Tax audit management system based on block chain

Abstract

The invention relates to the technical field of tax data processing, in particular to a blockchain-based tax audit management system, which comprises a semantic business activity acquisition module for capturing business data from an operating system of a factoring business participant, converting the business data into a semantic unit and storing the semantic unit in a blockchain; the dynamic tax rule evolution module optimizes a tax payment obligation identification model through a federated learning framework; the tax payment event association module is integrated with an oracle machine network to construct a tax payment event atlas; the cash flow tracing and sharing module adopts an ant colony optimization algorithm to generate a cash flow incidence matrix; the auditing deduction and feedback module generates a calibration signal through simulation deduction and feeds the calibration signal back to the rule evolution module. All the modules form a closed-loop system through a data flow and a feedback mechanism, dynamic association and automatic matching of multi-subject sub-link tax payment obligations and cash flow paths are achieved, and the tax auditing efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of tax data processing technology, and in particular to a blockchain-based tax audit management system. Background Technology

[0002] Blockchain technology, based on distributed ledgers and consensus mechanisms, can provide a transparent and tamper-proof way of recording data for tax audit management. In tax scenarios, all transaction information is linked by hash values ​​and permanently stored, allowing auditors to trace the complete historical flow. The automatic execution function of smart contracts helps to standardize tax calculation and reporting processes, thereby improving audit efficiency and data reliability.

[0003] Existing blockchain-based tax audit management technologies suffer from the following technical pain points: Firstly, while blockchain technology achieves global consistency and immutability of digital asset transfer records through distributed ledgers, in factoring service tax scenarios, the debt transfer process involves multiple parties, including the original creditor, factoring company, and investors. Furthermore, tax obligations arise in stages based on transaction phases such as accounts receivable transfer, interest recognition, and profit distribution, exhibiting dynamism and complexity. Secondly, because blockchain's recording mechanism focuses on linear tracking of asset transfers, it lacks the ability to dynamically correlate and intelligently decompose tax events and cash flow paths among multiple parties, resulting in the system's inability to automatically and accurately match specific cash flows with corresponding tax responsibilities. For example, in asset securitization transactions, after multiple transfers and packaging of the underlying debt, the proceeds ultimately flow to multiple investors. Although blockchain fully records the debt transfer history, it struggles to automatically identify the value-added tax or income tax obligations generated at each transfer stage and associate them with specific cash flows. This necessitates manual tax allocation and declaration, increasing compliance risks and operational costs. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a blockchain-based tax audit management system. This system solves the technical problem that, while blockchain provides global consistency in the transfer records of digital assets on the chain, it also contradicts the characteristics of factoring services, which involve multiple entities and stages of tax obligations that are difficult to accurately separate and dynamically link. This contradiction leads to the inability to automatically match tax responsibilities with cash flow paths in complex financial transactions such as asset securitization.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: This invention provides a blockchain-based tax audit management system, including a semantic business activity collection module, a dynamic tax rule evolution module, a tax event association module, a cash flow tracing and allocation module, and an audit simulation and feedback module; The semantic business activity acquisition module is used to capture business data from the operating system of the factoring business participants, extract entity relationships from the original vouchers to transform them into semantic business activity units, and submit them to the blockchain network after identity authentication to form an on-chain activity trajectory. The dynamic tax rule evolution module is a smart contract cluster deployed on blockchain nodes. It loads basic tax law rules and integrates a federated learning framework. It continuously optimizes the tax liability identification model by aggregating the encrypted model parameter updates of authorized nodes. The tax event association module connects to a decentralized oracle network to obtain off-chain events, performs streaming processing on-chain activity trajectories and off-chain events, and generates a tax event graph representing the relationship between tax obligations and business logic through pattern recognition and causal analysis. The cash flow tracing and allocation module receives the tax event graph, uses a smart contract group with simulated ant colony optimization algorithm, and iteratively calculates the contribution probability of the cash flow path to the tax event node to output the correlation matrix between tax liability and cash flow. The audit simulation and feedback module provides a sandbox environment with configurable parameters, performs discrete event simulation based on the correlation matrix, compares the simulation results with on-chain historical data to generate a calibration signal, and feeds the calibration signal back to the dynamic tax rule evolution module. The output of the semantic business activity acquisition module is simultaneously supplied to the dynamic tax rule evolution module and the tax event association module; the optimized model output by the dynamic tax rule evolution module is used for pattern recognition by the tax event association module; the graph output by the tax event association module drives the traceability analysis of the cash flow tracing and allocation module; the association matrix output by the cash flow tracing and allocation module supports the simulation and deduction of the audit deduction and feedback module; the calibration signal output by the audit deduction and feedback module drives the update of the dynamic tax rule evolution model, forming a closed-loop feedback system to realize the dynamic association and accurate matching of tax obligations and cash flow paths of multiple entities in factoring services.

[0006] Furthermore, in the blockchain-based tax audit management system of the present invention, the semantic business activity collection module includes: A lightweight agent program deployed in enterprise resource planning systems and bank settlement systems to monitor and capture structured transaction records and images of unstructured original vouchers; The optical character recognition unit analyzes the voucher image to extract text information; The natural language processing unit locates tax elements in the text using a named entity recognition model; Semantic mapping unit, which converts tax elements into standardized semantic units based on predefined business ontology; The digital signature unit uses the elliptic curve digital signature algorithm to sign the semantic unit; The blockchain communication unit broadcasts the signed semantic unit to a blockchain network that employs a practical Byzantine fault-tolerant consensus mechanism for storage.

[0007] Furthermore, in the blockchain-based tax audit management system of the present invention, the dynamic tax rule evolution module includes: The rule storage unit stores the basic tax law rules in an extensible markup language structure; The local training unit uses local de-identified historical data to train the tax identification model at each authorized node; The parameter encryption unit applies differential privacy technology and uses homomorphic encryption algorithm to encrypt the model parameters; The secure aggregation unit is used by on-chain smart contracts to aggregate encrypted parameters and generate global model parameters through a secure multi-party computation protocol. The model update unit distributes global parameters to each node to update the local model.

[0008] Furthermore, in the blockchain-based tax audit management system of the present invention, the tax event association module includes: Oracle gateway connects to a decentralized oracle network to obtain off-chain event data that has been verified by threshold signatures; The complex event processing unit defines event patterns and association rules, and performs streaming processing on-chain activity trajectories and off-chain event data. The time series alignment unit uses a dynamic time warping algorithm to normalize event sequences with different timestamps; The causal graph construction unit, when identifying complex business patterns, activates smart contracts to analyze the causal relationships between events and constructs a tax event graph in the form of a directed acyclic graph with edge weights.

[0009] Furthermore, in the blockchain-based tax audit management system of the present invention, the cash flow tracing and allocation module includes: The path initialization unit abstracts the cash flow paths in the tax event graph into an ant colony search space and initializes the pheromones. Parallel exploration unit simulates multiple computation threads selecting paths based on pheromone concentration and heuristic rules; The pheromone update unit dynamically updates the pheromone distribution based on path matching quality. The probability calculation unit uses a Bayesian network model based on a stable pheromone distribution to calculate the probability of cash flow contributing to taxation events and generates a correlation matrix.

[0010] Furthermore, in the blockchain-based tax audit management system of the present invention, the audit simulation and feedback module includes: The sandbox interface unit provides a graphical interface for auditors to set simulation parameters; The discrete event simulation unit uses the surrogate basis modeling method to simulate the interaction process of business flow, capital flow and tax obligation flow; The difference analysis unit calculates the degree of difference between the simulation results and the actual historical data on the chain. The signal generation unit generates a calibration signal that includes rule revision suggestions when the difference exceeds a threshold. The secure transmission unit encrypts the calibration signal and submits it to the blockchain network to trigger a model update in the dynamic tax rule evolution module.

[0011] Furthermore, the blockchain-based tax audit management system of the present invention further includes a data quality verification unit in the semantic business activity collection module, which is used to perform logical consistency verification on the extracted tax elements, including verifying the matching degree between the debt amount and the contract terms, the legality of the participants' identities, and the continuity of the time series; when data contradictions are detected, a re-collection process is triggered, and the abnormal record is marked as pending manual review; the verified semantic unit must also pass the business rule compliance check before digital signature to ensure that it meets the basic requirements of current tax regulations for factoring business accounting.

[0012] Furthermore, the blockchain-based tax audit management system of the present invention further includes a model version management unit in the dynamic tax rule evolution module, which is used to maintain multiple model versions generated during the federated learning process and record the performance indicators and applicable scenarios of each version; when the accuracy of the aggregated new global model on the test set is lower than the historical best version, it automatically rolls back to the stable version and triggers an alarm; at the same time, the model version management unit monitors the contribution of parameter updates of each node, and temporarily isolates nodes with long-term low-quality updates until their local data quality is improved and verified.

[0013] Furthermore, the blockchain-based tax audit management system of the present invention further includes a graph dynamic update unit in the tax event association module, which is used to continuously monitor newly generated on-chain activities and oracle events. When a new event with a potential causal relationship with the existing tax event graph is detected, the graph reconstruction process is automatically triggered. The graph dynamic update unit adopts an incremental learning algorithm, recalculates only the changed parts, and identifies business model variants through subgraph isomorphism detection technology, so that the graph can reflect the dynamic changes of complex financial transaction structures in real time.

[0014] Furthermore, the blockchain-based tax audit management system of the present invention further includes an uncertainty quantification unit in the cash flow tracing and allocation module, which is used to evaluate the credibility of the probability allocation results output by the ant colony optimization algorithm. The uncertainty quantification unit generates multiple possible cash flow path distributions through Monte Carlo simulation and calculates the confidence interval of the allocation amount for each tax event node. When the variance of the probability distribution of the critical path exceeds the threshold, the pheromone update strategy is automatically adjusted, and a time decay factor is introduced to dynamically discount the weights of historical paths, thereby improving the responsiveness to recent transaction patterns.

[0015] Beneficial effects of this invention; This invention transforms multi-source heterogeneous data from factoring transactions into standardized on-chain activity trajectories through a semantic business activity acquisition module, providing a high-quality and reliable data foundation for the system. A dynamic tax rule evolution module utilizes a federated learning framework to achieve continuous adaptive optimization of tax rules, enhancing adaptability to complex business scenarios. A tax event correlation module constructs a tax event graph through on-chain and off-chain event fusion and causal analysis, accurately representing the correlation between tax obligations of multiple entities at different stages. A cash flow tracing and allocation module employs an ant colony optimization algorithm to achieve intelligent tracing and probability allocation of cash flow paths, solving the problem of matching funds in multi-layered financial transactions. An audit simulation and feedback module generates calibration signals through simulation and comparison with historical data, forming a closed-loop learning mechanism from practice to rule optimization. These modules are deeply coupled through data flow and feedback mechanisms; semantic data supports rule evolution, optimized models improve event recognition accuracy, event graphs guide cash flow analysis, correlation matrices drive audit simulation, and calibration signals optimize rules in reverse. Ultimately, based on globally consistent records on the blockchain, automatic and accurate matching of tax obligations and cash flow paths is achieved, effectively reducing manual intervention costs and improving the efficiency and accuracy of tax audits. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the accompanying drawings without creative effort.

[0017] Figure 1 This is a system architecture diagram of the blockchain-based tax audit management system of the present invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0019] Please see Figure 1 This invention provides a blockchain-based tax audit management system, including a semantic business activity collection module, a dynamic tax rule evolution module, a tax event association module, a cash flow tracing and allocation module, and an audit simulation and feedback module; The semantic business activity acquisition module is used to capture business data from the operating system of the factoring business participants, extract entity relationships from the original vouchers to transform them into semantic business activity units, and submit them to the blockchain network after identity authentication to form an on-chain activity trajectory. The dynamic tax rule evolution module is a smart contract cluster deployed on blockchain nodes. It loads basic tax law rules and integrates a federated learning framework. It continuously optimizes the tax liability identification model by aggregating the encrypted model parameter updates of authorized nodes. The tax event association module connects to a decentralized oracle network to obtain off-chain events, performs streaming processing on-chain activity trajectories and off-chain events, and generates a tax event graph representing the relationship between tax obligations and business logic through pattern recognition and causal analysis. The cash flow tracing and allocation module receives the tax event graph, uses a smart contract group with simulated ant colony optimization algorithm, and iteratively calculates the contribution probability of the cash flow path to the tax event node to output the correlation matrix between tax liability and cash flow. The audit simulation and feedback module provides a sandbox environment with configurable parameters, performs discrete event simulation based on the correlation matrix, compares the simulation results with on-chain historical data to generate a calibration signal, and feeds the calibration signal back to the dynamic tax rule evolution module. The output of the semantic business activity acquisition module is simultaneously supplied to the dynamic tax rule evolution module and the tax event association module; the optimized model output by the dynamic tax rule evolution module is used for pattern recognition by the tax event association module; the graph output by the tax event association module drives the traceability analysis of the cash flow tracing and allocation module; the association matrix output by the cash flow tracing and allocation module supports the simulation and deduction of the audit deduction and feedback module; the calibration signal output by the audit deduction and feedback module drives the update of the dynamic tax rule evolution model, forming a closed-loop feedback system to realize the dynamic association and accurate matching of tax obligations and cash flow paths of multiple entities in factoring services.

[0020] The semantic business activity acquisition module captures structured transaction records and unstructured original document images in factoring transactions in real time through lightweight agent programs deployed in the enterprise resource planning system and bank settlement system. The lightweight agent program listens to the message queue of the business system and captures newly generated data streams. An optical character recognition engine parses the document images to generate text information, and a natural language processing engine uses a named entity recognition model to locate key tax elements in the text, such as creditor, debtor, debt amount, and payment period. The semantic mapping unit converts the extracted elements into a standardized resource description framework triple format based on a predefined business ontology model, forming business activity units with unified semantics. The digital signature unit uses an elliptic curve digital signature algorithm to sign the semantically encoded data packets, ensuring data source credibility and integrity. The blockchain communication unit broadcasts the signed data to a blockchain network employing a practical Byzantine fault-tolerant consensus mechanism. After node verification, it is written into a new block, forming an on-chain activity trajectory with timestamps and contextual semantics. This module provides a high-quality, machine-understandable baseline data source for subsequent processing.

[0021] The Dynamic Tax Rule Evolution Module, deployed as a cluster of smart contracts on blockchain nodes, initially loads basic tax law rules stored in an extensible markup language structure. The module integrates a federated learning framework, allowing authorized nodes to perform distributed model training using locally anonymized historical audit data. The local training unit uses stochastic gradient descent to optimize the tax liability identification model, while the parameter encryption unit applies differential privacy technology and homomorphic encryption to the model weight parameters. The secure aggregation unit, as an on-chain smart contract, aggregates encrypted parameters to generate global model parameters through a secure multi-party computation protocol without decrypting any node's local data. The model update unit distributes the aggregated new parameters to each node to update its local model, enabling the rule logic to dynamically evolve with policy and market practices. This module continuously optimizes the model through privacy protection mechanisms, providing adaptive capabilities for tax event identification.

[0022] The tax event association module connects to a decentralized oracle network to obtain off-chain event data verified by threshold signatures, such as court judgments or property registration information. The complex event processing engine defines event patterns and association rules, performing streaming processing on-chain activity trajectories and off-chain event data. The time series alignment unit uses a dynamic time warping algorithm to normalize event sequences with different timestamps and collection frequencies. The causal graph construction unit, when detecting sequences conforming to complex business patterns such as multiple debt transfers, activates a smart contract to analyze the causal relationships between events and constructs a directed acyclic graph (DAG) tax event graph based on a structural causal model. The edge weights of the graph are determined by both transition probabilities and causal strength. This module deeply associates trusted off-chain events with on-chain records, identifying cross-chain and cross-cycle business logic sequences.

[0023] The cash flow tracing and allocation module receives a tax event graph as input and uses a smart contract group with a simulated ant colony optimization algorithm to handle the cash flow path tracing problem. The path initialization unit abstracts the cash flow paths in the graph into an ant colony search space and initializes the pheromone concentration. The parallel exploration unit simulates multiple computing threads representing virtual ants, selecting paths based on pheromone concentration and heuristic rules such as time proximity and amount matching. The pheromone update unit dynamically updates the pheromone distribution based on path matching quality, enhancing pheromones on high-quality paths. The probability calculation unit uses a Bayesian network model based on a stable pheromone distribution to calculate the contribution probability of each cash flow path to a specific tax event node, generating a multidimensional correlation matrix between tax liability and cash flow. This module solves the cash flow allocation problem in multi-layered financial transactions through a positive feedback optimization mechanism.

[0024] The audit simulation and feedback module provides a configurable sandbox environment. The sandbox interface allows auditors to set simulation parameters such as interest rates or transaction structures via a graphical interface. The discrete event simulation unit employs a proxy basis modeling approach, simulating the dynamic interaction of business flows, cash flows, and tax liability flows over a specific period based on an association matrix. The difference analysis unit compares the simulation results with historical on-chain data, calculating difference indices such as root mean square error. The signal generation unit generates a calibration signal, including rule revision suggestions, when the difference exceeds a threshold. The secure transmission unit encrypts the calibration signal and submits it to the blockchain network, triggering a new round of federated learning in the dynamic tax rule evolution module. This module establishes a feedback loop from audit practice to rule optimization, driving continuous self-calibration of the system.

[0025] The modules are deeply coupled through standardized interfaces and shared data models: the output of the semantic business activity acquisition module is simultaneously supplied to the dynamic tax rule evolution module and the tax event association module; the optimized model of the dynamic tax rule evolution module improves the pattern recognition accuracy of the tax event association module; the graph generated by the tax event association module guides the traceability analysis of the cash flow tracing and allocation module; the correlation matrix output by the cash flow tracing and allocation module supports the simulation and deduction of the audit deduction and feedback module; and the calibration signal of the audit deduction and feedback module drives the update of the dynamic tax rule evolution module. This nonlinear bidirectional data flow and feedback mechanism enables the system as a whole to exhibit synergistic effects, achieving adaptive and accurate matching of tax obligations and cash flow paths for multiple entities at different stages in factoring services.

[0026] This invention's semantic business activity acquisition module is deployed in enterprise resource planning (ERP) systems and bank settlement systems via a lightweight proxy program. It listens to message queues to capture structured transaction records and unstructured original voucher images. An optical character recognition (OCR) unit parses the voucher images to generate text information. A natural language processing (NLP) unit uses a named entity recognition model to locate tax elements in the text, including creditors, debtors, debt amounts, and payment terms. A semantic mapping unit converts the extracted elements into resource description framework (RDL) triples based on a predefined business ontology model. A digital signature unit signs the semantic units using an elliptic curve digital signature algorithm. A blockchain communication unit broadcasts the signed data to a blockchain network using a practical Byzantine fault-tolerant consensus mechanism, forming an on-chain activity trajectory. This module automates the process of business data acquisition and on-chain processing, providing a foundation for data credibility and semantic consistency.

[0027] This invention's dynamic tax rule evolution module stores basic tax law rules in an extensible markup language structure through a rule storage unit; a local training unit trains a taxpayer identification model on each authorized node using locally anonymized historical data; a parameter encryption unit applies differential privacy technology and encrypts the model parameters using a homomorphic encryption algorithm; a secure aggregation unit, acting as an on-chain smart contract, aggregates the encrypted parameters to generate global model parameters through a secure multi-party computation protocol; and a model update unit distributes the global parameters to each node to update the local model. This module enables distributed learning and dynamic optimization of tax rules, improving model recognition accuracy and environmental adaptability.

[0028] This invention's tax event association module connects to a decentralized oracle network via an oracle gateway to obtain off-chain event data verified by threshold signatures; the complex event processing unit defines event patterns and association rules, performing streaming processing on-chain activity trajectories and off-chain event data; the time series alignment unit uses a dynamic time warping algorithm to normalize event sequences with different timestamps; and the causal graph construction unit activates smart contracts to analyze causal relationships when complex business patterns such as multiple debt transfers are identified, constructing a directed acyclic graph (DAG) form of a tax event graph. This module achieves deep integration of on-chain and off-chain events and causal reasoning, generating complete tax obligation association relationships.

[0029] This invention's cash flow tracing and allocation module abstracts cash flow paths in the tax event graph into an ant colony search space and initializes pheromone concentration through a path initialization unit; a parallel exploration unit simulates multiple computation threads to select paths based on pheromone concentration and heuristic rules, including time proximity and monetary matching degree; a pheromone update unit dynamically updates the pheromone distribution based on path matching quality; and a probability calculation unit calculates the contribution probability of cash flow to tax events using a Bayesian network model based on a stable pheromone distribution, generating a correlation matrix between tax liability and cash flow. This module enables intelligent tracing and probability allocation of cash flow in multi-level transactions, solving the problem of matching tax obligations with funding paths.

[0030] The audit simulation and feedback module of this invention receives simulation parameters set by auditors through a sandbox interface unit, including interest rate adjustments and changes in transaction structure; the discrete event simulation unit simulates the dynamic interaction process of fund flow and tax liability flow in factoring business based on the correlation matrix and the proxy basis modeling method; the difference analysis unit calculates the root mean square error between the simulation results and on-chain historical data as a difference index; the signal generation unit generates a calibration signal including rule revision suggestions when the difference exceeds a preset threshold; the secure transmission unit encrypts the calibration signal using an asymmetric encryption algorithm and transmits it to the dynamic tax rule evolution module through the blockchain network. This module realizes the simulation verification and rule optimization feedback of the audit process, forming a self-calibration closed loop of the system.

[0031] The data quality verification unit of the semantic business activity acquisition module of this invention performs logical consistency verification on the tax elements extracted by the natural language processing unit. This includes comparing the debt amount with the contract terms, verifying the validity of the participants' identities in the business registration database, and checking the continuity of the transaction time series. When data contradictions are detected, a re-acquisition process is automatically triggered, and the abnormal record is stored in the pending review database. Semantic units that pass the verification must also undergo a business rule compliance check, matching the current tax regulations for factoring business accounting requirements before entering the digital signature process. This unit improves the quality of the source data, providing reliable input for subsequent processing.

[0032] The model version management unit of the dynamic tax rule evolution module of this invention maintains multiple model versions generated during the federated learning process, recording the accuracy and applicable scenario indicators of each version. When the accuracy of the new global model generated by the security aggregation unit on the test set is lower than the historical best version, a version rollback mechanism is automatically triggered and an alarm signal is sent to the administrator interface. Simultaneously, this unit monitors the contribution of parameter updates from each authorized node, temporarily isolating nodes whose update quality repeatedly falls below a threshold until the local data quality is verified and the nodes are rejoined for federated learning. This unit ensures the stability and reliability of model evolution. The graph dynamic update unit of the tax event association module of this invention continuously monitors newly generated business activity trajectories and off-chain events input by oracles on the blockchain network. When a new event sequence with a potential causal relationship to the existing tax event graph is detected, the graph reconstruction process is automatically triggered. This unit uses an incremental learning algorithm to recalculate only the changed parts and identifies business model variants through subgraph isomorphism detection technology, updating the edge weights and node relationships of the directed acyclic graph (DAG) form of the tax event graph. This unit achieves real-time graph updates, adapting to dynamic business environments.

[0033] The uncertainty quantification unit of the cash flow tracing and allocation module of this invention generates multiple possible cash flow path probability distributions through Monte Carlo simulation and calculates the confidence interval of the allocated amount for each tax event node. When the variance of the probability distribution of the critical path exceeds a threshold, the pheromone update strategy is automatically adjusted, and a time decay factor is introduced to dynamically discount the weights of historical paths, giving priority to recent transaction patterns. The adjusted parameters are fed back to the parallel exploration unit to optimize path selection. This unit improves the reliability and timeliness of the allocation results.

[0034] This invention captures business data from the operating systems of factoring participants through a semantic business activity acquisition module, transforming original vouchers into business activity units with unified semantics and storing them on the blockchain to form a trusted on-chain activity trajectory, providing a high-quality data foundation for subsequent processing. The dynamic tax rule evolution module utilizes a federated learning framework to aggregate the cryptographic model parameters of authorized nodes, continuously optimizing the tax obligation identification model so that the rules can adapt to policy changes and business dynamics. The tax event correlation module connects to a decentralized oracle network to obtain off-chain events, constructing a tax event graph through complex event processing and causal analysis to characterize the correlation between multi-entity, multi-stage tax obligations and business logic. The cash flow tracing and allocation module receives the tax event graph, iteratively calculates the contribution probability of cash flow paths to tax event nodes using a simulated ant colony optimization algorithm in smart contract groups, and outputs a correlation matrix between tax liability and cash flow, solving the problem of accurate allocation of cash flow in multi-level transactions. The audit simulation and feedback module performs discrete event simulation based on the correlation matrix, compares the results with historical on-chain data to generate calibration signals, and feeds them back to the dynamic tax rule evolution module, forming a closed-loop learning system. Each module works collaboratively through data flow and feedback mechanisms. The output of the semantic business activity collection module simultaneously drives the dynamic tax rule evolution module and the tax event association module, optimizes the model to improve the accuracy of event recognition, guides cash flow tracing analysis through event graphs, supports audit simulation through association matrices, and optimizes rules through calibration signals. Thus, based on the globally consistent records of the blockchain, the dynamic association and automatic matching of tax obligations and cash flow paths of multiple entities in factoring services are realized.

Claims

1. A blockchain-based tax audit management system, characterized in that, It includes a semantic business activity collection module, a dynamic tax rule evolution module, a tax event association module, a cash flow tracing and allocation module, and an audit simulation and feedback module; The semantic business activity acquisition module is used to capture business data from the operating system of the factoring business participants, extract entity relationships from the original vouchers to transform them into semantic business activity units, and submit them to the blockchain network after identity authentication to form an on-chain activity trajectory. The dynamic tax rule evolution module is a smart contract cluster deployed on blockchain nodes. It loads basic tax law rules and integrates a federated learning framework. It continuously optimizes the tax liability identification model by aggregating the encrypted model parameter updates of authorized nodes. The tax event association module connects to a decentralized oracle network to obtain off-chain events, performs streaming processing on-chain activity trajectories and off-chain events, and generates a tax event graph representing the relationship between tax obligations and business logic through pattern recognition and causal analysis. The cash flow tracing and allocation module receives the tax event graph, uses a smart contract group with simulated ant colony optimization algorithm, and iteratively calculates the contribution probability of the cash flow path to the tax event node to output the correlation matrix between tax liability and cash flow. The audit simulation and feedback module provides a sandbox environment with configurable parameters, performs discrete event simulation based on the correlation matrix, compares the simulation results with on-chain historical data to generate a calibration signal, and feeds the calibration signal back to the dynamic tax rule evolution module. The output of the semantic business activity acquisition module is simultaneously supplied to the dynamic tax rule evolution module and the tax event association module; the optimized model output by the dynamic tax rule evolution module is used for pattern recognition by the tax event association module; the graph output by the tax event association module drives the traceability analysis of the cash flow tracing and allocation module; the association matrix output by the cash flow tracing and allocation module supports the simulation and deduction of the audit deduction and feedback module; the calibration signal output by the audit deduction and feedback module drives the update of the dynamic tax rule evolution model, forming a closed-loop feedback system to realize the dynamic association and accurate matching of tax obligations and cash flow paths of multiple entities in factoring services.

2. The blockchain-based tax audit management system of claim 1, wherein, The semantic business activity collection module includes: A lightweight agent program deployed in enterprise resource planning systems and bank settlement systems to monitor and capture structured transaction records and images of unstructured original vouchers; The optical character recognition unit analyzes the voucher image to extract text information; The natural language processing unit locates tax elements in the text using a named entity recognition model; Semantic mapping unit, which converts tax elements into standardized semantic units based on predefined business ontology; The digital signature unit uses the elliptic curve digital signature algorithm to sign the semantic unit; The blockchain communication unit broadcasts the signed semantic unit to a blockchain network that employs a practical Byzantine fault-tolerant consensus mechanism for storage. 3.The blockchain-based tax audit management system of claim 1, wherein, The dynamic tax rule evolution module includes: The rule storage unit stores the basic tax law rules in an extensible markup language structure; The local training unit uses local de-identified historical data to train the tax identification model at each authorized node; The parameter encryption unit applies differential privacy technology and uses homomorphic encryption algorithm to encrypt the model parameters; The secure aggregation unit is used by on-chain smart contracts to aggregate encrypted parameters and generate global model parameters through a secure multi-party computation protocol. The model update unit distributes global parameters to each node to update the local model.

4. The blockchain-based tax audit management system of claim 1, wherein, The tax event association module includes: Oracle gateway connects to a decentralized oracle network to obtain off-chain event data that has been verified by threshold signatures; The complex event processing unit defines event patterns and association rules, and performs streaming processing on-chain activity trajectories and off-chain event data. The time series alignment unit uses a dynamic time warping algorithm to normalize event sequences with different timestamps; The causal graph construction unit, when identifying complex business patterns, activates smart contracts to analyze the causal relationships between events and constructs a tax event graph in the form of a directed acyclic graph with edge weights.

5. The blockchain-based tax audit management system of claim 1, wherein, The cash flow tracing and allocation module includes: The path initialization unit abstracts the cash flow paths in the tax event graph into an ant colony search space and initializes the pheromones. Parallel exploration unit simulates multiple computation threads selecting paths based on pheromone concentration and heuristic rules; The pheromone update unit dynamically updates the pheromone distribution based on path matching quality. The probability calculation unit uses a Bayesian network model based on a stable pheromone distribution to calculate the probability of cash flow contributing to taxation events and generates a correlation matrix.

6. The blockchain-based tax audit management system of claim 1, wherein, The audit simulation and feedback module includes: The sandbox interface unit provides a graphical interface for auditors to set simulation parameters; The discrete event simulation unit uses the surrogate basis modeling method to simulate the interaction process of business flow, capital flow and tax obligation flow; The difference analysis unit calculates the degree of difference between the simulation results and the actual historical data on the chain. The signal generation unit generates a calibration signal that includes rule revision suggestions when the difference exceeds a threshold. The secure transmission unit encrypts the calibration signal and submits it to the blockchain network to trigger a model update in the dynamic tax rule evolution module.

7. The blockchain-based tax audit management system according to claim 2, characterized in that, The semantic business activity acquisition module also includes a data quality verification unit, which is used to perform logical consistency verification on the extracted tax elements, including verifying the matching degree between the debt amount and the contract terms, the legality of the participants' identities, and the continuity of the time series; when data contradictions are detected, a re-acquisition process is triggered, and the abnormal record is marked as pending manual review; the verified semantic unit must also pass the business rule compliance check before digital signature to ensure that it meets the basic requirements of current tax regulations for factoring business accounting.

8. The blockchain-based tax audit management system according to claim 3, characterized in that, The dynamic tax rule evolution module also includes a model version management unit, which is used to maintain multiple model versions generated during the federated learning process and record the performance indicators and applicable scenarios of each version. When the accuracy of the aggregated new global model on the test set is lower than the historical best version, it automatically rolls back to the stable version and triggers an alarm. At the same time, the model version management unit monitors the contribution of parameter updates of each node and temporarily isolates nodes with long-term low-quality updates until their local data quality is improved and verified.

9. The blockchain-based tax audit management system according to claim 4, characterized in that, The tax event association module also includes a graph dynamic update unit, which is used to continuously monitor newly generated on-chain activities and oracle events. When a new event with a potential causal relationship with the existing tax event graph is detected, the graph reconstruction process is automatically triggered. The graph dynamic update unit uses an incremental learning algorithm to recalculate only the changed parts and identifies business model variants through subgraph isomorphism detection technology, enabling the graph to reflect the dynamic changes of complex financial transaction structures in real time.

10. The blockchain-based tax audit management system according to claim 5, characterized in that, The cash flow tracing and allocation module also includes an uncertainty quantification unit, which is used to evaluate the credibility of the probability allocation results output by the ant colony optimization algorithm. The uncertainty quantification unit generates multiple possible cash flow path distributions through Monte Carlo simulation and calculates the confidence interval of the allocation amount for each tax event node. When the variance of the probability distribution of the critical path exceeds the threshold, the pheromone update strategy is automatically adjusted, and a time decay factor is introduced to dynamically discount the weights of historical paths, thereby improving the responsiveness to recent transaction patterns.

Images