Financial data intelligent automatic accounting method and system

Abstract

The application discloses a kind of financial data intelligent automation accounting method and system, it is related to financial data accounting technical field, when system runs, through intelligent intervention suggestion generation module, operation type, specific, responsibility attribution explicit intervention action suggestion Ψs, such as supplementary invoice, suspend payment and budget structure reexamine etc., are deduced based on exception type R and scene vector A, greatly reduce the judgment burden of financial personnel.This mechanism makes up the short board that traditional system cannot provide business intervention suggestion, strengthens the operation orientation and landing ability of system.Periodic trajectory normalization and adaptive learning module takes historical behavior trajectory and intervention response result as system learning material, continuously iterates abnormal identification mechanism and trajectory expectation model, dynamically optimizes comprehensive abnormal index ΔF and outputs revised abnormal deviation index nΔF, so that the system has feedback→adjustment→evolution learning closed loop ability.

Description

Technical Field

[0001] This invention relates to the field of financial data accounting technology, specifically to an intelligent and automated financial data accounting method and system. Background Technology

[0002] Automated financial systems not only handle traditional accounting and reconciliation functions, but are increasingly taking on the task of judging the logical rationality and verifying the compliance of income and expenditure behavior in complex business scenarios. Especially in project-based, contract-based, or multi-organizational collaborative enterprise management structures, specific judgments such as whether contract progress milestones match cash flow behavior, whether counterparty credit behavior is stable, and whether budget adjustments are reasonable increasingly rely on the intelligent processing capabilities of financial systems. Therefore, the deep integration of financial behavior with scenario semantics has become a key touchpoint for accounting systems to move towards "automatic identification of abnormal deviations."

[0003] While most current financial accounting systems possess functions such as automatic data entry, automatic classification, and automatic voucher generation, they still suffer from a significant flaw in financial judgment scenarios involving multiple scenarios and business types: they only process numerical strengths and ignore the semantic layer of behavior. Especially in actual business scenarios such as "accounts receivable not collected," "cash inflow exceeding budget," and "payment before performance," the system often fails to automatically identify whether such behavior constitutes a compliance deviation, relying on post-event discovery by financial personnel and manual analysis.

[0004] The root cause of the above problems lies in the fact that most current systems adopt a "single voucher processing mechanism," ignoring the trajectory chain and semantic context formed by financial behavior over time, and failing to integrate scenario-based factors such as contract progress, project status, and counterparty performance history into the trajectory logic. This lack of a holistic modeling approach integrating behavior, time, and context prevents the system from promptly capturing and intervening in atypical deviations during business execution. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent and automated accounting method and system for financial data, which solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a method and system for intelligent and automated accounting of financial data, including a scenario factor extraction and semantic aggregation module, a financial behavior trajectory reconstruction module, a deviation pattern recognition module, a multi-type risk automatic labeling module, an intelligent intervention suggestion generation module, and a periodic trajectory normalization and adaptive learning module;

[0007] The scenario factor extraction and semantic aggregation module is responsible for collecting scenario parameters that affect the compliance of financial behavior, preprocessing them, and fitting them into a scenario vector set A.

[0008] The financial behavior trajectory reconstruction module relies on the scenario vector set A to decompose income and expenditure operations into events and constructs the structural trajectory of the actual behavior according to the cycle to obtain the intensity T of the behavior trajectory.

[0009] The deviation pattern recognition module uses trajectory intensity T as a basis and combines it with scene vector set A to calculate the degree of deviation between the actual financial operation trajectory and the expected trajectory, and obtains the comprehensive anomaly index ΔF.

[0010] The multi-type risk automatic labeling module obtains the dynamic anomaly threshold Thf based on the acquired comprehensive anomaly index ΔF, and performs semantic classification on the comprehensive anomaly index ΔF to obtain the anomaly type R;

[0011] The intelligent intervention suggestion generation module generates an intelligent intervention operation set Ψs based on the anomaly type R and the scene vector set A, obtains the intervention results, and generates suggested actions to be performed.

[0012] The periodic trajectory normalization and adaptive learning module records the trajectory intensity T and intervention results within historical periods, readjusts the comprehensive anomaly index ΔF, and obtains the anomaly deviation index nΔF.

[0013] Preferably, the scene factor extraction and semantic aggregation module includes a scene parameter extraction unit and a normalization and semantic fitting unit;

[0014] The scenario parameter extraction unit extracts scenario parameters from the data in the financial system, including the contract performance stage coefficient Ahx, the counterparty performance stability Ads, and the project funding tension coefficient Axm, and fits them into the original scenario dataset YA.

[0015] Among them, the contract performance stage coefficient Ahx is obtained by analyzing the contract performance platform. The contract performance stage coefficient Ahx is obtained by reading the ratio of the current contract performance node progress Preal to the total performance progress plan Pplan for each project.

[0016] Counterparty performance stability (Ads) is obtained by analyzing past payment records and payment cycle logs;

[0017] Extract the series of payment deviation values ​​within 5 historical periods, then calculate the series standard deviation σds to construct the counterparty performance stability Ads;

[0018] Counterparty performance stability (Ads) is obtained using the following formula:

[0019] ;

[0020] The project funding tension coefficient Axm is obtained by collecting data from the financial budget and actual payment execution system.

[0021] The project funding tension coefficient Axm is obtained using the following formula:

[0022] ;

[0023] In the formula, Bm represents the project budget balance, and Pm represents the total amount of current accounts payable;

[0024] The normalization and semantic fitting unit normalizes the acquired original scene dataset YA to obtain the scene vector set A;

[0025] Scene vector set A is obtained using the following formula:

[0026] ;

[0027] In the formula, Ao represents the o-th data in the scene vector set A, YAo represents the o-th data in the original scene dataset YA, minYAo represents the valley value of the o-th data in the original scene dataset YA, and maxYAo represents the peak value of the o-th data in the original scene dataset YA.

[0028] Preferably, the financial behavior trajectory reconstruction module includes an income and expenditure event extraction unit and a behavior trajectory intensity construction unit;

[0029] The revenue and expenditure event extraction unit decomposes the revenue and expenditure operations in the original financial behavior into event-based decomposition, transforms them into structured event tuples e, and maps and groups the behavioral events according to the natural cycle based on the accounting period to obtain the event cycle behavior Ej.

[0030] The structured event tuple e is obtained using the following formula:

[0031] ek={Typek,stk,etk,mk,ϕk};

[0032] In the formula, ek represents the k-th financial action in the structured event tuple, Typek represents the event type of the k-th financial action, including invoice, receipt and carry-forward, stk represents the start time of the k-th financial action, etk represents the end time of the k-th financial action, mk represents the amount involved in the k-th financial action, and ϕk represents the event context label of the k-th financial action, such as project number, contract number, etc.

[0033] The event lifecycle behavior Ej is obtained using the following formula:

[0034] Ej = {e1, e2, ..., ek | etk ∈ period j};

[0035] Based on the contract performance stage coefficient Ahx and project funding tension coefficient Axm in the scenario vector set A, a time-series weighting function ωj is constructed;

[0036] The time-series weighting function ωj is obtained by summing the contract performance stage coefficient Ahx and the project funding tension coefficient Axm and then dividing by 2.

[0037] Preferably, the behavior trajectory intensity construction unit calculates the contribution intensity of all event period behaviors Ej in each period to the trajectory surface and obtains the trajectory intensity T.

[0038] The trajectory intensity T is obtained using the following formula:

[0039] ;

[0040] In the formula, T(i,j) represents the trajectory intensity of the i-th type of behavior within period j, Eji represents the event set of the i-th type of behavior within period j, Δtk represents the behavior time span, which is obtained by the difference between the start time stk and the end time etk of the k-th financial behavior, and Xk represents the behavior density correction function.

[0041] The behavior density correction function Xk is obtained by the following formula:

[0042] ;

[0043] In the formula, nTypek represents the frequency of the event type of the k-th financial behavior within period j, Nj represents the total number of all behaviors within period j, and ln represents the logarithmic function.

[0044] Preferably, the deviation pattern recognition module includes a trajectory expectation model fitting unit and a multi-dimensional deviation evaluation calculation unit;

[0045] The trajectory expectation model fitting unit establishes the expected trajectory model based on the trajectory intensity T, and constructs the trajectory expectation intensity G by combining the scene vector set A;

[0046] By analyzing the trajectory intensity T, historical periodic trajectory data T(i,j)(t-1), ..., T(i,j)(tN) are obtained; a scenario adjustment factor θ is established using the scenario vector set A.

[0047] The situational adjustment factor θ is obtained using the following formula:

[0048] ;

[0049] The expected intensity G of the trajectory is obtained by the following formula:

[0050] ;

[0051] In the formula, G(i,j) represents the expected intensity of the trajectory of the i-th type of behavior within period j, N represents the length of the historical observation window, and T(i,j)(t) represents the intensity of the trajectory of the i-th type of behavior within period j at time t.

[0052] Preferably, the multi-dimensional deviation assessment calculation unit compares the expected trajectory intensity G with the trajectory intensity T dimension by dimension to generate a comprehensive anomaly index ΔF;

[0053] The behavior difference vector is constructed by using the expected trajectory intensity G and the trajectory intensity T to obtain the deviation magnitude PC;

[0054] The deviation magnitude PC is obtained using the following formula:

[0055] PCi=|T(i,j)-G(i,j)|;

[0056] In the formula, PCi represents the deviation magnitude of the i-th type of behavior in the current period;

[0057] The weighted deviation function ΔHS is obtained by calculating the deviation magnitude PC.

[0058] The weighted bias function ΔHS is obtained by the following formula:

[0059] ΔHS=PCi*Aa;

[0060] In the formula, Aa represents the sensitivity weight, which is obtained by the formula: Aa=1+α1*Ahx+α2*Ads+α3*Axm, where α1, α2 and α3 represent the preset weight values ​​of the contract performance stage coefficient Ahx, the counterparty's performance stability Ads and the project funding tension coefficient Axm, respectively.

[0061] The comprehensive anomaly index ΔF is obtained using the following formula:

[0062] ;

[0063] In the formula, ΔHSi represents the weighted deviation function of the i-th type of behavior, M represents the total number of behavior types, p represents the deviation amplification index, which represents the order of nonlinear amplification of the deviation of a single behavior. It is often p=2 to simulate the Euclidean norm, and q represents the comprehensive convergence index, which represents the index for convergent compression of the overall deviation value. It is often q=2 or q=3 to control the fluctuation range of the abnormal index.

[0064] Preferably, the multi-type risk automatic labeling module includes a dynamic threshold generation unit and a risk type determination unit;

[0065] The dynamic threshold generation unit generates a dynamically adjusted dynamic anomaly threshold Thf based on the distribution of the historical comprehensive anomaly index ΔF and the adaptive behavioral trajectory trend.

[0066] The trajectory trend factor is obtained by calculating the trajectory intensity T. ;

[0067] Trajectory Trend Factor Obtain it using the following formula:

[0068] ;

[0069] In the formula, T(i, j-1) represents the trajectory intensity of the i-th type of behavior within period j-1, and E represents a constant to avoid the denominator being zero;

[0070] The dynamic anomaly threshold Tef is obtained using the following formula.

[0071] ;

[0072] In the formula, μΔF represents the mean of the comprehensive anomaly index, σΔF represents the standard deviation of the comprehensive anomaly index, and η represents the trend sensitivity adjustment factor. This represents the mean of the trajectory trend factor;

[0073] The risk type determination unit compares and classifies the comprehensive anomaly index ΔF with the dynamic anomaly threshold Thf to obtain the anomaly type R, and marks the behavior type and semantic label.

[0074] Exception type R is obtained by matching in the following way:

[0075] When the comprehensive anomaly index ΔF ≤ dynamic anomaly threshold Thf*0.5, it indicates a Level I anomaly, the behavior type is reimbursement carry-forward, and the semantic label is minor outstanding account anomaly.

[0076] When the dynamic anomaly threshold Thf*0.5 < the comprehensive anomaly index ΔF ≤ the dynamic anomaly threshold Thf, it indicates a level II anomaly type, the behavior type is payment collection, and the semantic label is accounts receivable not collected.

[0077] When the dynamic anomaly threshold Thf < the comprehensive anomaly index ΔF ≤ the dynamic anomaly threshold Thf * 1.5, it indicates a Level III anomaly type, the behavior type is payment, and the semantic label is unauthorized payment;

[0078] When the dynamic anomaly threshold Thf*1.5 < the comprehensive anomaly index ΔF, it indicates a level IV anomaly type, with the behavior type being project allocation and the semantic label being budget transfer risk.

[0079] Preferably, the intelligent intervention suggestion generation module establishes an anomaly impact path chain structure based on the anomaly type R and the scene vector set A, and obtains the causal path function CLH;

[0080] The causal path function CLH is obtained through the following formula:

[0081] CLH=f(Type,Ahx,Axm);

[0082] In the formula, f() represents the mapping function, and Type represents the event type;

[0083] The intelligent intervention operation set Ψs is obtained through the causal path function CLH;

[0084] The set of intelligent intervention operations Ψs is obtained through the following formula:

[0085] ;

[0086] In the formula, The knowledge graph representing intervention strategies is organized in the form of rules, graph structure, and knowledge base;

[0087] Retrieve intervention logic blocks that match the causal path function CLH from the strategy library and generate a set of intelligent intervention operations Ψs;

[0088] The feasibility of the intelligent intervention operation set Ψs is screened and prioritized to obtain an operation instruction suggestion table, including:

[0089] Recommendation (1): Issue a supplementary invoice, contract number, recommended person in charge: system status: invoice not issued;

[0090] Recommendation (2): Delay payment approval. High risk. Payment order number. It is recommended to initiate the approval process.

[0091] Recommendation (3): Generate a budget structure revision proposal.

[0092] Preferably, the periodic trajectory normalization and adaptive learning module analyzes the feedback results after each intervention action to construct the intervention response success rate factor Gxz;

[0093] The intervention response success rate factor Gxz is obtained using the following formula:

[0094] ;

[0095] In the formula, Y represents the number of intervention recommendations, CGy represents whether the y-th recommendation was successfully implemented (0 indicates failure, 1 indicates success), and PC represents the deviation magnitude.

[0096] The comprehensive anomaly index ΔF is readjusted by the pre-response success rate factor Gxz to obtain the anomaly deviation index nΔF, and then re-introduced into the multi-type risk automatic labeling module for anomaly classification.

[0097] The abnormal deviation index nΔF is obtained by the following formula:

[0098] nΔF = (1 - β * Gxz) * ΔF;

[0099] In the formula, β represents the response adjustment coefficient.

[0100] A method for intelligent and automated accounting of financial data includes the following steps:

[0101] Step 1: The scenario factor extraction and semantic aggregation module is responsible for collecting scenario parameters that affect the compliance of financial behavior, preprocessing them, and fitting them into a scenario vector set A;

[0102] Step 2: The financial behavior trajectory reconstruction module relies on the scenario vector set A to decompose the income and expenditure operations into events, and constructs the structural trajectory of the actual behavior according to the cycle to obtain the intensity T of the behavior trajectory;

[0103] Step 3: The deviation pattern recognition module uses the trajectory intensity T as a basis and combines it with the scene vector set A to calculate the degree of deviation between the actual financial operation trajectory and the expected trajectory, and obtains the comprehensive anomaly index ΔF.

[0104] Step 4: The multi-type risk automatic labeling module obtains the dynamic anomaly threshold Thf based on the acquired comprehensive anomaly index ΔF, and performs semantic classification on the comprehensive anomaly index ΔF to obtain the anomaly type R;

[0105] Step 5: The intelligent intervention suggestion generation module generates an intelligent intervention operation set Ψs based on the anomaly type R and the scene vector set A, obtains the intervention results, and generates suggested actions to be performed.

[0106] Step Six: The periodic trajectory normalization and adaptive learning module records the trajectory intensity T and intervention results within the historical period, readjusts the comprehensive anomaly index ΔF, and obtains the anomaly deviation index nΔF.

[0107] This invention provides a method and system for intelligent and automated accounting of financial data, which has the following beneficial effects:

[0108] (1) During system operation, the intelligent intervention suggestion generation module derives operational, specific, and clearly defined intervention action suggestions Ψs based on the anomaly type R and the scenario vector A, such as reissuing invoices, suspending payments, and reviewing the budget structure, which greatly reduces the judgment burden on financial personnel. This mechanism makes up for the shortcomings of traditional systems in providing business-oriented intervention suggestions and strengthens the operational orientation and implementation capability of the system. The periodic trajectory normalization and adaptive learning module uses historical behavior trajectories and intervention response results as learning materials for the system, continuously iterates the anomaly identification mechanism and trajectory expectation model, dynamically optimizes the comprehensive anomaly index ΔF, and outputs the corrected anomaly deviation index nΔF, thereby enabling the system to have a learning closed-loop capability of "feedback → adjustment → evolution".

[0109] (2) By mapping event behaviors to natural cycles, various behaviors can be aggregated to form event sets, thereby constructing a clear timeline trajectory. This behavior flow modeling method based on cycles breaks through the limitations of traditional single-point data judgment, realizes process tracking and trend expression of behavior, and is particularly helpful in identifying typical risk patterns such as inter-period transfer, delayed payment, and advance payment.

[0110] (3) By setting up a trajectory expectation model fitting unit, a desirable trajectory model that is highly adapted to the behavior category and time period is constructed based on historical periodic trajectory data, and this model is used as the theoretical baseline for deviation judgment. The introduction of this mechanism enables the system to have trend perception capability and periodic comparison logic, effectively solving the static limitation of "judging whether it is abnormal based solely on the current performance". By introducing key elements in the scenario vector set A, including contract performance status, counterparty performance stability and project funding tension, a context adjustment factor θ is constructed and used to adjust the output level of the trajectory expectation model. The system realizes the transformation of judging whether behavior is reasonable not only by numerical value, but also by business background. This mechanism significantly enhances the contextual semantic sensitivity of anomaly identification, effectively avoids misjudgment or omission due to ignoring business status, and improves the accuracy and interpretability of judgment.

[0111] (4) By setting a dynamic threshold generation unit, the system adaptively generates dynamically adjusted anomaly judgment criteria based on the distribution characteristics of historical anomaly indices and the trend changes of behavioral trajectories. This solves the limitation that the commonly used fixed thresholds in traditional financial systems cannot adapt to business fluctuations and diverse scenarios. This mechanism avoids misjudging minor fluctuations as anomalies and can also identify potential sudden deviation behaviors in advance, significantly improving the system's adaptability and judgment accuracy to the complexity of actual business. The periodic trajectory normalization and adaptive learning module records the execution feedback of each intervention action and adjusts the original anomaly index by constructing an intervention response success rate factor to generate a corrected anomaly deviation index, which is then reintroduced into the anomaly classification process. Attached Figure Description

[0112] Figure 1 This is a flowchart illustrating the intelligent automated accounting system for financial data according to the present invention.

[0113] Figure 2 This is a schematic diagram illustrating the steps of an intelligent automated accounting method for financial data according to the present invention.

[0114] Figure 3 This is a schematic diagram of the abnormal deviation index acquisition process of the present invention;

[0115] Figure 4 A bar chart showing the dynamic anomaly threshold constructed according to the present invention. Detailed Implementation

[0116] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0117] Example 1

[0118] This invention provides an intelligent automated accounting system for financial data. Please refer to [link / reference]. Figures 1 to 4 It includes a scenario factor extraction and semantic aggregation module, a financial behavior trajectory reconstruction module, a deviation pattern recognition module, a multi-type risk automatic labeling module, an intelligent intervention suggestion generation module, and a periodic trajectory normalization and adaptive learning module;

[0119] The scenario factor extraction and semantic aggregation module is responsible for collecting scenario parameters that affect the compliance of financial behavior, preprocessing them, and fitting them into a scenario vector set A.

[0120] The financial behavior trajectory reconstruction module relies on the scenario vector set A to decompose income and expenditure operations into events and constructs the structural trajectory of the actual behavior according to the cycle to obtain the intensity T of the behavior trajectory.

[0121] The deviation pattern recognition module uses trajectory intensity T as a basis and combines it with scene vector set A to calculate the degree of deviation between the actual financial operation trajectory and the expected trajectory, and obtains the comprehensive anomaly index ΔF.

[0122] The multi-type risk automatic labeling module obtains the dynamic anomaly threshold Thf based on the acquired comprehensive anomaly index ΔF, and performs semantic classification on the comprehensive anomaly index ΔF to obtain the anomaly type R;

[0123] The intelligent intervention suggestion generation module generates an intelligent intervention operation set Ψs based on the anomaly type R and the scene vector set A, obtains the intervention results, and generates suggested actions to be performed.

[0124] The periodic trajectory normalization and adaptive learning module records the trajectory intensity T and intervention results within historical periods, readjusts the comprehensive anomaly index ΔF, and obtains the anomaly deviation index nΔF.

[0125] In this embodiment, the scenario factor extraction and semantic aggregation module structures deep business semantic parameters such as contract performance status, counterparty behavior characteristics, and project budget tension into a scenario vector set A, which is then embedded into the entire accounting process, establishing a basic channel for the linkage between financial behavior and business scenarios. Through the financial behavior trajectory reconstruction module and periodic modeling mechanism, the system no longer judges deviation behavior on a per-voucher basis, but introduces the concept of behavior trajectory intensity T to structurally reconstruct and dynamically calculate the evolution of financial behavior. This enables the system to identify anomalies from a trend perspective, achieving early identification and warning of atypical deviation behaviors such as early payment, delayed invoices, and periodic bank runs, overcoming the shortcomings of existing systems that only focus on static values.

[0126] This system does not employ the traditional, crude method of using fixed thresholds. Instead, a multi-type risk automatic labeling module calculates a dynamic anomaly threshold (Thf) based on historical behavior patterns and trajectory trends, and combines this with behavioral type to achieve semantic classification. This avoids misjudging minor fluctuations as abnormal or ignoring hidden risks, significantly improving the system's accuracy in identifying risky behaviors and its adaptability to different situations.

[0127] The intelligent intervention suggestion generation module derives operational, specific, and clearly defined intervention action suggestions Ψs based on the anomaly type R and scenario vector A, such as issuing supplementary invoices, suspending payments, and reviewing the budget structure. This significantly reduces the judgment burden on finance personnel, enabling the system to possess a complete "discovery-reasoning-response" chain. This mechanism compensates for the shortcomings of traditional systems in providing business-oriented intervention suggestions, strengthening the system's operational orientation and implementation capabilities. The periodic trajectory normalization and adaptive learning module uses historical behavioral trajectories and intervention response results as learning materials for the system, continuously iterating the anomaly identification mechanism and trajectory expectation model, dynamically optimizing the comprehensive anomaly index ΔF, and outputting the corrected nΔF, thus enabling the system to possess a learning closed-loop capability of "feedback → adjustment → evolution." This mechanism breaks the defects of rigid rules and static models in traditional systems, enabling the accounting system to continuously optimize with changes in the business environment, achieving true "intelligent evolution" of the financial system.

[0128] Example 2

[0129] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically: the scene factor extraction and semantic aggregation module includes a scene parameter extraction unit and a normalization and semantic fitting unit;

[0130] The scenario parameter extraction unit extracts scenario parameters from the data in the financial system, including the contract performance stage coefficient Ahx, the counterparty performance stability Ads, and the project funding tension coefficient Axm, and fits them into the original scenario dataset YA.

[0131] Among them, the contract performance stage coefficient Ahx is obtained by analyzing the contract performance platform. The contract performance stage coefficient Ahx is obtained by reading the ratio of the current contract performance node progress Preal to the total performance progress plan Pplan for each project.

[0132] Counterparty performance stability (Ads) is obtained by analyzing past payment records and payment cycle logs;

[0133] The project funding tension coefficient Axm is obtained by collecting data from the financial budget and actual payment execution system.

[0134] The normalization and semantic fitting unit normalizes the acquired original scene dataset YA to obtain the scene vector set A;

[0135] Scene vector set A is obtained using the following formula:

[0136] ;

[0137] In the formula, Ao represents the o-th data in the scene vector set A, YAo represents the o-th data in the original scene dataset YA, minYAo represents the valley value of the o-th data in the original scene dataset YA, and maxYAo represents the peak value of the o-th data in the original scene dataset YA.

[0138] In this embodiment, by setting up a dedicated scenario parameter extraction unit, the system can proactively extract implicit factors closely related to the business process from the contract performance platform, accounts receivable and payable data, and budget execution system, such as the contract performance stage, counterparty stability, and project funding tension. This information has long been considered "external variables" by traditional financial systems and has not been incorporated into the accounting judgment logic. The structural design of this embodiment breaks down the barriers between the financial system and the business management system, enabling the accounting system to perceive actual business scenarios and significantly improving the business rationality of its judgments.

[0139] This embodiment establishes a semantic support foundation for subsequent modules such as behavior trajectory reconstruction and deviation analysis by uniformly collecting and fitting business scenario factors from different sources and with different scales into an original scenario dataset YA. Compared with the traditional method of judging behavior based on business type labels or fixed account codes, this method portrays scenario changes in a more granular and realistic way, which helps subsequent modules build more sensitive and personalized trajectory judgment models, fundamentally enhancing the system's ability to identify and interpret complex financial behaviors.

[0140] The system's normalization and semantic fitting unit normalizes the original scene dataset and transforms it into a scene vector set A with a unified structure. This effectively solves the problems of inconsistent dimensions and scales of data from different sources in traditional systems. This not only ensures the comparability and fusion of different parameters in the calculation of behavioral trajectories and anomaly indices, but also enhances the system's adaptability to cross-project and cross-cycle business differences. It provides a high-quality, structurally standardized feature data source for subsequent model inputs, avoiding the "semantic idling" problem of previous model inference.

[0141] Example 3

[0142] This embodiment is an explanation based on Embodiment 2. Please refer to it. Figure 1 Specifically: the financial behavior trajectory reconstruction module includes an income and expenditure event extraction unit and a behavior trajectory intensity construction unit;

[0143] The revenue and expenditure event extraction unit decomposes the revenue and expenditure operations in the original financial behavior into event-based decomposition, transforms them into structured event tuples e, and maps and groups the behavioral events according to the natural cycle based on the accounting period to obtain the event cycle behavior Ej.

[0144] The structured event tuple e is obtained using the following formula:

[0145] ek={Typek,stk,etk,mk,ϕk};

[0146] In the formula, ek represents the k-th financial action in the structured event tuple, Typek represents the event type of the k-th financial action, stk represents the start time of the k-th financial action, etk represents the end time of the k-th financial action, mk represents the amount involved in the k-th financial action, and ϕk represents the event context label of the k-th financial action.

[0147] The event lifecycle behavior Ej is obtained using the following formula:

[0148] Ej = {e1, e2, ..., ek | etk ∈ period j};

[0149] Based on the contract performance stage coefficient Ahx and project funding tension coefficient Axm in the scenario vector set A, a time-series weighting function ωj is constructed;

[0150] The time-series weighting function ωj is obtained by summing the contract performance stage coefficient Ahx and the project funding tension coefficient Axm and then dividing by 2.

[0151] The behavior trajectory intensity construction unit calculates the contribution intensity of all event-period behaviors Ej in each cycle to the trajectory surface and obtains the trajectory intensity T.

[0152] The trajectory intensity T is obtained using the following formula:

[0153] ;

[0154] In the formula, T(i,j) represents the trajectory intensity of the i-th type of behavior within period j, Eji represents the event set of the i-th type of behavior within period j, Δtk represents the behavior time span, which is obtained by the difference between the start time stk and the end time etk of the k-th financial behavior, and Xk represents the behavior density correction function.

[0155] The behavior density correction function Xk is obtained by the following formula:

[0156] ;

[0157] In the formula, nTypek represents the frequency of the event type of the k-th financial behavior within period j, Nj represents the total number of all behaviors within period j, and ln represents the logarithmic function.

[0158] In this embodiment, by setting up an income and expenditure event extraction unit, the original financial behaviors, including receipts, payments, transfers, and invoice registrations, are uniformly abstracted into structured event tuples. The system no longer uses voucher entries as the basic unit of analysis, but processes them as complete event units with time, amount, behavior type, and contextual labels. This approach avoids the problem of losing semantic dimensions during behavior decomposition in traditional systems, achieving a unified three-dimensional modeling of financial behaviors across "behavioral dimension, temporal dimension, and contextual dimension," greatly improving the analyzability of the data structure.

[0159] This module maps event behaviors to natural cycles, allowing various behaviors to aggregate into event sets and thus construct a clear timeline trajectory. This behavior flow modeling approach, based on cycles, breaks through the limitations of traditional systems' single-point data judgment, enabling process tracking and trend expression of behaviors. It is particularly helpful in identifying typical risk patterns such as inter-period transfers, delayed payments, and advance payments.

[0160] By introducing a time-series weighting factor based on the contract performance stage and project funding constraints during the trajectory construction process, the same behavior is assigned different judgment intensities under different scenario conditions. For example, when contract performance is incomplete or the project is under financial strain, the system automatically increases the trajectory intensity calculation for a certain payment behavior, thereby more sensitively capturing the "time-specific irrationality" of the behavior. This makes trajectory modeling not only reflect the behavior itself but also the rationality evaluation of the environment in which the behavior occurs, forming a dynamic behavior recognition mechanism that is highly aligned with business logic.

[0161] By logarithmizing the frequency of behavior types within a period, the system can identify whether behaviors are excessively concentrated or occur frequently within a certain period, thereby identifying high-density behavioral anomalies that do not conform to normal patterns, such as "frequent collections" and "multiple budget transfers in a short period of time." This mechanism enhances the system's ability to perceive structural and periodic behavioral anomalies, and is particularly suitable for detecting behavioral changes in complex projects or under centralized group funding models.

[0162] Example 4

[0163] This embodiment is an explanation based on Embodiment 3. Please refer to it. Figure 1 and Figure 3 Specifically: the deviation pattern recognition module includes a trajectory expectation model fitting unit and a multi-dimensional deviation evaluation calculation unit;

[0164] The trajectory expectation model fitting unit establishes the expected trajectory model based on the trajectory intensity T, and constructs the trajectory expectation intensity G by combining the scene vector set A;

[0165] By analyzing the trajectory intensity T, historical periodic trajectory data T(i,j)(t-1), ..., T(i,j)(tN) are obtained; a scenario adjustment factor θ is established using the scenario vector set A.

[0166] The situational adjustment factor θ is obtained using the following formula:

[0167] ;

[0168] The expected intensity G of the trajectory is obtained by the following formula:

[0169] ;

[0170] In the formula, G(i,j) represents the expected intensity of the trajectory of the i-th type of behavior within period j, N represents the length of the historical observation window, and T(i,j)(t) represents the intensity of the trajectory of the i-th type of behavior within period j at time t.

[0171] The multi-dimensional deviation assessment calculation unit compares the expected trajectory intensity G with the trajectory intensity T dimension by dimension to generate a comprehensive anomaly index ΔF.

[0172] The behavior difference vector is constructed by using the expected trajectory intensity G and the trajectory intensity T to obtain the deviation magnitude PC;

[0173] The deviation magnitude PC is obtained using the following formula:

[0174] PCi=|T(i,j)-G(i,j)|;

[0175] In the formula, PCi represents the deviation magnitude of the i-th type of behavior in the current period;

[0176] The weighted deviation function ΔHS is obtained by calculating the deviation magnitude PC.

[0177] The weighted bias function ΔHS is obtained by the following formula:

[0178] ΔHS=PCi*Aa;

[0179] In the formula, Aa represents the sensitivity weight, which is obtained by the formula: Aa=1+α1*Ahx+α2*Ads+α3*Axm, where α1, α2 and α3 represent the preset weight values ​​of the contract performance stage coefficient Ahx, the counterparty's performance stability Ads and the project funding tension coefficient Axm, respectively.

[0180] The comprehensive anomaly index ΔF is obtained using the following formula:

[0181] ;

[0182] In the formula, ΔHSi represents the weighted deviation function of the i-th type of behavior, M represents the total number of behavior types, p represents the deviation amplification index, and q represents the comprehensive convergence index.

[0183] This embodiment constructs a desired trajectory model fitting unit based on historical periodic trajectory data to build a trajectory model highly adapted to the behavior category and time period, which serves as the theoretical baseline for deviation judgment. This mechanism enables the system to possess trend perception capabilities and periodic comparison logic, effectively overcoming the static limitation of judging anomalies solely based on current performance. By introducing key elements from the scenario vector set A, including contract performance status, counterparty performance stability, and project financial tension, a context adjustment factor θ is constructed and used to adjust the output level of the trajectory expectation model. This allows the system to consider not only numerical values ​​but also the business context when determining the rationality of behavior. This mechanism significantly enhances the contextual semantic sensitivity of anomaly identification, effectively avoiding misjudgments or omissions due to neglecting business status, and improving the accuracy and interpretability of the judgment.

[0184] By comparing the actual behavioral trajectory with the expected trajectory dimension by dimension, a structured deviation magnitude vector is formed. Based on this, a weighted deviation function is generated using preset sensitivity weights, enabling differentiated expression of deviation sensitivity for different behavioral categories in different scenarios. This design can not only identify local anomalies of a certain type of behavior, such as advance payments, but also identify systemic structural anomalies when multiple behavioral categories deviate collaboratively, such as budget imbalances and cash flow mismatches. By introducing two adjustable parameters, the deviation amplification index and the comprehensive convergence index, the weighted deviation results are comprehensively integrated. The system can configure sensitivity according to different enterprise scenarios or risk preferences, achieving flexible and strategic control of anomaly judgment. This is significantly superior to the traditional "one-size-fits-all" anomaly threshold mechanism, making the system scalable, portable, and configurable, providing a technical foundation for accurate verification under various business models.

[0185] The final output comprehensive anomaly index ΔF is a multi-source integrated index generated through behavioral modeling, context weight adjustment, and deviation fusion calculation. It not only provides quantitative support for anomaly level judgment but also provides a unified decision-making basis for subsequent modules such as risk labeling, intervention suggestions, and learning feedback. Compared with the traditional rule-based classification method, the introduction of this module represents a key upgrade for the financial system from a "result-triggered verification mechanism" to a "behavioral modeling-driven intelligent identification system."

[0186] Example 5

[0187] This embodiment is an explanation based on Embodiment 4. Please refer to it. Figure 1 and Figure 4 Specifically: the multi-type risk automatic labeling module includes a dynamic threshold generation unit and a risk type determination unit;

[0188] The dynamic threshold generation unit generates a dynamically adjusted dynamic anomaly threshold Thf based on the distribution of the historical comprehensive anomaly index ΔF and the adaptive behavioral trajectory trend.

[0189] The trajectory trend factor is obtained by calculating the trajectory intensity T. ;

[0190] Trajectory Trend Factor Obtain it using the following formula:

[0191] ;

[0192] In the formula, T(i, j-1) represents the trajectory intensity of the i-th type of behavior within period j-1, and E represents a constant;

[0193] The dynamic anomaly threshold Tef is obtained using the following formula.

[0194] ;

[0195] In the formula, μΔF represents the mean of the comprehensive anomaly index, σΔF represents the standard deviation of the comprehensive anomaly index, and η represents the trend sensitivity adjustment factor. This represents the mean of the trajectory trend factor;

[0196] Specific examples:

[0197] Table 1: Dynamic Anomaly Threshold Calculation Table;

[0198]

[0199] The risk type determination unit compares and classifies the comprehensive anomaly index ΔF with the dynamic anomaly threshold Thf to obtain the anomaly type R, and marks the behavior type and semantic label.

[0200] Exception type R is obtained by matching in the following way:

[0201] When the comprehensive anomaly index ΔF ≤ dynamic anomaly threshold Thf*0.5, it indicates a Level I anomaly, the behavior type is reimbursement carry-forward, and the semantic label is minor outstanding account anomaly.

[0202] When the dynamic anomaly threshold Thf*0.5 < the comprehensive anomaly index ΔF ≤ the dynamic anomaly threshold Thf, it indicates a level II anomaly type, the behavior type is payment collection, and the semantic label is accounts receivable not collected.

[0203] When the dynamic anomaly threshold Thf < the comprehensive anomaly index ΔF ≤ the dynamic anomaly threshold Thf * 1.5, it indicates a Level III anomaly type, the behavior type is payment, and the semantic label is unauthorized payment;

[0204] When the dynamic anomaly threshold Thf*1.5 < the comprehensive anomaly index ΔF, it indicates a level IV anomaly type, with the behavior type being project allocation and the semantic label being budget transfer risk.

[0205] The intelligent intervention suggestion generation module establishes an anomaly impact path chain structure based on the anomaly type R and the scene vector set A, and obtains the causal path function CLH.

[0206] The causal path function CLH is obtained through the following formula:

[0207] CLH=f(Type,Ahx,Axm);

[0208] In the formula, f() represents the mapping function, and Type represents the event type;

[0209] The intelligent intervention operation set Ψs is obtained through the causal path function CLH;

[0210] The set of intelligent intervention operations Ψs is obtained through the following formula:

[0211] ;

[0212] In the formula, Represents a knowledge graph of intervention strategies;

[0213] Retrieve intervention logic blocks that match the causal path function CLH from the strategy library and generate a set of intelligent intervention operations Ψs;

[0214] The feasibility of the intelligent intervention operation set Ψs is screened and prioritized to obtain an operation instruction suggestion table, including:

[0215] Recommendation (1): Issue a replacement invoice;

[0216] Recommendation (2): Delay payment approval;

[0217] Recommendation (3): Generate a budget structure revision proposal.

[0218] The periodic trajectory normalization and adaptive learning module analyzes the feedback results after each intervention action and constructs the intervention response success rate factor Gxz;

[0219] The intervention response success rate factor Gxz is obtained using the following formula:

[0220] ;

[0221] In the formula, Y represents the number of intervention recommendations, CGy represents whether the y-th recommendation was successfully implemented, and PC represents the deviation magnitude;

[0222] The comprehensive anomaly index ΔF is readjusted by the pre-response success rate factor Gxz to obtain the anomaly deviation index nΔF, and then re-introduced into the multi-type risk automatic labeling module for anomaly classification.

[0223] The abnormal deviation index nΔF is obtained by the following formula:

[0224] nΔF = (1 - β * Gxz) * ΔF;

[0225] In the formula, β represents the response adjustment coefficient.

[0226] In this embodiment, by setting a dynamic threshold generation unit, the system adaptively generates dynamically adjusted anomaly judgment criteria based on the distribution characteristics of historical anomaly indices and the trend changes of behavioral trajectories. This solves the limitation of traditional financial systems where fixed thresholds cannot adapt to business fluctuations and diverse scenarios. This mechanism avoids misjudging minor fluctuations as anomalies while also identifying potential sudden deviations in advance, significantly improving the system's adaptability and judgment accuracy to the complexity of actual business operations.

[0227] This embodiment, after comparing the anomaly index with a dynamic threshold, sets up multiple levels of anomaly type judgment intervals and binds them to specific behavior types and semantic tags, achieving dual recognition from "degree of anomaly" to "nature of anomaly." This mechanism not only improves the granularity and interpretability of the system's anomaly recognition but also provides a clear classification basis for subsequent intervention suggestions, avoiding the fuzzy judgment phenomenon of "identifying anomalies but lacking semantic support" in general systems. The system not only identifies and classifies anomalies but also constructs a causal path function (CLH) based on key indicators in the scenario vector, such as contract status and financial tension, and retrieves corresponding response actions from the intervention strategy knowledge graph accordingly. This mechanism establishes a logical chain between anomaly recognition and intelligent response, enabling the system to have a closed-loop response capability with traceable anomaly causes and predictable intervention suggestions, greatly improving the automation and intelligence level of business processing.

[0228] Once the set of intervention operations is formed, the system does not recommend all operations indiscriminately. Instead, it scores feasibility and prioritizes them based on factors such as scenario adaptability, execution cost, and historical effectiveness, ultimately outputting a list of suggested operation instructions. This structured and hierarchical approach effectively alleviates the decision-making pressure on finance personnel while improving the adoption rate and actual implementation effect of the suggestions, thus addressing the problem of traditional systems' vague output of "suggestions without logical hierarchy and operational guidance."

[0229] The periodic trajectory normalization and adaptive learning module records the execution feedback of each intervention action and adjusts the original anomaly index by constructing an intervention response success rate factor, generating a "corrected anomaly deviation index," which is then reintroduced into the anomaly classification process. This mechanism achieves a complete closed loop of data-identification-intervention-feedback-correction, enabling the system to continuously optimize judgment thresholds and model outputs based on historical processing results. It possesses dynamic learning and long-term evolution capabilities, marking the system's transformation and upgrade from a "tool-based financial assistant" to an "autonomous intelligent accounting system."

[0230] Example 6

[0231] A method for intelligent and automated accounting of financial data, please refer to... Figure 2 Specifically, it includes the following steps:

[0232] Step 1: The scenario factor extraction and semantic aggregation module is responsible for collecting scenario parameters that affect the compliance of financial behavior, preprocessing them, and fitting them into a scenario vector set A;

[0233] Step 2: The financial behavior trajectory reconstruction module relies on the scenario vector set A to decompose the income and expenditure operations into events, and constructs the structural trajectory of the actual behavior according to the cycle to obtain the intensity T of the behavior trajectory;

[0234] Step 3: The deviation pattern recognition module uses the trajectory intensity T as a basis and combines it with the scene vector set A to calculate the degree of deviation between the actual financial operation trajectory and the expected trajectory, and obtains the comprehensive anomaly index ΔF.

[0235] Step 4: The multi-type risk automatic labeling module obtains the dynamic anomaly threshold Thf based on the acquired comprehensive anomaly index ΔF, and performs semantic classification on the comprehensive anomaly index ΔF to obtain the anomaly type R;

[0236] Step 5: The intelligent intervention suggestion generation module generates an intelligent intervention operation set Ψs based on the anomaly type R and the scene vector set A, obtains the intervention results, and generates suggested actions to be performed.

[0237] Step Six: The periodic trajectory normalization and adaptive learning module records the trajectory intensity T and intervention results within the historical period, readjusts the comprehensive anomaly index ΔF, and obtains the anomaly deviation index nΔF.

[0238] The method of this invention, in its first step, structures scenario factors such as contract performance progress, counterparty performance behavior, and project funding tensions into vector model inputs through "scenario factor extraction and semantic aggregation." This provides deeply embedded business logic support for subsequent trajectory modeling and anomaly identification, overcoming the blind spots in the ability to understand context in previous accounting processes. Through the second step of financial behavior trajectory reconstruction, the system, for the first time, events-basedizes income and expenditure behaviors and maps them to natural cycles, constructing trajectory intensity indicators according to behavior category, time span, and frequency, truly reflecting the dynamic rhythm and structural characteristics of financial operations. This approach overcomes the problem of traditional systems that "only look at single-point values ​​of financial behavior without considering the continuity of behavior," providing a quantifiable and traceable behavioral basis for deviation assessment and trend judgment.

[0239] In the third step, the method calculates the intensity of the deviation between actual behavior and the expected trajectory to form a comprehensive anomaly index ΔF. This index is then combined with scenario factors to set adjustment factors, differentiating behavioral deviations across different scenarios. Compared to the traditional "one-size-fits-all" judgment logic, in the fourth step, the system automatically classifies abnormal behavior into multiple levels of risk based on the relationship between ΔF and dynamic anomaly thresholds, matching corresponding behavioral categories and semantic labels to form a refined classification result. This not only enhances the financial audit's ability to quickly understand the nature of anomalies but also lays a hierarchical foundation for generating subsequent intervention recommendations, overcoming the previous fuzzy identification phenomenon of "only identifying anomalies but not knowing their categories."

[0240] In the fifth step, a causal path function is constructed based on the anomaly type and scenario vector, and matching intervention strategies are retrieved to generate a set of executable suggestions. This not only achieves a complete closed-loop response path from anomaly detection to handling suggestions but also provides operational suggestions such as reissue of invoices and budget revisions, effectively alleviating the burden of manual verification and decision-making, and demonstrating practical feasibility and automated processing capabilities. In the sixth step, the system records the behavioral trajectory and intervention feedback for each cycle, evaluates the intervention success rate, and adjusts ΔF accordingly to generate a corrected anomaly index nΔF, which is then re-entered into the risk labeling module for cyclical updates. This mechanism signifies that the system has upgraded from a one-time calculation system to a dynamic learning system, possessing a long-term optimization judgment mechanism and the ability to adapt to changes in multiple scenarios, truly forming an accounting model with intelligent evolutionary capabilities.

[0241] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A financial data intelligent automated accounting system, characterized in that: It includes a scenario factor extraction and semantic aggregation module, a financial behavior trajectory reconstruction module, a deviation pattern recognition module, a multi-type risk automatic labeling module, an intelligent intervention suggestion generation module, and a periodic trajectory normalization and adaptive learning module; The scenario factor extraction and semantic aggregation module is responsible for collecting scenario parameters that affect the compliance of financial behavior, preprocessing them, and fitting them into a scenario vector set A. The scene factor extraction and semantic aggregation module includes a scene parameter extraction unit and a normalization and semantic fitting unit; The scenario parameter extraction unit extracts scenario parameters from the data in the financial system, including the contract performance stage coefficient Ahx, the counterparty performance stability Ads, and the project funding tension coefficient Axm, and fits them into the original scenario dataset YA. Among them, the contract performance stage coefficient Ahx is obtained by analyzing the contract performance platform. The contract performance stage coefficient Ahx is obtained by reading the ratio of the current contract performance node progress Preal to the total performance progress plan Pplan for each project. Counterparty performance stability (Ads) is collected by analyzing past payment records and payment cycle logs; The project funding tension coefficient Axm is obtained by collecting data from the financial budget and actual payment execution system. The normalization and semantic fitting unit normalizes the acquired original scene dataset YA to obtain the scene vector set A; Scene vector set A is obtained using the following formula: ； In the formula, Ao represents the o-th data in the scene vector set A, YAo represents the o-th data in the original scene dataset YA, minYAo represents the valley value of the o-th data in the original scene dataset YA, and maxYAo represents the peak value of the o-th data in the original scene dataset YA. The financial behavior trajectory reconstruction module relies on the scenario vector set A to decompose income and expenditure operations into events and constructs the structural trajectory of the actual behavior according to the cycle to obtain the intensity T of the behavior trajectory. The deviation pattern recognition module uses trajectory intensity T as a basis and combines it with scene vector set A to calculate the degree of deviation between the actual financial operation trajectory and the expected trajectory, and obtains the comprehensive anomaly index ΔF. The multi-type risk automatic labeling module obtains the dynamic anomaly threshold Thf based on the acquired comprehensive anomaly index ΔF, and performs semantic classification on the comprehensive anomaly index ΔF to obtain the anomaly type R; The intelligent intervention suggestion generation module generates an intelligent intervention operation set Ψs based on the anomaly type R and the scene vector set A, obtains the intervention results, and generates suggested actions to be executed. The periodic trajectory normalization and adaptive learning module records the trajectory intensity T and intervention results within historical periods, readjusts the comprehensive anomaly index ΔF, and obtains the anomaly deviation index nΔF.

2. The intelligent automated accounting system for financial data according to claim 1, characterized in that: The financial behavior trajectory reconstruction module includes an income and expenditure event extraction unit and a behavior trajectory intensity construction unit; The revenue and expenditure event extraction unit decomposes the revenue and expenditure operations in the original financial behavior into event-based decomposition, transforms them into structured event tuples e, and maps and groups the behavioral events according to the natural cycle based on the accounting period to obtain the event cycle behavior Ej. The structured event tuple e is obtained using the following formula: ek={Typek,stk,etk,mk,ϕk}; In the formula, ek represents the k-th financial action in the structured event tuple, Typek represents the event type of the k-th financial action, stk represents the start time of the k-th financial action, etk represents the end time of the k-th financial action, mk represents the amount involved in the k-th financial action, and ϕk represents the event context label of the k-th financial action. The event lifecycle behavior Ej is obtained using the following formula: Ej = {e1, e2, ..., ek | etk ∈ period j}; Based on the contract performance stage coefficient Ahx and project funding tension coefficient Axm in the scenario vector set A, a time-series weighting function ωj is constructed; The time-series weighting function ωj is obtained by summing the contract performance stage coefficient Ahx and the project funding tension coefficient Axm and then dividing by 2.

3. The intelligent automated accounting system for financial data according to claim 2, characterized in that: The behavior trajectory intensity construction unit calculates the contribution intensity of all event-period behaviors Ej in each cycle to the trajectory surface and obtains the trajectory intensity T. The trajectory intensity T is obtained using the following formula: ； In the formula, T(i,j) represents the trajectory intensity of the i-th type of behavior within period j, Eji represents the event set of the i-th type of behavior within period j, Δtk represents the behavior time span, which is obtained by the difference between the start time stk and the end time etk of the k-th financial behavior, and Xk represents the behavior density correction function. The behavior density correction function Xk is obtained through the following formula: ； In the formula, nTypek represents the frequency of the event type of the k-th financial behavior within period j, Nj represents the total number of all behaviors within period j, and ln represents the logarithmic function.

4. The intelligent automated accounting system for financial data according to claim 3, characterized in that: The deviation pattern recognition module includes a trajectory expectation model fitting unit and a multi-dimensional deviation evaluation calculation unit; The trajectory expectation model fitting unit establishes the expected trajectory model based on the trajectory intensity T, and constructs the trajectory expectation intensity G by combining the scene vector set A; By analyzing the trajectory intensity T, historical periodic trajectory data T(i,j)(t-1), ..., T(i,j)(tN) are obtained; a scenario adjustment factor θ is established using the scenario vector set A. The situational adjustment factor θ is obtained using the following formula: ； The expected intensity G of the trajectory is obtained by the following formula: ； In the formula, G(i,j) represents the expected intensity of the trajectory of the i-th type of behavior within period j, N represents the length of the historical observation window, and T(i,j)(t) represents the intensity of the trajectory of the i-th type of behavior within period j at time t.

5. The intelligent automated accounting system for financial data according to claim 4, characterized in that: The multi-dimensional deviation assessment calculation unit compares the expected trajectory intensity G with the trajectory intensity T dimension by dimension to generate a comprehensive anomaly index ΔF. The behavior difference vector is constructed by using the expected trajectory intensity G and the trajectory intensity T to obtain the deviation magnitude PC; The deviation magnitude PC is obtained using the following formula: PCi=|T(i,j)-G(i,j)|; In the formula, PCi represents the deviation magnitude of the i-th type of behavior in the current period; The weighted deviation function ΔHS is obtained by calculating the deviation magnitude PC. The weighted bias function ΔHS is obtained by the following formula: ΔHS=PCi*Aa; In the formula, Aa represents the sensitivity weight, which is obtained by the formula: Aa=1+α1*Ahx+α2*Ads+α3*Axm, where α1, α2 and α3 represent the preset weight values ​​of the contract performance stage coefficient Ahx, the counterparty's performance stability Ads and the project funding tension coefficient Axm, respectively. The comprehensive anomaly index ΔF is obtained using the following formula: ； In the formula, ΔHSi represents the weighted deviation function of the i-th type of behavior, M represents the total number of behavior types, p represents the deviation amplification index, and q represents the comprehensive convergence index.

6. The intelligent automated accounting system for financial data according to claim 5, characterized in that: The multi-type risk automatic labeling module includes a dynamic threshold generation unit and a risk type determination unit; The dynamic threshold generation unit generates a dynamically adjusted dynamic anomaly threshold Thf based on the distribution of the historical comprehensive anomaly index ΔF and the adaptive behavioral trajectory trend. The trajectory trend factor is obtained by calculating the trajectory intensity T. ; Trajectory Trend Factor Obtain it using the following formula: ； In the formula, T(i, j-1) represents the trajectory intensity of the i-th type of behavior within period j-1, and E represents a constant; The dynamic anomaly threshold Tef is obtained using the following formula. ； In the formula, μΔF represents the mean of the comprehensive anomaly index, σΔF represents the standard deviation of the comprehensive anomaly index, and η represents the trend sensitivity adjustment factor. This represents the mean of the trajectory trend factor; The risk type determination unit compares and classifies the comprehensive anomaly index ΔF with the dynamic anomaly threshold Thf to obtain the anomaly type R, and marks the behavior type and semantic label. Exception type R is obtained by matching in the following way: When the comprehensive anomaly index ΔF ≤ dynamic anomaly threshold Thf*0.5, it indicates a Level I anomaly, the behavior type is reimbursement carry-forward, and the semantic label is minor outstanding account anomaly. When the dynamic anomaly threshold Thf*0.5 < the comprehensive anomaly index ΔF ≤ the dynamic anomaly threshold Thf, it indicates a level II anomaly type, the behavior type is payment collection, and the semantic label is accounts receivable not collected. When the dynamic anomaly threshold Thf < the comprehensive anomaly index ΔF ≤ the dynamic anomaly threshold Thf * 1.5, it indicates a Level III anomaly type, the behavior type is payment, and the semantic label is unauthorized payment; When the dynamic anomaly threshold Thf*1.5 < the comprehensive anomaly index ΔF, it indicates a level IV anomaly type, with the behavior type being project allocation and the semantic label being budget transfer risk.

7. The intelligent automated accounting system for financial data according to claim 6, characterized in that: The intelligent intervention suggestion generation module establishes an anomaly impact path chain structure based on the anomaly type R and the scene vector set A, and obtains the causal path function CLH. The causal path function CLH is obtained through the following formula: CLH=f(Type,Ahx,Axm); In the formula, f() represents the mapping function, and Type represents the event type; The intelligent intervention operation set Ψs is obtained through the causal path function CLH; The set of intelligent intervention operations Ψs is obtained through the following formula: ； In the formula, Represents a knowledge graph of intervention strategies; Retrieve intervention logic blocks that match the causal path function CLH from the strategy library and generate a set of intelligent intervention operations Ψs; The feasibility of the intelligent intervention operation set Ψs is screened and prioritized to obtain an operation instruction suggestion table, including: Recommendation (1): Issue a replacement invoice; Recommendation (2): Delay payment approval; Recommendation (3): Generate a budget structure revision proposal.

8. The intelligent automated accounting system for financial data according to claim 7, characterized in that: The periodic trajectory normalization and adaptive learning module analyzes the feedback results after each intervention action and constructs the intervention response success rate factor Gxz; The intervention response success rate factor Gxz is obtained using the following formula: ； In the formula, Y represents the number of intervention recommendations, CGy represents whether the y-th recommendation was successfully implemented, and PC represents the deviation magnitude; The comprehensive anomaly index ΔF is readjusted by the pre-response success rate factor Gxz to obtain the anomaly deviation index nΔF, and then re-introduced into the multi-type risk automatic labeling module for anomaly classification. The abnormal deviation index nΔF is obtained by the following formula: nΔF = (1 - β * Gxz) * ΔF; In the formula, β represents the response adjustment coefficient.

9. A method for intelligent and automated accounting of financial data, applied to the intelligent and automated accounting system for financial data as described in any one of claims 1 to 8, characterized in that: Includes the following steps: Step 1: The scenario factor extraction and semantic aggregation module is responsible for collecting scenario parameters that affect the compliance of financial behavior, preprocessing them, and fitting them into a scenario vector set A; Step 2: The financial behavior trajectory reconstruction module relies on the scenario vector set A to decompose the income and expenditure operations into events, and constructs the structural trajectory of the actual behavior according to the cycle to obtain the intensity T of the behavior trajectory; Step 3: The deviation pattern recognition module uses the trajectory intensity T as a basis and combines it with the scene vector set A to calculate the degree of deviation between the actual financial operation trajectory and the expected trajectory, and obtains the comprehensive anomaly index ΔF. Step 4: The multi-type risk automatic labeling module obtains the dynamic anomaly threshold Thf based on the acquired comprehensive anomaly index ΔF, and performs semantic classification on the comprehensive anomaly index ΔF to obtain the anomaly type R; Step 5: The intelligent intervention suggestion generation module generates an intelligent intervention operation set Ψs based on the anomaly type R and the scene vector set A, obtains the intervention results, and generates suggested actions to be performed. Step Six: The periodic trajectory normalization and adaptive learning module records the trajectory intensity T and intervention results within the historical period, readjusts the comprehensive anomaly index ΔF, and obtains the anomaly deviation index nΔF.

Images