Finance and tax data fusion and audit traceability management system based on block chain

Abstract

The invention belongs to the technical field of block chains, and particularly relates to a financial and tax data fusion and audit traceability management system based on a block chain, which comprises an under-chain multi-source data acquisition layer, an on-chain dynamic fusion verification layer and a cross-domain collaborative feedback control layer, and is pre-bound to a data source through a business rule engine. Through combination of a lightweight fusion unit of space-time alignment and an incremental consensus mechanism based on a state channel, semantic consistency, fine-grained tracing and real-time compliance intervention of finance and tax data in a full life cycle are realized. Through a business rule pre-binding and source event capturing mechanism, finance and taxation data is prevented from breaking away from compliance context in a generation stage; a space-time alignment fusion unit is utilized to construct a composite transaction object with causal logic, and semantic segmentation among multi-source data is eliminated; the incremental consensus architecture based on the state channel is realized, and the final consistency of key transactions is ensured while the high throughput performance is maintained.

Description

Blockchain-based financial and tax data fusion and audit traceability management system Technical Field

[0001] This invention belongs to the field of blockchain technology, specifically a blockchain-based financial and tax data fusion and audit traceability management system. Background Technology

[0002] The rapid development of the digital economy has driven the transformation of financial and tax management from traditional paper-based vouchers to a fully electronic and intelligent system. Ensuring the authenticity, integrity, and traceability of massive amounts of heterogeneous financial and tax data has become a core issue supporting the modernization of tax collection and administration, preventing financial risks, and improving corporate compliance and governance capabilities. Blockchain technology, with its decentralized, immutable, and verifiable characteristics, has been widely explored and applied in financial and tax scenarios such as electronic invoices and tax declarations. It has initially established a data storage and sharing mechanism based on distributed ledgers, enhancing the credibility of audit evidence and providing a technological foundation for cross-institutional data mutual recognition. Existing blockchain-based financial and tax data management systems mostly focus on static storage of single data sources, with an architecture that essentially follows a linear process of generating data first and then uploading it to the blockchain. Enterprise financial, tax declaration, and bank payment systems operate independently, only uploading summary data such as invoice amounts and tax payment vouchers to the blockchain after the business is completed. Although this model can provide a post-audit anchor, it does not solve the structural gap in the process of financial and tax data collection, dynamic flow, and multi-source integration. It treats the blockchain as a passive trust container, resulting in a semantic gap and temporal disconnect between on-chain data and off-chain business logic.

[0003] In existing technologies, the need for fine-grained tracking of the entire lifecycle required for auditing and tracing is mutually constrained by the high storage overhead, consensus latency, and limited system throughput caused by putting all intermediate states on the chain. There is a lack of unified modeling and spatiotemporal alignment mechanisms for multi-source heterogeneous financial and tax data, and cross-domain data correlation analysis relies on manual methods. There is a lack of closed-loop feedback capabilities for business rules and data flows, making it impossible to achieve dynamic compliance verification and anomaly warning. Only passive post-event remediation is possible, which increases the cost of error correction and is prone to distorting audit conclusions due to data silos. Simply optimizing algorithms or expanding storage cannot fundamentally resolve the conflict.

[0004] To this end, the present invention provides a blockchain-based financial and tax data fusion and audit traceability management system. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows: The blockchain-based financial and tax data fusion and audit traceability management system of this invention includes a financial and tax data source access module, a multimodal data preprocessing module, a spatiotemporal alignment fusion unit, a blockchain main ledger node cluster, a smart contract execution environment, a state channel manager, a compliance verification trigger, an abnormal behavior feedback controller, and an audit evidence extraction interface; the above modules interact in a structured manner through a standardized communication protocol, and use hardware security elements as the root of trust to ensure that key operations are non-repudiable and key materials are stored in isolation.

[0007] The financial and tax data source access module is deployed in the enterprise's internal financial system, tax declaration platform, bank payment gateway, electronic contract signing platform, and logistics information system. It is used to capture original business events and their contextual metadata in real time. The module has a built-in business rule awareness unit, which can synchronously parse and bind the corresponding financial and tax compliance logic identifier at the moment the business event is triggered. The identifier is generated according to the national unified financial and tax coding system and is mapped one-to-one with elements such as tax type, tax rate, tax base, and deduction chain relationship in specific business scenarios. All captured data packets are attached with digital timestamps and device identity certificates and transmitted to edge computing nodes via power line carrier or RS485 bus.

[0008] The multimodal data preprocessing module is located on edge computing nodes. Its function is to convert raw data from different systems into an intermediate representation with a unified semantic model. This module includes a structured parsing submodule, an unstructured text extraction submodule, and a semi-structured information normalization submodule. The structured parsing submodule performs syntax validation and field integrity checks on fields such as received accounting records, invoice details, and payment vouchers. The unstructured text extraction submodule uses natural language processing technology to extract key entities and clause constraints from documents such as electronic contracts and purchase orders, and converts them into attribute-value pairs. The semi-structured information normalization submodule flattens hierarchical information such as logistics trajectories and warehousing records, retaining path indexes and associated key values. The preprocessed data is encapsulated into standardized fusion units, each carrying a unique sequence number, its associated business session ID, data type label, and integrity hash value.

[0009] The spatiotemporal alignment fusion unit is located in the off-chain collaborative processing layer. It is responsible for aggregating multiple preprocessed fusion units according to the business logic sequence and spatial topology. This unit maintains a local event log buffer and uses dual-port RAM as a shared cache. The DMA controller automatically triggers the aggregation logic after detecting a new fusion unit write interruption. The aggregation process is executed according to the preset financial and tax business process template. For example, the issuance of VAT special invoices must correspond to a valid sales contract number and outbound document number. If any precondition is missing, the fusion unit is marked as pending verification and the subsequent on-chain process is suspended. After the aggregation is completed, a fusion transaction object is generated, which contains a complete causal dependency graph, reference pointers of each sub-unit, and an overall consistency signature.

[0010] The blockchain master ledger node cluster adopts a consortium blockchain architecture and is jointly maintained by tax regulatory agencies, commercial banks, third-party auditing institutions, and large enterprises. Each master ledger node is equipped with an independent security element chip, which is used to store the node's private key and execute zero-knowledge proof verification logic. The master ledger adopts a sharding storage strategy, storing high-frequency transaction data and low-frequency archive data in different physical partitions to optimize read and write performance. Merged transaction objects are not directly written to the master ledger, but are first submitted to the state channel manager.

[0011] The State Channel Manager is a second-layer network built outside the main ledger to carry out temporary consensus for high-throughput converged transactions. This manager establishes a dedicated state channel for each pair of frequently interacting participants (such as suppliers and buyers). An optimistic concurrency control mechanism is used within the channel to allow both parties to quickly confirm the validity of the converged transaction under the premise of no conflict. When the cumulative amount in the channel reaches the preset batch threshold or a potential rule conflict is detected, the State Channel Manager compresses the current channel snapshot into a Merkle root and calls the smart contract execution environment to initiate a main chain anchoring request. This mechanism effectively reduces the load on the main chain while retaining the ability to verify the results afterward.

[0012] The smart contract execution environment runs in a trusted execution environment, loaded with a tax compliance rule base and a dynamic risk assessment model. Whenever a new snapshot of a fusion transaction is anchored to the main ledger, the environment starts a compliance verification trigger. The compliance verification trigger matches the corresponding verification logic script based on the business rule identifier carried in the fusion transaction, and performs context-sensitive analysis in conjunction with historical on-chain data. For example, when verifying the eligibility for input tax deduction, it not only checks the authenticity of the invoice and the amount, but also traces the trend of changes in the tax credit rating of upstream enterprises and whether there is a history of false invoicing. If a violation of the preset compliance conditions is found, the compliance verification trigger immediately activates the abnormal behavior feedback controller.

[0013] The abnormal behavior feedback controller sends blocking instructions or correction suggestions to the relevant financial and tax data source access modules through the message queue service. For high-risk operations (such as large-amount abnormal invoicing), the controller can remotely lock the corresponding business interface until manual review is completed. For low-risk deviations (such as inconsistent date formats), a standardized correction template is pushed for the system to automatically fill in. All feedback actions generate operation logs and write them back to the blockchain to form a closed-loop governance evidence chain.

[0014] The audit evidence extraction interface provides standardized query services, supporting the retrieval of fused transactions and their complete traceability paths by taxpayer identification number, business session ID, time interval, or multi-dimensional combination conditions. The interface integrates a graph database engine, which can efficiently traverse the causal dependency graph and restore the historical change trajectory of any node. In addition, the interface supports the generation of encapsulated packages that conform to the legal electronic evidence format, including the original data hash, on-chain evidence storage location, timestamp authoritative authentication, and operator identity credentials, meeting the requirements of judicial evidence collection.

[0015] Preferably, the business rule perception unit in the financial and tax data source access module maintains a synchronization update mechanism with the latest policy documents issued by the State Taxation Administration, obtains rule change notifications by subscribing to government information broadcast channels, and implements version control and gray-scale switching strategies in the local rule cache area to ensure data compatibility during the transition between old and new rules.

[0016] Preferably, the unstructured text extraction submodule in the multimodal data preprocessing module adopts a sequence labeling model based on an attention mechanism. However, the parameters of this model are fixed in read-only memory and can only be replaced as a whole through firmware upgrades to prevent semantic drift caused by runtime tampering.

[0017] Preferably, the business process templates used by the spatiotemporal alignment and fusion unit are written in a declarative language. The template definition includes triggering conditions, participating roles, a set of required data items, and timeout rollback strategies. All templates must be digitally signed and verified by the tax authorities before they can take effect.

[0018] Preferably, the state channel manager adopts a channel admission mechanism based on credit scores. Only participants with credit scores higher than a preset security threshold are allowed to establish long-term state channels, while other interactions are forced to follow the main chain consensus path, thereby balancing efficiency and security.

[0019] Preferably, the dynamic risk assessment model in the smart contract execution environment adopts an online learning architecture. Its training samples are derived from the comparative feature set of confirmed abnormal cases and normal business flows. The model weight update requires multi-party joint signature authorization and differential privacy perturbation processing is performed before and after the update to prevent model inversion attacks.

[0020] Preferably, the abnormal behavior feedback controller is deeply integrated with the approval workflow of the enterprise ERP system. When a medium-level warning is triggered, a pending task is automatically created and assigned to a designated risk control position. The task status changes are synchronized to the blockchain in real time to ensure that the handling process is auditable.

[0021] Preferably, the audit evidence extraction interface supports cross-chain query function, which can access other industry blockchains (such as customs import and export record chain, market supervision registration information chain) through the relay chain bridging mechanism to achieve multi-dimensional cross verification. However, cross-chain data reference must be accompanied by source chain consensus proof and cross-chain transmission integrity verification code.

[0022] The beneficial effects of this invention are as follows: The blockchain-based financial and tax data fusion and audit traceability management system described in this invention, through the pre-binding of business rules and the source event capture mechanism, avoids financial and tax data from deviating from the compliance context during the generation stage; it can utilize spatiotemporal aligned fusion units to construct composite transaction objects with causal logic, eliminating semantic fragmentation between multi-source data; it realizes an incremental consensus architecture based on state channels, ensuring the eventual consistency of key transactions while maintaining high throughput performance; it ensures real-time linkage between compliance verification logic and on-chain evidence, transforming risk identification from passive response to proactive interception; and through closed-loop feedback control and cross-domain audit interfaces, it supports the need for penetrating supervision across the entire process, all elements, and all subjects. Attached Figure Description

[0023] The invention will now be further described with reference to the accompanying drawings.

[0024] Figure 1 is a structural block diagram of the blockchain-based financial and tax data fusion and audit traceability management system of the present invention; Figure 2 is a schematic diagram of the incremental consensus mechanism in which the state channel manager and the blockchain main ledger node cluster work together in the present invention. Detailed Implementation

[0025] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0026] As shown in Figures 1 and 2, the blockchain-based financial and tax data fusion and audit traceability management system of this invention includes a financial and tax data source access module deployed within the enterprise's internal financial system, tax declaration platform, bank payment gateway, electronic contract signing platform, and logistics information system. This module is used to capture original business events and their contextual metadata in real time. It has a built-in business rule awareness unit that can synchronously parse and bind corresponding financial and tax compliance logic identifiers the instant a business event is triggered. These identifiers are generated based on the national unified financial and tax coding system and form a unique identifier with elements such as tax type, tax rate, tax base, and deduction chain relationship in specific business scenarios. In one mapping, all captured data packets are appended with a digital timestamp and device identity certificate, and transmitted to the edge computing node via power line carrier or RS485 bus. For example, when an enterprise completes a sales shipment operation, its ERP system triggers a sales shipment event. The financial and tax data source access module immediately extracts the commodity code, customer taxpayer identification number, shipment quantity, unit price, and contract number from the event, and automatically generates the identifier VAT-INV-2023-11-00123456 in accordance with Article 12 of the "Regulations on the Use of Special VAT Invoices". This identifier clearly points to the compliance requirement that general taxpayers must associate a valid contract and shipment order when issuing special VAT invoices.

[0027] The multimodal data preprocessing module is located on edge computing nodes. Its function is to convert raw data from different systems into an intermediate representation with a unified semantic model. This module includes a structured parsing submodule, an unstructured text extraction submodule, and a semi-structured information normalization submodule. The structured parsing submodule performs syntax validation and field integrity checks on received fields such as transaction records, invoice details, and payment vouchers. For example, it performs hard rule filtering to check if the invoice code length is 12 digits, if the amount field conforms to the numerical format, and if the taxpayer identification number passes the Luhn algorithm verification. The unstructured text extraction submodule uses natural language processing technology to extract key entities and clause constraints from documents such as electronic contracts and purchase orders, and converts them into attribute-value pairs. In a specific embodiment, the unstructured text extraction submodule loads an attention-based sequence labeling model with parameters stored in read-only memory. Its input is the full text of a PDF-formatted purchase contract, and its output is {Buyer: XX Technology Co., Ltd., Seller: YY Materials Factory, Delivery Date: 2023-12-15, Penalty Ratio: 5%}, etc. The structured triplet is processed; the semi-structured information normalization submodule flattens hierarchical information such as logistics trajectories and warehousing records, while retaining path indexes and associated key values. For example, a logistics path from warehouse A to sorting area B to loading point C to transport vehicle D is converted into four independent records. Each record carries a global session ID, step number, location coordinates, and timestamp, and the topological order is maintained through the parent_step_id field. The preprocessed data is encapsulated into standardized fusion units. Each fusion unit carries a unique sequence number (using UUIDv4 format), the business session ID to which it belongs (assigned by the financial and tax data source access module), data type tags (such as INVOICE, CONTRACT, LOGISTICS), and an integrity hash value (calculated using the SHA3-256 algorithm).

[0028] In some embodiments, the spatiotemporal alignment fusion unit is located in the off-chain collaborative processing layer and is responsible for aggregating multiple preprocessed fusion units according to business logic timing and spatial topology. The spatiotemporal alignment fusion unit maintains a local event log buffer, using dual-port RAM as a shared cache. The aggregation logic is automatically triggered by the DMA controller upon detecting a write interrupt of a new fusion unit. The aggregation logic process is executed according to a preset financial and tax business process template. The financial and tax business process template is written in a declarative language and defines triggering conditions, participating roles, required data item sets, and timeout rollback strategies. All templates must be digitally signed and verified by the tax authorities before deployment; for example, for the scenario of issuing value-added tax special invoices, the template definition is as follows: in, The string represents the event type. As a unique identifier for the contract, For the outbound order number, The taxpayer identification number is used for customers, and the timeout threshold is 300 seconds. If any precondition is missing, the fusion unit is marked as pending verification and the subsequent on-chain process is suspended. After aggregation, a fusion transaction object is generated, which contains a complete causal dependency graph (stored in the form of a directed acyclic graph DAG), reference pointers of each sub-unit (indexed by the unique sequence number of the fusion unit), and an overall consistency signature (signed by the hardware security element using the ECDSA-P256 algorithm to sign the hash value of the DAG root node).

[0029] In some embodiments, the blockchain master ledger node cluster adopts a consortium blockchain architecture, jointly maintained by tax regulatory agencies, commercial banks, third-party auditing institutions, and large enterprises, with a total of 7 nodes to meet the requirements of Byzantine fault-tolerant consensus. Redundancy requirements ( Each master ledger node is equipped with an independent Secure Element chip (such as Infineon OPTIGA™ TPM2.0) to store the node's private key and execute zero-knowledge proof verification logic. The master ledger employs a sharding storage strategy, storing high-frequency transaction data (such as daily invoice streams) and low-frequency archival data (such as annual audit reports) in different physical partitions. The high-frequency partition uses LevelDB as the underlying storage engine, supporting 10,000 writes per second; the low-frequency partition uses the IPFS distributed file system, storing only content hashes, with the original data retrieved through content addressing; merged transaction objects are not directly written to the master ledger but are first committed to the state channel manager.

[0030] The State Channel Manager is a second-layer network built outside the main ledger to carry high-throughput, converged transactional ad hoc consensus. The State Channel Manager establishes a dedicated state channel for each pair of frequently interacting participants (such as suppliers and buyers). The channel admission mechanism is based on a credit scoring model, which comprehensively considers historical transaction fulfillment rates, tax credit ratings (A / B / C / D), frequency of abnormal behavior, and other dimensions. The calculation formula is as follows:

[0031] in, This represents the contract fulfillment rate over the past 12 months (value range [0,1]). These are the tax credit rating mapping values ​​(A=1.0, B=0.8, C=0.5, D=0.0). Frequency of abnormal behavior (unit: times / thousand transactions). This is a normalized value for the enterprise's registration duration within the system (maximum value is 1.0); it only applies when... Only when a long-term state channel is established is it allowed; otherwise, the interaction is forced to follow the main chain consensus path. An optimistic concurrency control mechanism is adopted within the channel, allowing both parties to quickly confirm the validity of the fusion transaction under the premise of no conflict. When the cumulative number of transactions in the channel reaches the preset batch threshold (default is 100 transactions) or a potential rule conflict is detected (such as two transactions referencing the same invoice number but with different amounts), the state channel manager compresses the current channel snapshot into a Merkle tree root and calls the smart contract execution environment to initiate a main chain anchoring request. The Merkle tree is constructed using the Keccak-256 hash function, and the tree height is limited to 8 layers to ensure that the verification path length does not exceed 8 hash operations.

[0032] The smart contract execution environment runs within the IntelSGX trusted execution environment, loaded with a tax compliance rule base and a dynamic risk assessment model. Whenever a new merged transaction snapshot is anchored to the main ledger, this environment initiates a compliance verification trigger. The compliance verification trigger matches the corresponding verification logic script based on the business rule identifier carried in the merged transaction and performs context-sensitive analysis using historical on-chain data. For example, when verifying input tax deduction eligibility, the following logic is executed: in, This indicates the result of the invoice authenticity verification (verified via tax UKey signature). This indicates that the difference between the invoice amount and the payment voucher amount is less than 0.5%. The seller's current tax credit rating is used; if a violation of preset compliance conditions is found, the compliance verification trigger will immediately activate the abnormal behavior feedback controller.

[0033] The abnormal behavior feedback controller sends blocking instructions or correction suggestions to the relevant financial and tax data source access modules through the RabbitMQ message queue service. For high-risk operations (such as a single-day invoice amount exceeding 10 times the company's registered capital), the controller can remotely lock the corresponding business interface until manual review is completed. For low-risk deviations (such as an invoice date format of 2023 / 11 / 05 instead of 2023-11-05), a standardized correction template is pushed for the system to automatically fill in. All feedback actions generate operation logs and are written back to the blockchain, forming a closed-loop governance evidence chain. In some embodiments, the abnormal behavior feedback controller is deeply integrated with the approval workflow of the enterprise ERP system. When a medium-level warning is triggered (such as three consecutive invoices where the buyer's industry does not match the seller's main business), a pending task is automatically created and assigned to a designated risk control position. The task status changes (such as accepted, rejected, or corrected) are synchronized to the blockchain in real time to ensure that the handling process is auditable.

[0034] The audit evidence extraction interface provides a standardized RESTful API service, supporting the retrieval of fused transactions and their complete traceability paths by taxpayer identification number, business session ID, time interval, or multi-dimensional combination conditions. Internally, the interface integrates the Neo4j graph database engine, enabling efficient traversal of causal dependency graphs and reconstruction of the historical change trajectory of any node. For example, querying all sales invoices for taxpayer 91310101MA1FLMXXXX in the fourth quarter of a certain year returns a list containing 237 fused transaction objects, each object accompanied by a full-link event node and timestamp from contract signing, warehousing, invoicing to bank payment. In addition, this interface supports generating legally valid electronic evidence packages that comply with the "Rules for Electronic Data Forensics," including the original data hash, on-chain evidence storage location (block height and transaction index), timestamp authoritative authentication, and operator identity credentials. In some embodiments, the audit evidence extraction interface supports cross-chain query functionality, accessing other industry blockchains, such as the customs import and export record chain and the market supervision registration information chain, through a relay chain bridging mechanism. Cross-chain data citations must be accompanied by source chain consensus proof (such as Fabric's endorsement policy signature set) and cross-chain transmission integrity check code (generated using the HMAC-SM3 algorithm) to ensure that the cited data has not been tampered with.

[0035] Furthermore, the business rule awareness unit in the financial and tax data source access module maintains a synchronization update mechanism with the latest policy documents issued by the State Taxation Administration; this unit subscribes to the government information broadcast channel to receive rule change notifications; the local rule cache implements version control and gray-scale switching strategies: new rules are loaded in shadow mode and only take effect on 1% of test traffic; if there are no abnormal alarms within 72 hours, the effect is gradually increased to 100%; this mechanism ensures data compatibility during the transition between old and new rules and avoids system interruption due to sudden policy changes.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A blockchain-based financial and tax data fusion and audit traceability management system, characterized in that, include: The tax and finance data source access module is used to capture raw business events and their contextual metadata in real time, and bind tax and finance compliance logic identifiers generated according to the tax and finance coding system at the moment the event is triggered. The identifiers are mapped to tax types, tax rates, tax bases, and deduction chain relationships. The multimodal data preprocessing module is set up at the edge computing node to convert raw data from different systems into an intermediate representation with a unified semantic model, and encapsulate it into a standardized fusion unit carrying a unique sequence number, business session ID, data type label, and integrity hash value. The spatiotemporal alignment fusion unit maintains a local event log buffer and aggregates multiple fusion units according to the business logic time sequence and spatial topology relationship based on a preset tax and finance business process template to generate a package. The system includes a fusion transaction object containing a causal dependency graph, sub-unit reference pointers, and an overall consistency signature; a state channel manager that establishes dedicated state channels for frequently interacting participants, employing an optimistic concurrency control mechanism within the channel to achieve temporary consensus on fusion transactions, and compressing a Merkle root into a channel snapshot and submitting it to the blockchain main ledger node cluster for anchoring when a batch threshold is reached or a rule conflict is detected; a blockchain main ledger node cluster using a consortium blockchain architecture, jointly maintained by tax regulatory agencies, commercial banks, third-party auditing institutions, and large enterprises, with the main ledger employing a sharding storage strategy to distinguish between high-frequency transaction data and low-frequency archive data; and a smart contract execution environment running in a trusted execution environment, loading a tax compliance rule base and a dynamic risk assessment model. The compliance verification trigger is activated after the fusion transaction snapshot is anchored. It matches the verification logic script based on the business rule identifier and performs context-sensitive analysis in combination with historical on-chain data. The abnormal behavior feedback controller is activated if the compliance verification trigger determines that there is a violation of the preset compliance conditions. It sends a blocking instruction or correction suggestion to the relevant financial and tax data source access module and writes the feedback action log back to the blockchain. The audit evidence extraction interface provides standardized query services, supports the retrieval of fused transactions and their complete traceability paths by taxpayer identification number, business session ID or time interval, and integrates a graph database engine to traverse the causal dependency graph and restore the historical change trajectory of any node.

2. The blockchain-based financial and tax data fusion and audit traceability management system according to claim 1, characterized in that, The business rule awareness unit in the financial and tax data source access module obtains policy change notifications by subscribing to the government information broadcast channel, and implements version control and gray-scale switching strategies in the local rule cache area. The new rules take effect on some test traffic in shadow mode, and after a preset observation period without any abnormalities, they are gradually switched to full application.

3. The blockchain-based financial and tax data fusion and audit traceability management system according to claim 1, characterized in that, The unstructured text extraction submodule in the multimodal data preprocessing module uses an attention-based sequence labeling model to extract key entities and clause constraints from electronic contracts or purchase orders. The model parameters are stored in read-only memory and can only be replaced as a whole through firmware upgrades.

4. The blockchain-based financial and tax data fusion and audit traceability management system according to claim 1, characterized in that, The financial and tax business process templates on which the spatiotemporal alignment and fusion unit is based are written in a declarative language and include triggering conditions, participating roles, a set of required data items, and timeout rollback strategies. All templates must be digitally signed and verified before they can take effect.

5. The blockchain-based financial and tax data fusion and audit traceability management system according to claim 1, characterized in that, The state channel manager adopts a channel access mechanism based on credit scores. The credit score comprehensively considers historical contract performance rate, tax credit rating, frequency of abnormal behavior and system registration time. Only when the credit score is not lower than the preset security threshold is the participant allowed to establish a long-term state channel; otherwise, the main chain consensus path is forced.

6. The blockchain-based financial and tax data fusion and audit traceability management system according to claim 1, characterized in that, The dynamic risk assessment model in the smart contract execution environment adopts an online learning architecture. Its training samples are derived from the comparative feature set of confirmed abnormal cases and normal business flows. The model weight update requires multi-party joint signature authorization, and differential privacy perturbation processing is performed before and after the update to prevent model inversion attacks.

7. The blockchain-based financial and tax data fusion and audit traceability management system according to claim 1, characterized in that, The abnormal behavior feedback controller is deeply integrated with the approval workflow of the enterprise ERP system. When a medium-level warning is triggered, a pending task is automatically created and assigned to a designated risk control position. The task status changes are synchronized to the blockchain in real time to ensure that the handling process is auditable.

8. The blockchain-based financial and tax data fusion and audit traceability management system according to claim 1, characterized in that, The audit evidence extraction interface supports cross-chain query functionality, accessing the customs import and export record chain or market supervision registration information chain through the relay chain bridging mechanism. The cross-chain referenced data includes source chain consensus proof and cross-chain transmission integrity verification code generated based on the HMAC-SM3 algorithm.

9. A blockchain-based method for financial and tax data fusion and audit traceability management, applicable to the blockchain-based financial and tax data fusion and audit traceability management system described in any one of claims 1-8, characterized in that, Includes the following steps: The tax and financial data source access module captures raw data and binds it to tax and financial compliance logic identifiers when a business event is triggered. At edge computing nodes, structured parsing, unstructured text extraction, and semi-structured information normalization are performed on multi-source heterogeneous data to generate standardized fusion units. Based on a preset financial and tax business process template, causal aggregation is performed on the fusion units in the spatiotemporal alignment fusion unit to generate fusion transaction objects. Temporary consensus is reached on the fusion transactions within the state channel. When the anchoring conditions are met, the channel snapshot is submitted to the blockchain main ledger for anchoring. Load a compliance rule base and risk assessment model into the smart contract execution environment to perform compliance verification that is sensitive to the anchored integrated transaction execution context; If the verification fails, an intervention instruction is sent to the data source through the abnormal behavior feedback controller, and the operation log is uploaded to the blockchain; the external query request is responded to through the audit evidence extraction interface, returning the fused transaction and its complete traceability path, and supporting the generation of a package that conforms to the legal electronic evidence format.

10. A blockchain-based method for financial and tax data fusion and audit traceability management according to claim 9, characterized in that, During the compliance verification process, the eligibility for input tax deduction is verified; during cross-chain queries, the validity of the source chain consensus and the integrity of cross-chain transmission are verified for the referenced external chain data.

Images