[0056] Step Six: The hierarchical anonymity mapping module determines the anonymity processing level as A1 based on user A's real-name level L3 and current risk score (calculated as 25 points by combining transaction amount, historical behavior, and other characteristics). It then calls the zero-knowledge proof module to generate range proof and set membership proof.

[0057] Step 7: The transaction is broadcast to the network with a zero-knowledge proof attached, and the verification node verifies it successfully.

[0058] Step 8: The programmable processing module determines the risk level to be low based on a risk score of 25, executes normal settlement, and the transaction is credited to the account in real time.

[0059] Step 9: The dynamic risk assessment engine updates the full lifecycle risk profile of the coin string and adds the characteristics of this transaction to the behavior sequence vector.

[0060] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A digital renminbi-based payment risk assessment method, characterized in that: Includes the following steps: S1: When a digital RMB payment transaction is triggered, the risk control rules embedded in the smart contract associated with the payment transaction are invoked to perform automated risk checks before the transaction. S2: When the payment transaction is in a dual offline state, the lightweight risk assessment model built into the wallet security chip is used to perform a local pre-assessment of the payment transaction, and during the offline period, the transaction hash of the current transaction is generated by linking the hash values ​​of the previous transactions through a hash function, and a transaction hash chain for locking the order of offline transactions is constructed; after the wallet is connected to the network, the transaction hash chain is used to detect conflicts with the confirmed transaction hash tree of the main ledger, and a post-verification of the double-spending risk for offline transactions is performed. S3: Based on the real-name authentication level and real-time risk level of the payment transaction parties, the anonymity processing level of this transaction is dynamically determined by querying a preset hierarchical anonymity risk mapping table, and a transaction certificate matching the anonymity processing level is generated using zero-knowledge proof technology; the transaction certificate is used to prove to the regulatory node that the transaction complies with preset compliance requirements without disclosing the specific identity information of the counterparty, and enables the regulator to selectively disclose the specified information contained in the transaction certificate using the key it holds; S4: Integrate the multi-dimensional features of the payment transaction, combine them with the full lifecycle dynamic risk profile constructed based on the unique identifier of the digital RMB, and use an adaptive algorithm to calculate the real-time risk score; wherein, the full lifecycle dynamic risk profile includes a behavioral sequence feature vector formed based on the on-chain circulation path of the digital RMB, and is triggered to update after each transaction is completed; S5: Based on the real-time risk score, trigger the programmable risk handling mechanism and execute the handling action corresponding to the current risk level.

2. The payment risk assessment method based on digital renminbi according to claim 1, characterized in that: The risk control rules embedded in the smart contract include one or more of the following: transaction amount threshold check, counterparty wallet address blacklist verification, and contract execution precondition verification.

3. The payment risk assessment method based on digital RMB according to claim 2, characterized in that: The method of using a lightweight risk assessment model built into the wallet's security chip to perform local pre-assessment of the payment transaction, constructing a transaction hash chain during offline operation, and performing post-event verification of double-spending risk after connecting to the network further includes: The lightweight risk assessment model is a binary classification model based on gradient boosting decision trees. It extracts feature data of offline transactions and outputs a risk pre-score. If the risk pre-score exceeds a preset threshold, the offline transaction is directly rejected. If the risk pre-score does not exceed the preset threshold, the transaction is executed and the transaction hash of the i-th offline transaction is calculated as H_i = Hash(T_i || H_{i-1}) during the offline period, where T_i is the current transaction data and H_{i-1} is the hash value of the previous transaction, thus constructing a transaction hash chain that locks the order of offline transactions; After the wallet is connected to the network, each transaction hash in the transaction hash chain is compared with the set of unspent transaction hashes that have been confirmed on the digital RMB main ledger. If there is a matching hash value, the transaction is determined to be a double-spending transaction, and subsequent transactions involving double-spending are rejected or rolled back. If there is no conflict, the transaction batch in the transaction hash chain is submitted to the main ledger for confirmation.

4. The payment risk assessment method based on digital RMB according to claim 3, characterized in that: The hierarchical anonymity risk mapping table stores a combination mapping relationship between wallet real-name level, real-time risk level and anonymity processing level; among them, transactions with low risk level and high real-name level are mapped to strong anonymity processing level, which allows the use of ring signature or group signature to hide the transaction party identifier; transactions with high risk level are mapped to controllable anonymity processing level, which forces the use of decryptable identifiers encrypted based on the regulator's public key. The zero-knowledge proof technology is used to generate non-interactive zero-knowledge proofs, which prove that the transaction amount is within the compliant range and that the wallet addresses of both parties to the transaction belong to the set of certified addresses and are not blacklisted.

5. The payment risk assessment method based on digital RMB according to claim 4, characterized in that: The multi-dimensional features include transaction amount, transaction time, transaction geographical location, device fingerprint, user's historical transaction behavior sequence, and smart contract call depth; The adaptive algorithm employs an online gradient boosting decision tree or an online deep neural network model, and continuously updates the model parameters online based on real-time feedback of the processing results.

6. The payment risk assessment method based on digital RMB according to claim 5, characterized in that: The programmable risk management mechanism includes at least one of the following actions: freezing transaction funds, delaying transaction settlement, triggering an enhanced identity verification process for the transacting parties, and sending an early warning notification to the regulatory system; the actions are executed automatically through the smart contract.

7. The payment risk assessment method based on digital RMB according to claim 6, characterized in that: In the case of an offline payment transaction, after the wallet is connected to the internet, the following is also included: The local pre-assessment records, transaction hash chains, and real-time risk scores of offline transactions are uploaded to the central risk control platform to update the full lifecycle dynamic risk profile and to serve as training data to optimize the lightweight risk assessment model.

8. A payment risk assessment system based on digital RMB, comprising the payment risk assessment method based on digital RMB according to claim 7, characterized in that: include: The smart contract risk control module is used to call the risk control rules embedded in the associated smart contract to perform automated risk checks before the transaction is triggered. The offline pre-evaluation module, deployed in the wallet's security chip, is used to perform local pre-evaluation of payment transactions using a built-in lightweight risk assessment model when both are offline. During the offline period, it generates a transaction hash chain by linking the hash values ​​of previous transactions using a hash function. After the wallet is connected to the network, it performs a post-event verification of double-spending risk for offline transactions by detecting conflicts between the transaction hash chain and the confirmed transaction hash tree in the main ledger. The tiered anonymity mapping module is used to query a preset tiered anonymity risk mapping relationship table based on the real-name level and real-time risk level of the transacting parties, dynamically determine the anonymity processing level of the transaction, and call the zero-knowledge proof module to generate transaction credentials that can be disclosed selectively. The dynamic risk assessment engine is used to integrate the multi-dimensional features of the payment transaction, combine them with the full lifecycle dynamic risk profile constructed based on the unique identifier of the digital RMB and recording the feature vector of the on-chain circulation behavior sequence, use an adaptive algorithm to calculate the real-time risk score, and trigger the update of the risk profile after each transaction is completed. The programmable response module is used to trigger and execute response actions corresponding to the current risk level based on the real-time risk score.

9. A payment risk assessment system based on digital RMB according to claim 8, characterized in that: The offline pre-evaluation module is further used to: extract offline transaction feature data and output a risk pre-score using a binary classification model based on gradient boosting decision trees; reject transactions when the risk pre-score exceeds a threshold; during the offline period, calculate the transaction hash of the i-th transaction as H_i = Hash(T_i || H_{i-1}) to construct a transaction hash chain; after connecting to the network, compare each transaction hash in the transaction hash chain with the set of unspent transaction hashes confirmed in the main ledger; if there is a conflict, it is determined to be a double-spending transaction and subsequent transactions are rejected or rolled back; if there is no conflict, batch confirmation is submitted.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images