Intelligent financial accounting system based on multi-dimensional data verification

Abstract

The invention relates to the technical field of intelligent financial accounting, and discloses an intelligent financial accounting system based on multi-dimensional data verification, comprising: S1, a multi-dimensional data verification module for performing real-time multi-dimensional data verification on a financial accounting system, comprising a business rule dimension verification unit, a time sequence continuity dimension verification unit, a multi-source association dimension verification unit and a risk scanning dimension verification unit. According to the intelligent financial accounting system based on multi-dimensional data verification, real-time risk scanning is realized through the multi-dimensional data verification module, threats can be rapidly isolated and resource utilization can be optimized in combination with hierarchical response control and a dynamic resource allocation mechanism when risks occur, and meanwhile, through trap data generation and a multi-dimensional two-way verification technology, the risk can be effectively eliminated. The method effectively improves the emergency disposal capability and data recovery reliability of the system in a complex risk scene, and has the advantage of dual improvement of financial data safety protection and system operation efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent financial accounting, and in particular to an intelligent financial accounting system based on multi-dimensional data verification. BACKGROUND

[0002] In the operation and management of the financial accounting system, the existing technology mainly focuses on the execution of the conventional accounting process and the verification of the basic data. Usually, according to the preset accounting task sequence, linear processing of financial data is carried out, such as sequentially completing data entry, calculation and summarization according to the transaction time sequence, subject category, etc., while simple rule verification is provided, such as balance check of borrowing and lending, basic format verification, etc., to ensure the preliminary accuracy of the accounting results. Such a mode can meet the basic accounting needs and support the routine operation of financial work when the business scenario is stable and the data risk is low.

[0003] However, the existing technology lacks systematicness in risk identification and response when dealing with complex risk scenarios. Most of them can only discover abnormalities after accounting, and it is difficult to perform real-time and multi-dimensional risk scanning before and during task execution. In the face of potential threats such as data tampering and process vulnerabilities, it is difficult to accurately locate the accounting tasks associated with risks, which can easily lead to risk diffusion, affecting the integrity and reliability of financial data. In addition, the resource scheduling mechanism is rigid, and there is no dynamic adjustment strategy adapted to the risk level. When the system encounters risks, it cannot reasonably allocate resources to prioritize critical tasks. In the context of high-risk and multiple risks, there is a lack of effective emergency disposal and vulnerability tracing means, which cannot quickly restore the normal accounting order of the system, nor can it deeply analyze and prevent the direction of risk attacks, increasing the risk of financial data security and business interruption. SUMMARY

[0004] The technical problem to be solved by the present application is that the existing technology has the drawback of response lag leading to limited defense. Therefore, we propose an intelligent financial accounting system based on multi-dimensional data verification.

[0005] In order to achieve the above purpose, the following technical scheme is adopted in the present application: an intelligent financial accounting system based on multi-dimensional data verification, comprising:

[0006] S1: a multi-dimensional data verification module for real-time multi-dimensional data verification of the financial accounting system, including a business rule dimension verification unit, a time sequence continuity dimension verification unit, a multi-source association dimension verification unit and a risk scanning dimension verification unit;

[0007] S2: a risk rating and association module for dividing risks into three levels of high, medium and low according to the type, impact range and severity of abnormal data when the multi-dimensional data verification module finds abnormal data, and identifying the accounting task chain associated with the risks;

[0008] S3: hierarchical response control module, comprising:

[0009] high-risk response unit for locking associated accounting tasks and data storage areas, prohibiting task execution and data modification, and releasing computing resources occupied by locked tasks;

[0010] low-risk response unit for adding a verification mechanism to the associated task node and limiting the upper limit of resource usage;

[0011] S4: resource dynamic allocation module for reallocating computing resources released by the hierarchical response control module in S3 to risk-free accounting tasks and improving their execution priority;

[0012] S5: emergency processing module for performing emergency disposal when multiple high risks are detected, comprising:

[0013] task control unit for stopping all accounting task execution;

[0014] data locking unit for locking all system accounting data;

[0015] decoy data generation unit for generating simulated transaction data based on historical real business data as false financial data, the false financial data containing invisible digital watermark marks; an independent monitoring channel is established to monitor false data tampering behavior;

[0016] S6: risk removal module for performing removal processing after a risk removal signal is triggered, restoring the priority of the accounting tasks and performing re-accounting.

[0017] Preferably, the business rule dimension verification unit is used to verify the balance of borrowing and lending and tax compliance, the time sequence continuity dimension verification unit is used to verify the hooking relationship of cross-period data, the multi-source association dimension verification unit is used to compare the consistency of contract, invoice and bank flow data, and the risk scanning dimension verification unit is used to detect abnormal data in system operation.

[0018] Preferably, after the risk removal signal is triggered, the removal processing is performed, the locked data and the original benchmark data are verified in multiple dimensions and bidirectionally based on the business rule dimension and the time sequence continuity dimension of S1, the difference points are marked and a risk analysis report is generated, and the data is unlocked through a two-factor authentication mechanism combining biometric recognition and dynamic password.

[0019] Preferably, the risk rating and association module is configured to: high risk is a threat to the completeness of core financial data; medium risk is a threat to non-core business processes; and low risk is an automatically repairable anomaly.

[0020] Preferably, the multi-dimensional data verification module is configured to perform verification in a preset period or real-time event triggering mode, wherein:

[0021] The business rule dimension verification unit is based on a preset rule library, which includes corporate accounting standards and tax regulations for verification; the risk scanning dimension verification unit identifies abnormal data in system operation through cluster analysis.

[0022] Preferably, the resource allocation module establishes a resource isolation pool to implement CPU and memory isolation, and adopts a weighted priority scheduling algorithm to allocate computing resources; implements a dynamic resource quota adjustment mechanism, and the formula is: new quota = basic quota × (1 + risk coefficient × task urgency)

[0023] The task urgency range is ; Risk factor Classification by risk level:

[0024] High risk , medium risk , low risk .

[0025] Preferably, the emergency response module establishes false financial data generation by using a GAN neural network to simulate a real transaction model; the digital watermark is implanted using an invisible watermark algorithm based on wavelet transform;

[0026] Tamper monitoring implements operational behavior profiling analysis, generates attack path heat maps, and establishes a risk management sandbox environment to isolate and trap operations.

[0027] Preferably, the risk relief module uses multi-dimensional two-way verification including business rule review, temporal logic backtracking and multi-source data re-matching, difference point marking uses a comparative learning model to identify abnormal patterns, two-factor authentication integrates the FIDO2 security authentication protocol, and the recalculation process implements differential data version rollback and compensation accounting.

[0028] Preferably, the risk rating and association module is configured with a risk scoring model, which satisfies: Risk score = Severity Factor+ Impact range coefficient+ Repair difficulty coefficient, where: the weight constraint relationship is , and the weight value range is limited to: severity weight , influence range weight , fix difficulty weight The severity coefficient is assigned based on the amount of data tampering and the compliance risk level, and the assignment range is The impact range coefficient is assigned according to the number of affected tasks and the number of affected modules, and the assignment range is .

[0029] The calculation resource released by the hierarchical response control module is preferably allocated according to a calculation formula for risk-free accounting tasks: , wherein the priority coefficient is determined by the task urgency, and the value range is ; the service value coefficient has a value range of , and is obtained by weighted calculation, the base priority is a base parameter preset by the system for normalized calculation, and has a value of a constant greater than 0.

[0030] Technical effects and advantages of the present application:

[0031] In the present application, real-time risk scanning is realized through the multi-dimensional data verification module, combined with hierarchical response control and dynamic resource allocation mechanism, which can quickly isolate threats and optimize resource utilization when risks occur, and through the trap data generation and multi-dimensional two-way verification technology, the emergency disposal capability and data recovery reliability of the system in complex risk scenarios are effectively improved, and the system has the advantages of double improvement of financial data security protection and system operation efficiency. Risk identification is accurate: through the multi-dimensional data verification module, the financial data is monitored in real time from the aspects of business rules, time sequence continuity, multi-source association and risk scanning, compared with the single-dimensional verification of the prior art, the abnormality can be found more comprehensively and timely, and the risk type and influence range can be accurately identified. BRIEF DESCRIPTION OF DRAWINGS

[0032] The disclosure of the present application will be described with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present application. In the drawings, the same reference numerals are used to refer to the same parts:

[0033] Fig. 1 is a flow chart of the present application;

[0034] Fig. 2 is a logic decision diagram of the present application;

[0035] Fig. 3 is a structure block diagram of the emergency processing module of the present application. DETAILED DESCRIPTION

[0036] It is easy to understand that, according to the technical scheme of the present application, those skilled in the art can propose a plurality of structure modes and implementation modes which can be replaced with each other without changing the essential spirit of the present application. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical scheme of the present application, and should not be regarded as the whole or as a limitation or restriction of the technical scheme of the present application.

[0037] Referring to Figs. 1-3 , the present application provides a technical scheme: an intelligent financial accounting system based on multi-dimensional data verification, comprising:

[0038] S1: Construct a multi-dimensional verification system, in which the multi-dimensional data verification module contains four parallel verification units: the business rule dimension verification unit is used to verify the credit and debit balance and tax compliance, which is specifically implemented by loading the enterprise accounting standards into the rule engine; the time series continuity dimension verification unit is used to detect the cross-period data cross-reference relationship, which is implemented through the time series analysis algorithm; the multi-source association dimension verification unit is used to compare the consistency of contracts, invoices and bank statements, which is implemented by data fusion technology; the risk scanning dimension verification unit is used to identify abnormal data in system operation.

[0039] The multidimensional data verification module is configured to perform verification according to preset cycles or real-time event triggering. The business rule dimension verification unit is based on a preset rule library, which contains corporate accounting standards and tax regulations for verification; the risk scanning dimension verification unit identifies abnormal data in system operation through cluster analysis.

[0040] Among them, the preset period trigger method means starting the data verification process at a fixed time interval, performing a full verification every 30 minutes, and immediately starting the verification when the system detects a specific business operation or data change. This is achieved by monitoring transaction events through the message queue.

[0041] The pre-set rule base in the business rule dimension verification unit converts normative documents such as corporate accounting standards and tax regulations into a programmable set of verification logic, which is then used by the Drools rule engine. Cluster analysis in the risk scanning dimension verification unit uses unsupervised machine learning algorithms to group financial data and identify anomalies that deviate from the normal data distribution.

[0042] The combination of preset periodic triggering and real-time event triggering can ensure the regular periodic full-volume scanning while performing instant verification for sudden transactions or data changes. The business rule dimension verification unit loads corporate accounting standards and tax regulations through the rule engine. In the invoice verification scenario, it automatically checks the matching of input tax and output tax, or verifies the compliance of tariff calculations for cross-border transactions. The dual-mode trigger mechanism covers full-process risk monitoring, combining rule-driven and data-driven dual verification, which not only ensures the rigid constraints of core business rules, but also identifies hidden abnormal patterns that traditional rule bases cannot cover. This is achieved through cluster analysis algorithms. When the system is running, the module continuously performs four-dimensional cross-validation on financial data, breaking through the limitations of single-dimensional verification and achieving early risk detection.

[0043] The business rule dimension verification unit refers to verifying the balance relationship of the accounting subjects of the financial data and the tax compliance through a preset rule library. Specifically, it can be implemented by using a rule engine combined with an expert system, and is used to ensure that the debit and credit amounts of each transaction match and meet the tax reporting requirements. The time sequence continuity dimension verification unit refers to performing time sequence logical verification on the financial data across accounting periods, which is implemented by using a timestamp verification algorithm combined with historical data trend analysis, and is used to detect accounting period connection abnormalities or data fault problems. The multi-source association dimension verification unit refers to cross-verification of three types of heterogeneous data, namely, contract text, electronic invoice image and bank transaction flow. Specifically, it can extract key contract clauses by using natural language processing technology, parse invoice information by using optical character recognition, and compare the amounts and timestamps with the bank flow. The risk scanning dimension verification refers to identifying abnormal operations through real-time monitoring of system operation logs and data change records, and detecting data points deviating from the normal mode based on a density-based clustering algorithm. The business rule dimension verification unit performs automatic balance verification on the debit and credit subject amounts of each transaction through the rule library of built-in accounting standards and tax clauses, and matches the tax reporting data with the account processing results. The time sequence continuity dimension verification unit establishes a cross-period data hook verification relationship model, identifies abnormal transactions that are not processed according to the accrual basis by analyzing the time sequence association of accounts receivable and collected funds. The multi-source association dimension verification unit constructs the mapping relationship of contract number, invoice code and bank flow number, and uses data fingerprint technology to verify the consistency of the transaction subject, amount and time information of the three. The risk scanning dimension verification unit triggers an alarm when detecting abnormal high-frequency operations or non-working time data modification by continuously collecting system operation logs and establishing a user behavior baseline model.

[0044] The traditional financial accounting system only performs basic debit and credit balance verification and format verification, and cannot verify the data logical association across accounting periods. The establishment of multi-source data association verification can find the differences between contract performance and fund collection and payment. The added risk scanning dimension can capture abnormal operation behaviors in system operation in real time. The multi-dimensional real-time verification of financial accounting data is realized, and four types of risks, i.e., business rule violation, cross-period data fault, multi-source data contradiction and system operation abnormality, are effectively identified. The data compliance, continuity, consistency and system security detection are completed synchronously during the execution of the accounting task, avoiding abnormal data from entering the subsequent accounting process.

[0045] S2: When the multi-dimensional data verification module detects an anomaly, the process enters the risk rating and association module. This module uses a weighted scoring model to calculate risk values ​​based on the anomaly type (e.g., data tampering, logical vulnerabilities), the impact on the number of accounting tasks, the financial modules, the severity, and the impact on data integrity and business continuity. It then categorizes the risk into high, medium, and low levels. It also uses a graph database to track task chain relationships and mark affected upstream and downstream task chains. High risk refers to abnormal data that directly compromises the accuracy of financial statements or the security of funds. This can be achieved through core database access log monitoring and transaction amount mutation detection, which can be used to block tampering with key information such as general ledger data and bank account balances.

[0046] Medium risk refers to abnormal data that interferes with the efficiency of business process execution or auxiliary accounting modules. This can be achieved through business process integrity verification and auxiliary accounting task dependency analysis to limit the impact on non-core processes such as expense reimbursement and accounts receivable settlement.

[0047] Low risk refers to abnormal data that does not affect the accounting results and complies with preset repair rules. This can be achieved by automatically replacing abnormal data or using a logical self-correction algorithm to repair problems that can be standardized, such as duplicate voucher numbers and missing auxiliary accounting items.

[0048] When the multidimensional data verification module detects an anomaly, the system first determines the risk level based on the object of the abnormal data. If the anomaly involves general ledger account balances or bank reconciliation data, it is marked as high risk and triggers a core data lock. If the anomaly exists in the expense allocation or accounts receivable write-off process, it is marked as medium risk and the process verification mechanism is activated. If the anomaly is due to an incorrect voucher entry format or a missing auxiliary accounting item, it is marked as low risk and the preset repair rules are executed. By classifying the threat objects into three categories: core data integrity, non-core process operation status, and standardizable anomalies, the system can specifically invoke three disposal modes: high-risk blocking, medium-risk isolation, and low-risk self-repair, avoiding the resource mismatch problem caused by risk level confusion in traditional solutions. By establishing a three-level classification mechanism, the system can prioritize blocking core data tampering, isolate non-core process anomalies in a targeted manner, and at the same time release low-risk disposal resources to achieve simultaneous improvement in risk disposal efficiency and accuracy. It can effectively isolate core financial data threats, reduce the probability of interruption of non-core business processes, and improve the processing efficiency of repairable anomalies, avoiding resource mismatch and response delays caused by confusion about risk levels in traditional solutions, accurately locating risk transmission paths, and providing a basis for subsequent graded responses.

[0049] The risk rating and association module is configured with a risk scoring model that satisfies: Risk score = Severity Factor+ Impact range coefficient+ The repair difficulty coefficient, wherein: the weight constraint relationship is , and the weight value interval is limited to: severity weight , impact range weight , repair difficulty weight The severity coefficient is valued according to the data tampering amount and compliance risk level, and the value range is The impact range coefficient is valued according to the number of affected tasks and the number of affected modules, and the value range is .

[0050] When the system detects abnormal data, the risk scoring model first calculates the severity coefficient according to the data tampering amount and compliance risk level, then calculates the impact range coefficient according to the number of affected tasks and the number of modules, and the repair difficulty coefficient is valued according to the preset repair resource consumption table. The weight of each coefficient is dynamically adjusted according to the constraint relationship. Finally, the risk score is generated by weighted summation, and the risk is divided into high, medium and low levels, providing a quantitative basis for subsequent resource scheduling.

[0051] By constructing a multi-dimensional scoring model, the severity, impact range and repair difficulty are dynamically weighted, solving the problem of mismatch between risk level division and actual threat level in traditional methods. For example, when a high-risk event has a small impact range, the scoring model can avoid over-response through weight adjustment; when a low-risk event involves multiple modules, the model can automatically increase its risk level.

[0052] S3: Based on the risk rating and associated modules output, the hierarchical response control module starts the corresponding strategy. For high-risk, the high-risk response unit realizes task lock and storage isolation mechanism through distributed lock technology, freezes the associated accounting task process, locks the corresponding data storage area write permission, blocks risk spread, and releases the CPU and memory resources occupied by the locked task; for medium and low risk, the medium and low risk response unit dynamically injects verification logic with the help of service mesh technology, adds additional verification nodes to the associated task nodes, uses secondary data cross-validation, manual review nodes, and sets the upper limit of resource usage to limit the computing resources that can be called by the task, realizing risk control and minimizing business impact.

[0053] S4: After the resource dynamic allocation module receives the released resources of the hierarchical response control module, it reallocates resources using the weight priority scheduling algorithm and achieves computing resource isolation and rapid migration through containerization technology. The released CPU, memory, etc. are preferentially allocated to risk-free core accounting tasks, and the weight is dynamically adjusted according to the task urgency, business value, etc. to improve the priority of core task execution. The resource allocation module establishes a resource isolation pool to implement CPU and memory isolation, and uses a weight priority scheduling algorithm to allocate computing resources; a dynamic resource quota adjustment mechanism is implemented, and the formula is: new quota = basic quota x (1 + risk coefficient x task urgency)

[0054] wherein the task urgency value range is ; the risk coefficient According to the risk level:

[0055] high risk , medium risk , low risk , establish a dynamic association between risk level and resource scheduling to ensure critical business while optimizing resource utilization.

[0056] The resource isolation pool deploys high-risk tasks and risk-free accounting tasks in independent computing environments through physical isolation mechanisms. When a high-risk task is detected, the CPU and memory resources it occupies are limited within the isolation pool to avoid resource preemption. By introducing a dynamic adjustment factor of risk coefficient and task urgency, a resource quota calculation model is established that links to the risk level, so that the resource allocation strategy can respond to changes in system risk state in real time. At the same time, the physical isolation mechanism of the resource isolation pool breaks through the logical isolation limitations of traditional virtualization resource pools and blocks resource abnormal occupation from the bottom hardware layer.

[0057] The computing resources released by the hierarchical response control module are allocated to risk-free accounting tasks using the following formula: wherein: the priority coefficient is determined by the task urgency, with a value range of ; the business value coefficient has a value range of , which is obtained by weighted calculation. The baseline priority is a baseline parameter preset by the system for normalization calculation, and its value is a constant greater than 0.

[0058] The mathematical calculation model is constructed to quantitatively couple the task urgency and the business value. The priority coefficient generates a dynamic value according to the remaining processing time of the task or the criticality of the associated business process, and the business value coefficient is given a fixed weight based on the support of the accounting task for the generation of financial statements or tax returns. The two are combined by weighted calculation to form the resource allocation ratio. The benchmark priority is a normalization parameter to ensure that the total amount of resources released under different risk levels dynamically matches the task demand. When high-risk scenarios release computing resources, high-urgency tax accounting tasks can obtain resource allocation based on formula calculation results, while low-business-value daily statistical tasks automatically reduce resource occupation proportion.

[0059] S5: When the system detects multiple high-risk events concurrently, trigger the emergency processing module. This module immediately stops all task execution through the task control unit, and locks the system data through the data locking unit; the decoy data generation unit learns historical data based on a generative adversarial network, simulates real transaction data as false financial data, and embeds invisible digital watermark marks in the false data through a wavelet transform algorithm; an independent monitoring channel is established to track data tampering behavior in real time,

[0060] The GAN neural network simulates real transaction patterns by learning the distribution of historical real business data using a generative adversarial network model, generating simulated transaction data with the same statistical characteristics. Through the adversarial training of the generator and the discriminator, the data generation quality is optimized. The invisible watermark algorithm of wavelet transform embeds digital watermark in the high-frequency sub-band coefficients of the financial data, realizing hidden marking on the premise of ensuring the integrity of the data surface. The operation behavior portrait analysis extracts the behavior feature vector by collecting user operation logs, and constructs the mapping relationship between operation features and risk levels.

[0061] Specifically, when multiple high-risk concurrency is detected, the emergency handling module starts the false financial data generation process. The GAN neural network generates simulated transaction records based on historical real transaction data, which is consistent with the real business mode in terms of transaction amount, time distribution, subject association, etc. At the same time, through the wavelet transform algorithm, invisible digital watermark is embedded in the data. The generated false data is put into an independent monitoring channel. When the attacker attempts to tamper with the data, the operation behavior portrait analysis unit collects the operation trajectory in real time, extracts the operation frequency, access path, modification mode and other characteristics, and identifies abnormal behavior through a pre-trained classification model. The detected abnormal operation is guided to the sandbox environment for execution. In this process, the attack path heat map is dynamically updated, and the task nodes triggered by the abnormal operation and their propagation paths are marked. The sandbox environment ensures that the real accounting data is not affected through resource isolation mechanism, while recording the operation steps and data modification traces of the attacker, and quickly generating high-fidelity trap data in high-risk concurrency, effectively confusing the attack target and recording the tampering behavior characteristics; through invisible watermarking, the data tampering is traced back, and the attacked data flow is accurately identified; using operation behavior analysis and heat map visualization technology, the attack path is tracked in real time and the diffusion direction is predicted; the sandbox isolation environment protects the integrity of the real accounting data while capturing the attack process data completely, providing a traceable operation evidence chain for risk disposal.

[0062] S6: After the risk is removed, the risk removal module starts the recovery process, ensures data integrity through multi-dimensional bidirectional verification, identifies the difference between the locked data and the original baseline data using a contrast learning model; marks the difference and generates a risk analysis report; completes the security unlocking by an administrator through a two-factor authentication mechanism combining biometric features and dynamic passwords; restores the original task priority, performs compensation accounting, and re-executes the accounting of tasks that were not completed or affected during the risk period according to the priority, ensuring the accuracy of financial data and business continuity.

[0063] After the risk removal signal is triggered, the removal process is performed, and the locked data and the original baseline data are verified in multiple dimensions based on business rule dimensions and time sequence continuity dimensions. The verification difference is marked and a risk analysis report is generated, and the data is unlocked through a two-factor authentication mechanism combining biometric features and dynamic passwords.

[0064] Among them, the multi-dimensional bidirectional verification refers to the bidirectional cross verification of the locked data and the original baseline data in the business rule dimension and the time sequence continuity dimension. Through business rule reinspection and time sequence logic backtracking algorithm, the contrast learning model identifies abnormal patterns, positively verifies whether the locked data conforms to the business rules, and reversely verifies whether the original baseline data has logical discontinuity, so as to find data tampering that cannot be identified by one-way verification.

[0065] Difference point marking refers to identifying data deviation locations and logical breakpoints through comparative learning models. Specifically, it can be implemented using an abnormal pattern detection algorithm based on the attention mechanism. For example, the business rule verification differences and time series continuity differences can be correlated and analyzed to generate structured difference marking data, providing precise coordinates for risk tracing.

[0066] The two-factor authentication mechanism refers to the combination of biometric recognition and dynamic passwords for identity authentication. It can be implemented using fingerprint recognition and a one-time password generator. The biometric collection module and the dynamic password verification interface are integrated through the FIDO2 security protocol to block the risks of password leakage and replay attacks.

[0067] When the risk release signal is triggered, the locked data and the original benchmark data are first bidirectionally verified in the business rule dimension. The forward verification is to see whether the locked data meets the credit and debit balance and tax compliance, and the reverse verification is to see whether the original benchmark data has a broken cross-period cross-checking relationship. If there is an abnormally modified invoice amount in the locked data, the forward verification will find its deviation from the contract amount, and the reverse verification can detect whether the corresponding bank flow in the original benchmark data has been tampered with. Subsequently, the temporal continuity dimension verification verifies whether the data changes conform to the logical order through time series analysis, and detects whether there are accounting operations that have not completed the pre-approval. The differences generated during the verification process are marked by the comparative learning model to identify the combination pattern of abnormal numerical fluctuations and logical conflicts, and generate a risk analysis report containing the difference location, type and association rules. Finally, the unlocking operation requires two-factor authentication of fingerprint recognition and dynamic password. The operator needs to complete fingerprint verification at the designated terminal and enter the 6-digit dynamic password generated in real time to ensure the legality of the operation authority. In the invoice data tampering scenario, two-way verification can simultaneously detect abnormalities in the tampered invoice amount and logical breaks in the associated contract data. The difference marking model automatically locates the tampered items and generates repair suggestions. Two-factor authentication prevents attackers from using stolen passwords to tamper with the data again.

[0068] The risk mitigation module utilizes multi-dimensional, two-way validation, including business rule revalidation, temporal logic backtracking, and multi-source data rematching. Difference marking utilizes a contrastive learning model to identify anomalous patterns. Two-factor authentication integrates the FIDO2 security authentication protocol. The recalculation process implements version rollback and compensation accounting for discrepancies. Business rule revalidation involves reloading the accounting standards and tax compliance rule base to verify the debit / credit balance and tax entries of locked data, eliminating recovery bias caused by rule updates or conflicts. Temporal logic backtracking utilizes a sliding time window mechanism to verify the continuity of cross-period data and detect data gaps by comparing cumulative values ​​from previous and subsequent periods. Multi-source data rematching establishes a cross-index table of contract numbers, invoice codes, and bank serial numbers, forcing a three-party data consistency check. During the difference marking phase, the contrastive learning model maps anomalous data and baseline data into a high-dimensional feature space, identifying patterns that deviate from the normal distribution through similarity calculations. The FIDO2 protocol requires users to provide both a fingerprint and a dynamically generated one-time key when performing an unlock operation, using cryptographic binding to prevent authentication information leakage. Data version rollback selects an uncontaminated baseline snapshot for recovery based on the version chain, while compensation accounting ensures auditability of the repair process by generating a difference log. Multi-dimensional, bidirectional verification covers rule, time series, and multi-source correlation risks, utilizing comparative learning to improve anomaly detection sensitivity. Integrating the FIDO2 protocol strengthens identity authentication security, and through version rollback and compensation accounting, data repair accuracy and traceability are ensured.

[0069] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. An intelligent financial accounting system based on multidimensional data verification, characterized in that: include: S1: Multidimensional data verification module, which performs real-time multidimensional data verification on the financial accounting system, including business rule dimension verification unit, time series continuity dimension verification unit, multi-source association dimension verification unit and risk scanning dimension verification unit; S2: Risk rating and association module, which is used to classify risks into high, medium, and low levels based on the type, impact scope, and severity of abnormal data when the multidimensional data verification module finds abnormal data, and identifies the accounting task chain associated with the risk; S3: Hierarchical response control module, including: High-risk response unit, used to lock associated accounting tasks and data storage areas, prohibit task execution and data modification, and release computing resources occupied by locked tasks; Medium and low-risk response units are used to add verification mechanisms to associated task nodes and limit resource usage; S4: Dynamic resource allocation module, used to reallocate the computing resources released by the hierarchical response control module in S3 to risk-free accounting tasks and increase their execution priority; S5: Emergency handling module, used to perform emergency handling when multiple high risks are detected, including: Task control unit, used to stop the execution of all accounting tasks; Data locking unit, used to lock the accounting data of the entire system; A trapping data generation unit is used to generate simulated transaction data based on historical real business data as false financial data, wherein the false financial data contains an invisible digital watermark; and establish an independent monitoring channel to monitor the tampering of false data; S6: Risk release module, used to execute release processing after the risk release signal is triggered, restore the accounting task priority and perform recalculation.

2. The intelligent financial accounting system based on multidimensional data verification according to claim 1, characterized in that: The business rule dimension verification unit is used to verify the credit and debit balance and tax compliance, the time series continuity dimension verification unit is used to verify the cross-reference relationship of cross-period data, the multi-source association dimension verification unit is used to compare the consistency of contracts, invoices and bank flow data, and the risk scanning dimension verification detects abnormal data in the system operation.

3. The intelligent financial accounting system based on multidimensional data verification according to claim 1 is characterized in that: After the risk release signal is triggered, the release process is executed. Based on the business rule dimension and time continuity dimension described in S1, the locked data and the original benchmark data are verified in a multi-dimensional and bidirectional manner. The verification differences are marked and a risk analysis report is generated. The data is unlocked through a two-factor authentication mechanism combining biometric recognition and dynamic passwords.

4. The intelligent financial accounting system based on multidimensional data verification according to claim 1, characterized in that: The risk rating and association module configuration is as follows: high risk is a threat that affects the completeness of core financial data; medium risk is a threat that affects non-core business processes; and low risk is anomalies that can be automatically repaired.

5. The intelligent financial accounting system based on multidimensional data verification according to claim 1 is characterized in that: The multidimensional data verification module is configured to perform verification according to a preset period or in a real-time event triggering manner, wherein: The business rule dimension verification unit is based on a preset rule library, which includes corporate accounting standards and tax regulations for verification; the risk scanning dimension verification unit identifies abnormal data in system operation through cluster analysis.

6. The intelligent financial accounting system based on multidimensional data verification according to claim 1, characterized in that: The resource allocation module establishes a resource isolation pool to implement CPU and memory isolation, and uses a weighted priority scheduling algorithm to allocate computing resources; it implements a dynamic resource quota adjustment mechanism, with the formula: new quota = basic quota × (1 + risk factor × task urgency) The task urgency range is ; Risk factor Classification by risk level: High risk , medium risk , low risk .

7. The intelligent financial accounting system based on multidimensional data verification according to claim 1, characterized in that: The emergency response module uses GAN neural network to simulate real transaction patterns to generate false financial data; digital watermarks are embedded in an invisible watermark algorithm based on wavelet transform; Tamper monitoring implements operational behavior profiling analysis, generates attack path heat maps, and establishes a risk management sandbox environment to isolate and trap operations.

8. The intelligent financial accounting system based on multidimensional data verification according to claim 1, characterized in that: The risk mitigation module uses multi-dimensional two-way verification, including business rule review, temporal logic backtracking, and multi-source data rematching. The difference point marking uses a comparative learning model to identify abnormal patterns. Two-factor authentication integrates the FIDO2 security authentication protocol. The recalculation process implements differential data version rollback and compensation accounting.

9. The intelligent financial accounting system based on multidimensional data verification according to claim 1, characterized in that: The risk rating and association module is configured with a risk scoring model that satisfies: Risk score = Severity Factor+ Impact range coefficient+ Repair difficulty coefficient, where: the weight constraint relationship is , and the weight value range is limited to: severity weight , influence range weight , fix difficulty weight The severity coefficient is assigned based on the amount of data tampering and the compliance risk level, and the assignment range is The impact range coefficient is assigned according to the number of affected tasks and the number of affected modules, and the assignment range is .

10. The intelligent financial accounting system based on multidimensional data verification according to claim 1, characterized in that: The calculation formula for allocating computing resources released by the hierarchical response control module and risk-free accounting tasks is: , where: the priority coefficient is determined by the urgency of the task and has a value range of ;Business value coefficient value range , obtained through weighted calculation, the benchmark priority is a benchmark parameter preset by the system for normalization calculation, and its value is a constant greater than 0.

Images