A cross-border accounting compliance processing system based on federated learning and trusted computing

Abstract

The application belongs to the technical field of financial data sharing, and particularly relates to a cross-border accounting compliance processing system based on federated learning and trusted computing, comprising: a federated learning module, which is used for synchronously updating learning models of all financial nodes through a federated averaging algorithm according to model parameters trained locally by each financial node; a dynamic account book mapping module, which is used for acquiring exchange rate data of countries where each financial node is located in real time, and automatically adjusting currency amounts in the account book through sliding window backtracking and recalculation; a trusted execution module, which is used for performing cross-chain contract execution in an isolated container; a tax atlas reasoning module, which is used for calculating correlation strength between each node in a locally configured knowledge graph through an attention mechanism, so as to generate a probability distribution of a compliance path and identify an optimal tax processing scheme; and a report generation module, which is used for generating a standard format report according to relevant data about accounting standard alignment and relevant data about exchange rate adjustment.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of financial data sharing, and particularly relates to a cross-border accounting compliance processing system based on federated learning and trusted computing. BACKGROUND

[0002] In the global accounting practice of multinational enterprises, there are great differences in accounting standards, currency systems and tax regulations adopted by different countries. Traditional ERP (Enterprise Resource Planning) systems rely on centralized data models and manual configuration, and cannot realize the real-time of standard conversion, the automation of data privacy protection and tax prediction. For example, a U.S. enterprise A has its subsidiaries in China and Germany, which adopt Chinese accounting standards and IFR (International Financial Reporting Standards) respectively for accounting, and need to manually convert the standards every quarter and submit consolidated reports, which takes an average of 4-6 weeks. SUMMARY

[0003] In view of the above-mentioned deficiencies of the prior art, the purpose of the present application is to provide a cross-border accounting compliance processing system based on federated learning and trusted computing, to solve the report conversion problem caused by the differences in accounting standards of multiple countries, and no longer need manual intervention. At the same time, the privacy and security of financial data in cross-border processing are guaranteed, the data compliance requirements of each country are met, and the automatic operations such as tax risk prediction and intelligent report generation are realized, and the processing efficiency is improved.

[0004] The application provides a cross-border accounting compliance processing system based on federated learning and trusted computing, which is applied to a multinational enterprise, and the multinational enterprise includes financial nodes in different countries; the system comprises: a federated learning module, which is used for synchronously updating learning models of all financial nodes through a federated averaging algorithm according to model parameters trained locally by each financial node; wherein, a corresponding learning model is deployed in each financial node in advance, and the learning model is configured to adapt to the accounting standards of the current country, is used for processing semantic understanding of local financial data, and learns mapping rules of the accounting standards; a dynamic account book mapping module, which is used for acquiring exchange rate data of the country where each financial node is located in real time, and updating the locally configured exchange rate conversion rule according to the exchange rate data, so that the currency amount in the account book is automatically adjusted through sliding window backtracking and recalculation after the exchange rate conversion rule is updated; a trusted execution module, which is used for performing cross-chain contract execution in an isolated container, including merging of cross-chain contracts and account offset of cross-chain contracts; a tax atlas reasoning module, which is used for calculating the correlation strength between each node in the locally configured knowledge graph through an attention mechanism according to a feature vector of a to-be-processed transaction, so as to generate a probability distribution of a compliance path and identify an optimal tax processing scheme; and a report generation module, which is used for filling related data about accounting standard alignment output by the federated learning module and related data about exchange rate adjustment output by the dynamic account book mapping module into a report template through a preset field mapping rule, so as to generate a report in a standard format.

[0005] In an embodiment of the application, the learning model adopts a BERT-Adapter model.

[0006] In an embodiment of the application, the federated learning module also uses an epsilon-differential privacy mechanism to protect sensitive financial data hidden in the model parameters in the process of training and aggregating the model parameters of each financial node.

[0007] In an embodiment of the application, when the learning model of any financial node is invalid, the federated learning module is also used for dynamically adjusting the weight proportion of the learning model corresponding to the financial node through weighted averaging.

[0008] In an embodiment of the application, the dynamic account book mapping module is also used for identifying outliers in the exchange rate data through an IQR method, and automatically eliminating or correcting abnormal data.

[0009] In an embodiment of the application, the dynamic account book mapping module is also configured with account book field corresponding rules of the accounting standards of each financial node, which is used for automatically identifying the field meaning of the local account book of any financial node, and performing field alignment and conversion.

[0010] In an embodiment of the application, the trusted execution module is configured to run in an SGX container.

[0011] In an embodiment of the present application, the trusted execution module is further configured to generate a contract execution state snapshot at a preset frequency to record the current execution progress, offset result and key parameters of the cross-chain contract; wherein the trusted execution module is further configured to compress the generated contract execution state snapshot by using a Merkle Root, and synchronize the hash value obtained by compression and the snapshot information to the main chain of the blockchain.

[0012] In an embodiment of the present application, the trusted execution module is further configured to perform a contract rollback operation after an error occurs in contract execution, and generate a unique rollback event ID and corresponding HASH record on the main chain of the blockchain; wherein the record of the contract rollback operation is stored in association with the record of the contract execution.

[0013] In an embodiment of the present application, the tax map reasoning module has a knowledge graph pre-constructed with nodes of country / region, tax type, tax rate, industry type, transaction type and compliance clause, and the nodes are connected by applicable relationship, conflict relationship or preferential relationship.

[0014] In an embodiment of the present application, the tax map reasoning module adopts a graph attention network as a core reasoning framework and is constructed as a 3-layer network structure.

[0015] In an embodiment of the present application, the report generation module is trained by using a natural language generation model.

[0016] In an embodiment of the present application, the report generation module further integrates an anomaly detection model for multi-dimensional detection of the generated report, including numerical anomaly detection, logical conflict detection and compliance deviation detection, and highlighting of risk points in the report.

[0017] In an embodiment of the present application, the anomaly detection model uses an isolation forest algorithm to identify risk points in the report.

[0018] The present application focuses on four core pain points of multi-standard differences, data compliance, tax conflicts and report inefficiency in cross-border accounting processing, and integrates technologies such as federated learning, trusted computing, blockchain and graph neural network to construct a full-process intelligent processing system, providing safe and compliant and efficient intelligent cross-border accounting solutions for multinational enterprises, financial shared centers and the like. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings, which are incorporated herein and form a part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application. It is to be understood that the drawings are designed solely for purposes of illustration to be used in conjunction with the description in

[0020] Figure 1 A structure schematic diagram of a cross-border accounting compliance processing system based on federated learning and trusted computing provided in an embodiment of the present application;

[0021] Figure 2 A process schematic diagram of processing logic of a cross-border accounting compliance processing system based on federated learning and trusted computing provided in an embodiment of the present application;

[0022] Figure 3 A process schematic diagram of processing logic of a trusted execution module provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] For the purpose of promoting the understanding and facilitating appreciation of the application, reference will be made to the accompanying drawings relating to embodiments of the application. Embodiments of the application are illustrated in the drawings and described in detail below. It should be noted that the application can be implemented in numerous ways, including the methods specified in the application and other methods. Instead of being restricted to the specific embodiments described herein, it is therefore intended that the application include all alternatives, modifications and equivalents falling within the spirit and scope of the present application.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0025] The specific embodiments of the present application will be described below with reference to the accompanying drawings. Based on the disclosure given herein, those skilled in the art can easily understand other advantages and purposes of the present application. The present application can be implemented or applied in other different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0026] In the following description, numerous specific details are discussed in order to provide a thorough understanding of the embodiments of the application. However, those skilled in the art will recognize that the embodiments of the application can be practiced without these specific details, in other embodiments, well-known structures and devices are shown in block diagram form rather than in detail in order to avoid obscuring the embodiments of the application.

[0027] Embodiment 1

[0028] Please refer to Figure 1 、 2 As shown in the figure, a cross-border accounting compliance processing system based on federated learning and trusted computing can be applied to a multinational enterprise, and the multinational enterprise can have subsidiaries in different countries, i.e., financial nodes (subsidiaries) in several different countries, thereby assisting the multinational enterprise to realize intelligent financial processing based on privacy protection, real-time multi-criteria alignment. The cross-border accounting compliance processing system includes five core modules: a federated learning module 10, a dynamic account book mapping module 20, a trusted execution module 30, a tax atlas reasoning module 40, and a report generation module 50, and covers the whole process of cross-border accounting of the multinational enterprise through collaborative work.

[0029] Among them, the federated learning module 10 can complete the alignment training of multi-country accounting standards without transmitting original financial data. Specifically, first of all, it needs to be explained that the multinational enterprise will pre-deploy a BERT-Adapter model, i.e., a learning model, at each financial node, which is responsible for adapting the accounting standards of the country (such as Chinese accounting standards, IFRS), processing the semantic understanding of local financial data, and learning the mapping rules of accounting standards.

[0030] It can be understood that the BERT-Adapter model as a lightweight adaptation module does not need to retrain the basic BERT model, but only needs to fine-tune the adaptation layer parameters, so as to quickly respond to the differences of different countries' standards. Moreover, BERT-Adapter can realize efficient fine-tuning of BERT model through very few additional parameters, while ensuring performance close to full fine-tuning, solving resource consumption, multi-task conflict and catastrophic forgetting problem, and accordingly serving as a preferred solution for the learning model in this embodiment.

[0031] Correspondingly, after each financial node completes a round of training on the BERT-Adapter deployed therein, it uploads the model parameters (rather than the original data) to the federated learning module 10. The federated learning module 10 performs weighted averaging on the model parameters fed back by all financial nodes to generate globally unified standard alignment model parameters, and then distributes them to each financial node to realize synchronous updating of multi-financial node models, i.e., through FedAvg (Federated Averaging) to ensure that each financial node follows consistent standard conversion logic.

[0032] Meanwhile, during the training and aggregation of model parameters, an ε-differential privacy mechanism (where the value of ε ranges from 0.1 to 1.0) is introduced to protect the sensitive financial data hidden in the model parameters. Specifically, by adding small, controllable noise to the model parameters, the sensitive financial data characteristics of individual financial nodes are masked, preventing the model parameters from being used to derive the original financial data in reverse, thereby meeting the data compliance requirements of various countries.

[0033] Furthermore, when the BERT-Adapter model of a certain country's financial node fails (e.g. due to network failure or hardware problems), the federated learning module 10 will automatically identify the status of the financial node and dynamically adjust the weight calculation rules of the global model. That is, it will adjust the weight ratio of the learning model corresponding to the financial node by weighted averaging, reduce the impact of the failed financial node on the global model, and trigger the remaining normal financial nodes to retrain locally and upload model parameters to ensure that the federated learning process is not interrupted and maintain the accuracy of criterion alignment.

[0034] The Dynamic Ledger Mapping Module 20 is mainly used to solve two key problems in cross-border accounting: "inconsistent fields across multiple standards" and "real-time exchange rate fluctuations". It enables automatic adaptation, linkage adjustment and accuracy correction of ledger data of subsidiaries in different countries of multinational corporations, providing a unified data foundation for subsequent consolidated financial statements and compliance processing.

[0035] Specifically, the dynamic ledger mapping module 20 has built-in rules for the correspondence of ledger fields for each financial node with the corresponding accounting standards of the country (such as Chinese accounting standards and IFRS). This allows it to automatically identify the meaning of fields in the local ledgers of the multinational corporation's subsidiaries, complete the field alignment and conversion under heterogeneous standards, and eliminate the need for manual adjustment of the standard caliber, thus solving the inefficiency problem of manual configuration required by traditional ERP systems.

[0036] Meanwhile, the dynamic ledger mapping module 20 also connects to authoritative exchange rate data sources such as the China Foreign Exchange Trading Center API and third-party market APIs, thereby obtaining real-time exchange rate data for the countries where each financial node is located, such as the exchange rate data for RMB, USD, EUR, and other major currencies, to update the locally configured exchange rate conversion rules accordingly. Furthermore, the dynamic ledger mapping module 20 supports sliding window rollback and recalculation, allowing for automatic adjustment of currency amounts in multiple ledgers when exchange rates change, i.e., after the local exchange rate conversion rules are updated. The sliding window length can be customized from 7 to 30 days to adapt to the exchange rate adjustment cycle needs of different enterprises.

[0037] It should be noted here that the dynamic ledger mapping module 20 uses the IQR (interquartile range) method to identify outliers in the exchange rate data (such as abnormal fluctuations in exchange rate values), and automatically removes or corrects abnormal data, avoiding calculation deviations caused by incorrect exchange rates, and ensuring the accuracy of multi-currency ledger data, providing a reliable currency basis for subsequent internal offset, tax calculation.

[0038] The trusted execution module 30 can be used to protect data privacy and complete contract merging, reconciliation and compliance recording of cross-border ledgers.

[0039] First, the trusted execution module 30 relies on SGX (Software Guard Extension) technology to build an independent trusted execution container (SGX enclave), i.e. running in the SGX container, so that all cross-chain accounting contract execution and data calculation are completed in the isolated container, which is physically isolated from external systems.

[0040] It can be understood that this environment can prevent contract code from being tampered with and the execution process from being monitored, ensuring the security of core logic such as internal transaction offset and tax rule verification, and meeting the requirements of cross-border accounting for data privacy and operation compliance.

[0041] Second, the execution of specific cross-chain accounting contracts includes cross-chain contract triggering and merging, for example, as shown in Figure 3 When the subsidiary chain (such as subsidiary chain A, subsidiary chain B) generates internal transaction contracts (such as internal procurement, fund transfer contracts), the trusted execution module 30 will automatically receive the contract merging request, and aggregate the multi-chain contracts into the SGX container for unified processing; and the accounting offset and conflict detection of cross-chain contracts, i.e. in the SGX container, the trusted execution module 30 will automatically complete the accounting offset of cross-chain contracts according to the preset accounting rules (such as internal offset logic). At the same time, the trusted execution module 30 will also detect conflicts in contract execution (such as exchange rate calculation differences, inconsistent criteria), and trigger the correction mechanism to ensure that the offset result meets the compliance requirements.

[0042] Further, the trusted execution module 30 will also generate contract execution state snapshots at a preset frequency, for example, once every minute, to record the current cross-chain contract execution progress, offset result, key parameters and other information, avoiding data loss due to system failure. And after each snapshot is generated, the trusted execution module 30 is also used to compress the generated contract execution state snapshot through the Merkle Root (Merkle Root) technology, and synchronize the compressed hash value (HASH) and snapshot information to the main chain of the blockchain, and the record is tamper-proof, providing traceable original data basis for subsequent audit.

[0043] In addition, if there is an error in contract execution (such as rule adaptation deviation, data anomaly), the trusted execution module 30 will automatically trigger contract rollback operation, and generate a unique rollback event ID and corresponding HASH record on the blockchain main chain, as shown in Figure 3 It should be noted that the rollback record of the contract is stored in association with the original execution record of the contract, clearly presenting the complete change trajectory of contract execution, meeting the compliance requirements of cross-border supervision on "traceable and auditable" accounting operations, and avoiding the risk of rollback without record.

[0044] The tax atlas reasoning module 40 solves the problem of cross-border tax rule conflict and compliance path matching through structured tax knowledge and intelligent reasoning algorithm.

[0045] It should be noted that the tax atlas reasoning module 40 is pre-constructed with a knowledge graph taking country / region, tax type, tax rate, industry type, transaction type, and compliance clause as nodes, and each node is connected through applicable relationship (such as a tax type applicable to a country), conflict relationship (such as the difference in tax recognition of the same transaction by two countries), or preferential relationship (such as a certain industry enjoying a certain tax rate preference), and is dynamically updated.

[0046] The tax atlas reasoning module 40 adopts graph attention network (GAT) as the core reasoning framework, realizes deep matching and conflict resolution of tax rules through a 3-layer network structure, calculates the correlation strength between each node in the locally configured knowledge graph according to the feature vector of the transaction to be processed through the attention mechanism, generates the probability distribution of the compliance path, and identifies the optimal tax processing scheme, and the network structure is as follows:

[0047] Input layer: receives the feature vector of the transaction to be processed (such as transaction country, transaction amount, industry attribute, etc.).

[0048] Hidden layer: calculate the correlation strength between nodes through attention mechanism (attention weight range can be set to 0.1~0.9), focus on capturing the core correlation of "transaction-tax-country".

[0049] Output layer: generate the probability distribution of the compliance path, and identify the optimal tax processing scheme.

[0050] It can be understood that through the attention mechanism, the key rules (such as high-priority international tax agreements are superior to domestic laws) can be automatically focused, and the secondary correlation can be weakened, thereby improving the reasoning accuracy.

[0051] In a specific embodiment, the tax graph reasoning module 40 receives the original data of cross-border transactions (such as related party transaction amount, transaction type, involved country), and converts it into entity features recognizable by the graph. Correspondingly, based on the GAT network, the applicable tax rules can be matched in the knowledge graph, the rule conflicts (such as A country recognizes as service income, B country recognizes as royalty) can be automatically identified, and the conflicts can be resolved according to the "agreement priority" and "territorial jurisdiction" meta-rules. Finally, a structured compliance path report is generated, including applicable tax types, tax rates, reporting deadlines, document requirements, etc., and the corresponding risk level is marked.

[0052] It should be noted that when new tax regulations are revised or reasoning biases occur, the knowledge graph and GAT model parameters can be updated through the federated learning module 10 to ensure that the reasoning results adapt to the real-time changes in the cross-border tax environment.

[0053] The report generation module 50 is used to convert the processed cross-border accounting data into reports that meet the regulatory requirements of each country, and accurately locate potential risk points.

[0054] To this end, a multi-lingual report template library can be pre-constructed to cover standardized report templates in countries where major financial nodes are located, including balance sheets, income statements, cash flow statements, and other basic reports, as well as country-specific reports, transfer pricing documents, etc. Special compliance reports, such as NLG (Natural Language Generation) models, can be constructed and trained with a large number of multi-lingual financial reports to support automatic translation of report fields and annotations, ensuring that terminology is consistent with local accounting standards.

[0055] Correspondingly, the report generation module 50 receives the relevant data about accounting standard alignment output by the federated learning module 10 and the relevant data about exchange rate adjustment output by the dynamic ledger mapping module 20, automatically fills in the report template through the pre-set field mapping rules (such as mapping the treatment results of "capitalization of research and development expenses" under different standards to specific rows of the report), and automatically adjusts the output style through format adaptation to generate standard format reports, which can effectively avoid manual typesetting errors.

[0056] In addition, the report generation module 50 also integrates an anomaly detection model that uses the Isolation Forest algorithm to perform multi-dimensional detection on the generated reports, including:

[0057] Numerical anomaly detection, such as a subject amount fluctuation exceeding 3 times the industry mean standard deviation, a sudden increase in related party transactions.

[0058] Logic conflict detection, such as the "total assets" and "total liabilities and owner's equity" on the balance sheet not being equal, and the cash flow table net amount and the profit table reconciliation relationship being abnormal.

[0059] Compliance deviation detection, such as failure to disclose certain related party information as required by local regulations, report submission deadline warning.

[0060] Correspondingly, the report generation module 50 highlights the identified risk points in the report through color labeling, annotation, and other means, and supports user clicks to view detailed verification logic.

[0061] In addition, after each report generation or modification, the report generation module 50 automatically records the version number, generation time, operator, and associates the underlying data source snapshot, supporting historical version backtracking and difference comparison. At the same time, the report generation module 50 is also used to synchronize the key operations (such as data filling rules, risk detection parameters) in the report generation process to the blockchain main chain of the trusted contract execution module, ensuring that the report is traceable and tamper-proof, meeting the requirements of cross-border audit.

[0062] Based on the cross-border accounting compliance processing system provided by the above embodiments, it is not limited to application in multinational enterprises, for example, it can also be applied to financial shared centers, financial technology platforms, etc. Corresponding financial nodes can also be set up in different countries, and no further limitation is made. Modifications and refinements made by those skilled in the art to the embodiments of the present application without departing from the spirit of the present application still fall within the scope of the present application.

[0063] In summary, the present application focuses on the four core pain points of multi-standard differences, data compliance, tax conflicts, and inefficient reports in cross-border accounting processing, and integrates technologies such as federated learning, trusted computing, blockchain, and graph neural networks to build a full-process intelligent processing system. It provides a safe and compliant and efficient intelligent cross-border accounting solution for multinational enterprises, financial shared centers, etc., which is embodied in the following four dimensions:

[0064] Cost reduction: replacing manual standard conversion, exchange rate adjustment, report preparation, and other repetitive work to reduce the manpower investment of cross-border financial teams;

[0065] Efficiency improvement: compressing core processes such as financial report generation and tax compliance from "weekly" to "daily" to support rapid decision-making by enterprises;

[0066] Risk control: predicting tax compliance risks and financial data anomalies in advance to avoid penalties or additional tax payments due to rule deviations;

[0067] Compliance: meeting the requirements of multiple countries' data privacy and accounting regulations to provide stable financial compliance support for cross-border business expansion.

[0068] The above-described embodiments of the application are merely descriptive of its principles and its application, and are not intended to limit the application to the precise details as disclosed. Any modification or variation which comes within the scope of the application as defined by the appended claims is intended to be included. Any person skilled in the art can make equivalent modifications or variations without departing from the spirit and scope of the application.

[0069] Reference throughout this specification to "one embodiment", "an embodiment", or "the embodiments" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application and not necessarily in all embodiments. Thus, appearances of the phrases "in one embodiment", "in an embodiment", or "in specific embodiments" in various places throughout this specification are not necessarily referring to the same embodiment. Furthermore, the particular features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that other variations and modifications can be made to the embodiments described and illustrated herein, and that the applications described and illustrated herein can include other embodiments that are within the scope of the application. It will be apparent to those skilled in the art that significant changes and modifications can be made in the application without departing from the spirit and scope thereof.

[0070] In addition, any arrows in the drawings are merely exemplary and not limiting. Further, the term "or" as used herein is generally intended to mean "and / or" unless otherwise indicated. Combinations of components or steps will also be perceived as being apparent to those skilled in the art upon a reading of this disclosure.

[0071] As used in the description and the appended claims of the application, the singular forms "a", "an" and "the" include plural referents unless otherwise indicated. Similarly, as used in the description and the appended claims of the application, the term "in" includes "in" and "on" unless otherwise indicated.

[0072] The above description of the illustrated embodiments of the application (including what is described in the abstract) is not intended to be exhaustive or to limit the application to the precise forms disclosed. While specific embodiments of, and examples for, the application are described herein for illustrative purposes, various equivalent modifications are possible within the spirit and scope of the application, as those skilled in the relevant art will recognize and appreciate. As indicated, these modifications or variations can be made in light of the above descriptions of the illustrated embodiments of the application and are to be included within the spirit and scope of the application.

[0073] The systems and methods have been described herein in general terms as helpful to an understanding of the details of the application. In addition, various specific details have been set forth in order to provide a general understanding of the overall structure and operation of embodiments of the present application. It will be appreciated, however, that implementations of the present application can be practiced without some or all of the specific details set forth herein, or with other apparatus, systems, assemblies, methods, components, materials, parts and / or the like. In other instances, well known structures, materials, and / or operations have not been shown or described in detail in order to avoid obscuring aspects of the embodiments of the present application.

[0074] Thus, although the present application has been described in reference to specific embodiments thereof, many changes in the details thereof will be suggested to those skilled in the art, and it is intended in the application to encompass all such changes and modifications that fall within the scope of the appended claims. Accordingly, various modifications can be made in carrying out the application described herein without departing from the scope and spirit of the application. Therefore, the scope of the present application is not to be limited to the specific embodiments discussed above but only by the claims that follow.

Claims

1. A cross-border accounting compliance processing system based on federated learning and trusted computing, characterized in that, The system is applied to a multinational enterprise, and the multinational enterprise includes a plurality of financial nodes in different countries; the system comprises: a federal learning module, configured to synchronize and update learning models of all financial nodes by a federal average algorithm according to model parameters of the learning models trained locally by the financial nodes; wherein, a corresponding learning model is deployed in each financial node in advance, and the learning model is configured to adapt to the accounting standards of the current country, and is used for processing semantic understanding of local financial data and learning mapping rules of the accounting standards; a dynamic account book mapping module, configured to obtain exchange rate data of a country where each financial node is located in real time, and update a locally configured exchange rate conversion rule according to the exchange rate data, so as to automatically adjust currency amounts in an account book by sliding window backtracking and recalculation after the exchange rate conversion rule is updated; a trusted execution module, configured to perform execution of cross-chain contracts in an isolated container, including merging of the cross-chain contracts and account offsetting of the cross-chain contracts; a tax atlas reasoning module, configured to calculate correlation strength between nodes in a locally configured knowledge graph by an attention mechanism according to a feature vector of a transaction to be processed, so as to generate a probability distribution of a compliance path and identify an optimal tax processing scheme; a report generation module, configured to fill related data about accounting standard alignment output by the federal learning module and related data about exchange rate adjustment output by the dynamic account book mapping module into a report template by a preset field mapping rule, so as to generate a report in a standard format.

2. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, wherein, In the process of training and aggregating model parameters of each financial node, the federal learning module also protects sensitive financial data hidden in the model parameters by an ε-differential privacy mechanism.

3. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, wherein, When a learning model of any financial node is invalid, the federal learning module is also configured to dynamically adjust a weight proportion of the learning model of the financial node by a weighted average method.

4. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, characterized in that, The dynamic account book mapping module also identifies outliers in exchange rate data by an IQR method, and automatically eliminates or corrects abnormal data.

5. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, characterized in that, The dynamic account book mapping module is also configured with corresponding rules of account book fields of the accounting standards of each financial node, and is used for automatically identifying field meanings of a local account book of any financial node, and performing field alignment and conversion.

6. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, wherein, The trusted execution module is configured to run in an SGX container.

7. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, characterized in that, The trusted execution module is also configured to generate contract execution state snapshots at a preset frequency, so as to record execution progress, offsetting results and key parameters of the current cross-chain contract; wherein, the trusted execution module is also configured to compress the generated contract execution state snapshots by a Merkle Root, and synchronize a hash value obtained by compression and snapshot information to a main chain of a blockchain.

8. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, characterized in that, The trusted execution module is also configured to perform a contract rollback operation after an error occurs in contract execution, and generate a unique rollback event ID and a corresponding HASH record on the main chain of the blockchain; wherein, the record of the contract rollback operation and the record of the contract execution are stored in association.

9. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, wherein, The tax map reasoning module is previously constructed with a knowledge graph taking a country / region, a tax type, a tax rate, an industry type, a transaction type and a compliance clause as nodes, and each node is connected through an applicable relationship, a conflict relationship or a preferential relationship.

10. The cross-border accounting compliance processing system based on federated learning and trusted computing according to claim 1, wherein, The report generation module is trained by using a natural language generation model, and the report generation module further integrates an anomaly detection model for multidimensional detection of the generated report, including numerical anomaly detection, logical conflict detection and compliance deviation detection, and highlighting of risk points in the report. The anomaly detection model uses an isolation forest algorithm to identify the risk points in the report.

Images