Financial business data risk early warning management method and system based on artificial intelligence

Abstract

The application belongs to the field of data security, and discloses a financial business data risk early warning management method and system based on artificial intelligence, which comprises the following steps: step 1, obtaining a time sequence of actual execution amount, and decomposing the time sequence of actual execution amount to obtain an abnormal score of a budget item; step 2, obtaining an abnormal confidence score of the budget item based on the abnormal score, and judging whether the budget item is an abnormal budget item based on the abnormal confidence score; step 3, constructing a directed graph based on business event data and the abnormal budget item; step 4, obtaining all paths satisfying path constraints with the abnormal budget node as the terminal point in the directed graph to obtain a path set, and calculating the path score of each path; and step 5, calculating the priority score of each path based on the path score, obtaining a main path based on the priority score, and generating a response suggestion based on the main path. The application improves the explanation ability and governance ability of the financial risk early warning system in a complex business environment.

Description

Technical Field

[0001] This invention relates to the field of data security, and in particular to a method and system for risk warning and management of financial business data based on artificial intelligence. Background Technology

[0002] In the financial management systems of large enterprises, multiple business modules such as budgeting, funds, assets, taxation, shared service centers, and infrastructure intertwine to form a complex financial business network. With the advancement of digital transformation, enterprise financial systems are gradually acquiring characteristics of real-time data and online business processes. However, this also brings greater management challenges in data consistency, anomaly identification, and risk linkage judgment. Taking budget management as an example, budget deviations occur frequently. However, because the execution rhythm is often affected by factors such as seasonal procurement and centralized project settlement, traditional judgment methods based on static thresholds often result in false alarms or omissions. Furthermore, the root cause of some budget deviations may not lie in the budgeting process itself, but may be caused by behavioral chains from other modules such as delayed processing of shared documents, abnormal asset recording, or contract execution delays. However, current mainstream risk warning systems mainly rely on indicator threshold comparisons, lacking deep modeling of causal relationships between business logics and the ability to trace the causes of abnormal events and assign responsibility. This makes it difficult for warning systems to provide accurate and interpretable judgment results in complex scenarios. In addition, most current systems still primarily provide one-way alerts, failing to achieve closed-loop feedback and management tracking of warning results, thus reducing the system's execution capability and management value. Therefore, how to build a comprehensive risk warning management system that is oriented towards financial business data and has intelligent recognition, causal reasoning and response capabilities is a key issue that urgently needs to be addressed in the field of smart finance. Summary of the Invention

[0003] The purpose of this invention is to disclose a method and system for risk warning management of financial business data based on artificial intelligence, and to solve the technical problems pointed out in the background art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] On the one hand, this invention provides a method for risk warning and management of financial business data based on artificial intelligence, including:

[0006] Step 1: Obtain the time series of the actual execution amount, and decompose the time series of the actual execution amount to obtain the anomaly score of the budget item;

[0007] Step 2: Obtain the anomaly confidence score for the budget item based on the anomaly score, and determine whether the budget item is an abnormal budget item based on the anomaly confidence score;

[0008] Step 3: Construct a directed graph based on business event data and abnormal budget items;

[0009] Step 4: Obtain all paths in the directed graph that satisfy the path constraints and terminate at the node with budget anomaly, thus obtaining a path set; calculate the path score for each path.

[0010] Step 5: Calculate the priority score for each path based on the path score, obtain the main path based on the priority score, and generate response suggestions based on the main path.

[0011] Preferably, obtaining the time series of budget items includes:

[0012] Obtain the company's budget execution data, which includes time, budget items, actual execution amount, and corresponding budget amount;

[0013] Convert the actual execution amount in the budget execution data into a time series to obtain the time series of the actual execution amount.

[0014] Preferably, the time series of the actual executed amount is decomposed to obtain anomaly scores for budget items, including:

[0015] The time series of actual execution amounts is decomposed into trend-seasonality-residual to obtain the long-term trend, periodic fluctuations and residual terms of budget execution.

[0016] Calculate the outlier score for the budget item based on the residual item.

[0017] Preferably, obtaining the anomaly confidence score for the budget item based on the anomaly score includes:

[0018] Calculate the median of historical rhythms based on anomaly scores;

[0019] Calculate the rhythm deviation of budget items based on the historical rhythm median;

[0020] Calculate the outlier confidence score for the budget item based on the rhythm deviation of the budget item.

[0021] Preferably, determining whether a budget item is an abnormal budget item based on anomaly confidence score includes:

[0022] Determine whether the anomaly confidence score is greater than the set anomaly confidence score threshold. If yes, the budget item is considered an anomaly; otherwise, the budget item is not considered an anomaly.

[0023] Preferably, step 3 includes:

[0024] Each event in the business event data is encoded as a node in a directed graph;

[0025] Connect the nodes in the directed graph according to the connection rules to obtain the edges in the directed graph;

[0026] Calculate the weight of each edge in the directed graph.

[0027] Preferably, all paths in the directed graph that satisfy the path constraints and terminate at the node with budget anomaly are obtained, resulting in a path set including:

[0028] In a directed graph, search for all paths that satisfy the business constraints to the reachable upstream event nodes in the directed graph, and store the obtained paths into a path set.

[0029] Preferably, obtaining the main path based on priority scoring includes:

[0030] The path with the highest priority score among all paths in the path set is selected as the primary path.

[0031] Preferably, generating response suggestions based on the main path includes:

[0032] The system populates a pre-defined response suggestion template with metadata corresponding to the main path and generates a response suggestion.

[0033] On the other hand, the present invention provides a financial business data risk early warning management system based on artificial intelligence, including a first acquisition module, a second acquisition module, a construction module, a third acquisition module and a generation module;

[0034] The first acquisition module is used to acquire the time series of the actual execution amount, and to decompose the time series of the actual execution amount to obtain the abnormal scores of the budget items;

[0035] The second acquisition module is used to obtain the anomaly confidence score of the budget item based on the anomaly score, and to determine whether the budget item is an abnormal budget item based on the anomaly confidence score.

[0036] The building module is used to construct a directed graph based on business event data and exception budget items;

[0037] The third acquisition module is used to acquire all paths in the directed graph that satisfy the path constraints and end at the node with budget anomaly, and obtain a path set; and calculate the path score for each path.

[0038] The generation module is used to calculate the priority score for each path based on the path score, obtain the main path based on the priority score, and generate response suggestions based on the main path.

[0039] Beneficial effects:

[0040] This invention addresses the common problems in current financial risk early warning systems, such as misjudgments, lack of causal analysis, and inadequate feedback loops. It proposes an artificial intelligence-based method and system for financial business data risk early warning management. This invention combines the actual data flow and logical dependencies between different modules of financial business to construct a behavioral modeling mechanism for business scenarios such as budgeting, shared services, assets, and taxation. This mechanism can identify abnormal deviation patterns with potential risks. Simultaneously, it introduces a causal analysis framework based on business behavior chains to uncover related events and potential sources behind anomalies, thereby explaining the causes of anomalies and initially identifying responsible parties. Furthermore, this invention designs a feedback response mechanism that integrates with existing systems, enabling closed-loop processing and tracking of risk events. Overall, this solution, with intelligent identification as the entry point, causal analysis as the core, and a risk management closed loop as the goal, significantly improves the interpretive and governance capabilities of financial risk early warning systems in complex business environments. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the financial business data risk early warning management method based on artificial intelligence according to the present invention.

[0043] Figure 2 This is a schematic diagram of the AI-based financial business data risk early warning management system of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0045] like Figure 1 As shown in one embodiment, the present invention provides an artificial intelligence-based financial business data risk early warning management method, including:

[0046] Step 1: Obtain the time series of the actual execution amount, and decompose the time series of the actual execution amount to obtain the anomaly score of the budget item.

[0047] This step, serving as the starting point of the entire patent system, primarily aims to identify abnormal behaviors that deviate from the normal budget execution rhythm by modeling the company's actual budget execution data. This invention introduces multi-scale time series decomposition and a dynamic threshold mechanism to establish an execution model that conforms to the business characteristics of each type of budget item, thereby achieving accurate identification of budget anomalies and providing fundamental data support for subsequent causal analysis and feedback.

[0048] The data for this step comes directly from the company’s existing information systems, mainly including the budget management system (BMS), enterprise resource planning system (ERP), and financial shared service platform.

[0049] Budget amount information is provided by the budget management system, actual expenditure information is recorded by the ERP system, and auxiliary information such as execution and accounting time is obtained through the financial shared service platform.

[0050] Preferably, obtaining the time series of budget items includes:

[0051] Obtain the company's budget execution data, which includes time, budget items, actual execution amount, and corresponding budget amount;

[0052] Convert the actual execution amount in the budget execution data into a time series to obtain the time series of the actual execution amount.

[0053] Data is collected via a RESTful API or a standard SQL database interface, and is aggregated by budget item each time it is collected, with a granularity of monthly.

[0054] The data collection fields include: time, budget item, actual amount executed (unit: yuan), and corresponding budget amount (unit: yuan).

[0055] Data is transmitted in standardized JSON or Parquet format.

[0056] All collected records are uniformly converted into time series for time series modeling.

[0057] Preferably, the time series of the actual executed amount is decomposed to obtain anomaly scores for budget items, including:

[0058] S11, perform trend-seasonal-residual decomposition on the time series of actual executed amounts to obtain the long-term trend, cyclical fluctuations, and residual terms of budget execution, including:

[0059] Convert all collected actual execution amounts into a time series structure ,in Indicates budget item In time The actual amount executed is used for subsequent trend modeling analysis. T represents the time span of the data.

[0060] This step uses the STL (Seasonal-Trend decomposition using LOESS) method to analyze the time series. Perform a trend-seasonality-residual triple decomposition. The decomposition expression is as follows:

[0061]

[0062] in:

[0063] Indicates budget item In time Actual expenditure amount (input data points);

[0064] This is a trend term, representing the long-term trend of budget expenditures;

[0065] This is a seasonal item, reflecting cyclical fluctuations (such as quarterly purchases, year-end rush, etc.).

[0066] These are residual terms used to capture potential non-seasonal anomalies. This invention employs an STL method with an annual period parameter (period set to 12) for each type of budget item. Independent modeling. The model is deployed on a data computing platform and supports parallel execution by budget item.

[0067] S12, calculates the outlier scores for budget items based on the residuals, including:

[0068] After decomposition, this invention focuses on the residual term. Through analysis Calculate the standard deviation of the distribution in the historical time series. As a reference for anomaly identification.

[0069]

[0070] in, Indicates time Budget items Abnormal scores; This is the residual term at that point in time; For the corresponding budget item The standard deviation of residuals in the historical series. If If it is, then it is considered a significant abnormality.

[0071] For example, suppose the residual item for travel expenses in March 2023 is 900 yuan, and the historical residual standard deviation for this type of budget item is 300 yuan, then... If the threshold for abnormality is reached, the system will automatically mark the budget item and time point as abnormal.

[0072] Step 2: Obtain the anomaly confidence score for the budget item based on the anomaly score, and determine whether the budget item is an abnormal budget item based on the anomaly confidence score.

[0073] Building upon the initial screening of budget anomalies in the previous stage, this step further constructs a behavioral rhythm anomaly judgment mechanism with business semantic understanding capabilities. The core objective is to accurately identify truly abnormal budget items, rather than relying solely on "superficial anomalies" judged by numerical deviations. In large enterprises, budget execution behavior exhibits significant rhythmicity and business-driven patterns. For example, some budget items tend to be released in concentrated bursts at the beginning of quarters or end of the year, while others show regular jumps. Ignoring these cyclical patterns and relying solely on numerical anomaly scores for judgment can easily lead to false alarms and wasted management resources. Therefore, this step introduces historical rhythm profile modeling, rhythm deviation factors, and budget item risk-driven field functions to further identify and classify abnormal behavior.

[0074] Preferably, obtaining the anomaly confidence score for the budget item based on the anomaly score includes:

[0075] S20, calculate the median of historical rhythms based on anomaly scores.

[0076] To improve the semantic accuracy of anomaly detection, this invention constructs a "historical behavior rhythm library," the basic idea of ​​which is: for each budget item... By constructing a typical budget rhythm template from the execution data of the same months over the past three years, a benchmark for behavioral deviations can be established. For example, travel expense budgets show significantly higher expenditures in February, July, and November each year than in other months; this rhythm should be considered "reasonable fluctuations due to inertia" rather than an anomaly.

[0077] This invention models this rhythm template as:

[0078]

[0079] in:

[0080] Indicates budget item The historical median of rhythm;

[0081] It was before Abnormal scores for the same month of the year;

[0082] pass A three-year baseline has been established cumulatively.

[0083] `median` means to get the median within the parentheses.

[0084] For a certain type of budget item (such as travel expenses, office supplies procurement, etc.), review the actual budget expenditure behavior of that month in the past three years, especially its abnormal score results at that time, in order to form a cross-year periodic behavior benchmark.

[0085] For example, if the current time is July 2024, we will look back at the abnormal scores of this budget item in July 2023, July 2022, and July 2021 as a reference for its historical budget execution rhythm at the point in "July".

[0086] S21, Calculate the rhythm deviation of the budget item based on the historical rhythm median:

[0087]

[0088] in:

[0089] This indicates the degree of deviation of the budget item from the rhythm at time t;

[0090] yes The median absolute deviation corresponding to historical rhythms is used to enhance robustness to extreme values;

[0091] The maximum value in the denominator is used to prevent division by zero, while also controlling the scoring scale.

[0092] The innovation lies in the fact that this invention does not directly... Instead of comparing to a static threshold, we compare to the historical median level of rhythm, and use MAD instead of standard deviation to enhance robustness to outliers. This design takes into account that the historical distribution of some budget items is highly volatile, and standard deviation may be distorted; using the median and MAD can more reliably reflect rhythmic behavior.

[0093] S22, Calculate the abnormal confidence score of the budget item based on the rhythm deviation of the budget item.

[0094] This invention introduces a business-driven factor when determining anomalies. , Simulate the "potential resistance" to budget anomalies in a real business environment, that is, the business rationale for anomalies in certain months or budget items is relatively strong, such as year-end asset liquidation or year-end centralized procurement. It can be built using the business calendar and major operational event data provided by the enterprise, such as:

[0095] A centralized settlement window is available every December;

[0096] The project's quarterly budget fluctuated significantly upon launch.

[0097] Policy changes have led to a concentrated adjustment of tax budgets.

[0098] A higher value indicates that the anomaly is more "reasonable," while a lower value suggests that it may be caused by a system problem.

[0099] Therefore, this invention defines the final "anomaly confidence score" as follows:

[0100]

[0101] in:

[0102] It is the anomaly confidence score ultimately used to determine whether a budget item is considered an anomaly;

[0103] This represents the business driver factor; the closer it is to 1, the more reasonable the budget fluctuations are for the business.

[0104] This is a control factor (e.g., 0.5~1.0) used to control the degree to which business rationality mitigates the impact of anomaly scoring.

[0105] Preferably, determining whether a budget item is an abnormal budget item based on anomaly confidence score includes:

[0106] The system determines whether the anomaly confidence score is greater than the set anomaly confidence score threshold. If it is, the budget item is considered an anomaly; otherwise, it is not. For example, if the set anomaly confidence score threshold is 3, then... The budget item will be included in the time frame. If it is determined to be abnormal, proceed to the subsequent cause-effect graph modeling process.

[0107] For example, if budget item c appears in December... Historical rhythm median , ,and (i.e., the business-driven factors are strong), then:

[0108]

[0109]

[0110] Because it does not exceed 3, it is not considered abnormal. Even The score was high, but due to the strong business rationale, it was ultimately not misjudged.

[0111] Step 3: Construct a directed graph based on business event data and abnormal budget items.

[0112] The core task of this step is to complete a structured, graphical modeling process to characterize the potential transmission path between budget anomalies and their possible business causes.

[0113] Preferably, step 3 includes:

[0114] S30 encodes each event in the business event data as a node in a directed graph;

[0115] This invention first obtains structured business event data from various enterprise business systems. Unlike log-style data, this step focuses on standard business event flows, i.e., completed data objects and their execution metadata in the system, including but not limited to:

[0116] The types of documents generated in the financial shared service platform (such as travel applications and invoice entries) and their submission, allocation, and review times;

[0117] Records of asset transfer to fixed assets, final settlement, and provisional entry in the asset management system;

[0118] Monthly tax burden rate, corporate income tax adjustment records, etc. in the tax system;

[0119] Budget system operations include budget modification, budget approval, and freeze operations.

[0120] The aforementioned business event data is imported into the graph building engine through a unified structure table. Fields include event type (e.g., type=invoice_submitted), occurrence time (e.g., 2023-02-03), amount (e.g., x=12000 yuan), and associated budget items (e.g., c=travel expenses).

[0121] Graph building engines can be implemented based on graph databases (such as Neo4j or TigerGraph), or they can be built using distributed graph computing platforms (such as Spark GraphFrames) or custom graph transformation modules to complete each business event. Encoded as a node in the graph The attribute vector of a node includes:

[0122] Belonging system (such as ERP, financial shared service platform);

[0123] Operation time ;

[0124] Related budget items ;

[0125] Percentage of the amount (e.g., the proportion of the monthly budget for this budget item);

[0126] Structured business metrics such as approval status and number of process interruptions.

[0127] S31, Connect the nodes in the directed graph according to the connection rules to obtain the edges in the directed graph;

[0128] The graph structure is initially set as a directed graph. , where the set of nodes The set of nodes for business events and nodes for budget anomalies, and the set of edges. This represents the business impact path between two events. The connection rules of the edges are set by the expert system according to the business process logic, such as "shared document submission → shared document review → accounting → budget expenditure" forming a standard process chain.

[0129] The reason for listing "business event nodes" and "budget anomaly nodes" side-by-side in the node set v is to emphasize the existence of two semantically different but functionally related node types in the graph. This is not a logical duplication, but rather a means to achieve a clearer modeling representation. In other words, business event nodes represent specific execution actions from the business system (such as document submission, approval, asset transfer, etc.); while budget anomaly nodes are result nodes calculated by the anomaly detection model in the previous stage, specifically used to indicate the fact that a budget item exhibits abnormal behavior at a certain moment. Although the formation of budget anomalies is related to business events, modeling them separately as nodes in the graph facilitates the subsequent establishment of a clear causal path between the anomaly and its potential causal events by graph neural networks or attention mechanisms.

[0130] S32, calculate the weight of each edge in the directed graph.

[0131] In edge weight learning, this invention proposes a lightweight weight modeling function that combines budget deviation, time distance, and business rationality to assign weights to edges in the graph:

[0132]

[0133] in:

[0134] Representing an edge The weights;

[0135] This represents the degree of deviation of the budget item from the rhythm at time t; a larger value indicates a higher degree of abnormality.

[0136] For nodes and The time difference between them (in months) represents their temporal distance;

[0137] 3 is the time penalty coefficient (e.g., 0.3~0.5), used to control the decreasing weight of long-term events;

[0138] It is a business-driven factor used to suppress association speculation in a highly reasonable context (the more reasonable the context, the weaker the edge).

[0139] This weighting formula embodies three core innovation mechanisms:

[0140] use By strongly linking the graph modeling process with budget anomalies, the structure in the graph is more likely to connect truly deviating budget anomalies.

[0141] Introducing an inverse proportional function of time distance reflects the "proximate importance" in the causal propagation process;

[0142] use Perform soft filtering on highly reasonable business behaviors to avoid creating incorrect maps based on reasonable deviations.

[0143] For example, if a budget item showed an abnormality in March 2023 Business rationality factors during anomalies (Relatively low), and if the approval time of a shared document is more than 2 days later than its posting date, then the weight of the edge is approximately:

[0144]

[0145] This value indicates that the connection between the two events in the graph has medium to high importance and will be retained in subsequent path reasoning.

[0146] After the graph is constructed, no path inference or scoring is performed; only the graph structure is output for use in subsequent steps.

[0147] Step 4: Obtain all paths in the directed graph that satisfy the path constraints and terminate at the node with budget anomaly, thus obtaining the path set; calculate the path score for each path.

[0148] This step aims to construct a path-level scoring model for budget anomalies, providing a structured modeling foundation for subsequent anomaly causal explanations and handling recommendations. This model is based on the directed graph reflecting the causality of business events constructed in the previous step. By combining the edge weights, historical co-occurrence statistics, and business semantic features of each path, a path importance modeling mechanism is established.

[0149] Preferably, all paths in the directed graph that satisfy the path constraints and terminate at the node with budget anomaly are obtained, resulting in a path set including:

[0150] In a directed graph, search for all paths that satisfy the business constraints to the reachable upstream event nodes in the directed graph, and store the obtained paths into a path set.

[0151] To characterize the potential causes of budget anomalies, this invention firstly... Each node with budget anomalies in China and Israel Using the endpoint as the destination, search for all paths that satisfy the business constraints to the upstream event nodes reachable from the graph, forming a path set. Each path It is a directed path of finite length with event time that is monotonically non-increasing.

[0152] Search for all paths that satisfy the business constraints among the reachable upstream event nodes in the graph, including:

[0153] In G, a node with a budget anomaly is used as the endpoint of a path. The system traces upstream to nodes of business events that could have caused the anomaly, searching layer by layer for the preceding nodes connected by their incoming edges, thus constructing all directed paths that satisfy specific business process rules. "Satisfying business constraints" includes, but is not limited to: the operation time between nodes must decrease from back to front (i.e., the time sequence must be reasonable); the path length must not exceed the set maximum number of hops (e.g., a maximum of 5 backtracking steps); and the combination of node types in the path must conform to the enterprise's standard business process sequence. For example, "budget freeze → document application → document approval → invoice payment → anomaly alarm" is considered a valid path; combinations that do not conform to process logic (e.g., invoice payment occurs before the application) will be eliminated. This path search process can be implemented using depth-first or breadth-first traversal algorithms, supplemented by a business process knowledge graph or rule base as constraints to ensure that the obtained paths have business rationality and causal traceability.

[0154] To process each path To establish an "anomaly formation probability" score, this invention proposes a path modeling method based on a path attention mechanism. The core idea is to construct a function for calculating the path score by integrating the structural weights, semantic factors, and historical support of each edge in the path.

[0155]

[0156] in:

[0157] For path Path scoring;

[0158] Indicates the number of hops (length) of the path;

[0159] The weight of the edge;

[0160] This is the semantic attention factor introduced in this step, representing the edge. Confidence level of historical co-occurrence with budget anomalies.

[0161] The calculation introduces a function design based on statistical frequency:

[0162]

[0163] in:

[0164] In the past In this budget anomaly event, the side The proportion of the combined event types and budget exception events that occur together;

[0165] For example, take an empirical threshold (e.g., take...) This is used to determine whether a chain belongs to a "high-frequency causal chain".

[0166] For adjustment coefficients (e.g., take...) ), control attention sensitivity.

[0167] The actual implementation of this design involves: performing offline statistical analysis on historical enterprise data; identifying which business event combinations frequently occurred before the anomalies in past budget anomaly events; and then transforming these combinations into... The numerical support basis.

[0168] For example, if the edge The phrase connects "Delay in Final Settlement of Project Completion → Delay in Transfer of Assets to Fixed Assets," and this combination has a high proportion of occurrences in historical budget anomalies. ,but ,like ,but This indicates that the semantic causal strength of this edge is relatively strong.

[0169] Finally, path scoring The weighted sum of all edges is then divided by the square root of the path length to balance the sudden behavior of short paths with the cumulative effect of long paths.

[0170] Step 5: Calculate the priority score for each path based on the path score, obtain the main path based on the priority score, and generate response suggestions based on the main path.

[0171] In the previous step, the present invention has obtained a score for each abnormal path ( ) and semantic attention factor ( This step requires generating clear response recommendations based on this, and triggering an executable feedback process through the company's internal system, thereby truly achieving a closed loop of "identification → interpretation → intervention".

[0172] This step strictly uses the output of step 4, and introduces a structured auxiliary data table based on it, specifically including:

[0173] Path collection For each abnormal budget node The corresponding set of high-confidence causal paths;

[0174] Path rating for each path : Measures the overall ability to explain anomalies in a path;

[0175] Semantic attention factor for edges in each path : Measures the strength of anomaly interpretation for key nodes in the path;

[0176] In addition, to complete the responsibility mapping generated by the suggestion, the following call is also required:

[0177] Responsibility Weighting Table This data, maintained by the enterprise master data system, represents the proportion of responsibility each business process node bears within the process. For example, the responsibility of the document submitter is 0.6, and the responsibility of the approver is 0.4. This table is a static configuration table, imported as auxiliary data during system deployment, and does not participate in model training.

[0178] To achieve closed-loop management of budget anomalies, this invention needs to extract core information from the scoring results, generate structured response suggestions, and push them to the responsible unit or person, while simultaneously initiating a processing feedback process in the enterprise business system.

[0179] The priority score of a path is calculated using the following formula:

[0180]

[0181] The explanation is as follows:

[0182] It is a path score that reflects the overall importance of the path, indicating the "strength of contribution" of the path to the current budget anomaly;

[0183] It is the edge with the strongest explanatory factor in the path, representing the most likely "root cause" in the path;

[0184] It is a response priority score; the higher the value, the more worthy the path is of priority.

[0185] Preferably, obtaining the main path based on priority scoring includes:

[0186] The path with the highest priority score among all paths in the path set is selected as the primary path.

[0187] Preferably, generating response suggestions based on the main path includes:

[0188] The system populates a pre-defined response suggestion template with metadata corresponding to the main path and generates a response suggestion.

[0189] Based on the key causal nodes in this path Based on its metadata (such as responsible person, system module, and event type), a preset response suggestion template is populated. Template fields include:

[0190] The person in charge (by) The approver, processor, or system responsible person fields in the node record can be read directly.

[0191] Suggested content (based on) Select a template from the suggestion library to fill in the relevant event type, such as "Please resubmit the document" or "Check the asset transfer process".

[0192] Urgency level (by The comparison is made against a threshold set by the company; values ​​exceeding 2.5 are considered "high priority".

[0193] Suggested deadline (default 3 business days, configurable).

[0194] The process of obtaining key causal nodes includes:

[0195] Traverse all edges (i,j) in the main path and find the edge that makes... Find the largest edge, and then take the node corresponding to the starting point of that edge as the key cause node of the current anomaly. .

[0196] For example:

[0197] Exceptional budget item: Travel expenses (March 2023);

[0198] main path This includes "delayed document submission → shared rollback → approval omission → accounting time deviation → budget anomaly";

[0199] Highest score node The incident was classified as "approval omission," the person responsible was "Li Si," and the corresponding event type was "shared system approval operation."

[0200] The system then generates the following suggestion:

[0201] Recipient: Li Si

[0202] Content: Your approved travel document (No. X) was delayed in March, resulting in a budget discrepancy. Please complete the approval information and submit an explanation within 3 days.

[0203] Urgency level: High

[0204] Feedback path: Unified process feedback platform URL

[0205] All suggestions are pushed through the company's internal notification system (such as OA, WeChat Work, email system, etc.). To ensure a closed loop, the system will call the business system interface to monitor whether the responsible person completes the rectification action (such as resubmitting documents, changing asset status, etc.) within the specified time. After the rectification is completed, the system will automatically mark it as "closed loop"; otherwise, an unprocessed alarm will be generated and written to the anomaly record center.

[0206] On the other hand, such as Figure 2 The present invention provides a financial business data risk early warning management system based on artificial intelligence, including a first acquisition module, a second acquisition module, a construction module, a third acquisition module and a generation module;

[0207] The first acquisition module is used to acquire the time series of the actual execution amount, and to decompose the time series of the actual execution amount to obtain the abnormal scores of the budget items;

[0208] The second acquisition module is used to obtain the anomaly confidence score of the budget item based on the anomaly score, and to determine whether the budget item is an abnormal budget item based on the anomaly confidence score.

[0209] The building module is used to construct a directed graph based on business event data and exception budget items;

[0210] The third acquisition module is used to acquire all paths in the directed graph that satisfy the path constraints and end at the node with budget anomaly, and obtain a path set; and calculate the path score for each path.

[0211] The generation module calculates the priority score for each path based on the path score, obtains the main path based on the priority score, and generates response suggestions based on the main path. The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for risk early warning management of financial and business data based on artificial intelligence, characterized in that: include: Step 1: Collect the company's budget execution data from the company's budget management system and enterprise resource planning system; The actual execution amount in the budget execution data is converted into a time series. The STL method is used to perform trend-seasonal-residual decomposition on the time series to obtain residual items. Based on the residual items and their historical residual standard deviations, the anomaly score of the budget item is calculated, as follows: ； in, This indicates the abnormal score of budget item c at time t; This is the residual term at that point in time; This represents the standard deviation of the residuals for the corresponding budget item c in the historical time series. Step 2: Obtain the anomaly confidence score for each budget item based on the anomaly score, and determine whether a budget item is an anomaly based on the anomaly confidence score, including: The historical tempo median is calculated based on anomaly scores, using the following formula: ； in, This represents the historical median of budget item c; These are abnormal scores for the same month in the previous i years, obtained through... A three-year baseline is constructed cumulatively; median indicates that the median within the parentheses is retrieved. The rhythm deviation of budget items is calculated based on the historical rhythm median, using the following formula: ； in: This indicates the degree of deviation of the budget item from the rhythm at time t; yes The median absolute deviation corresponding to historical rhythms; The outlier confidence score for each budget item is calculated based on its rhythm deviation, using the following formula: ； in, This indicates the abnormal confidence score for the budget item. Indicates business driving factors; As a regulatory factor; Step 3: Encode each business event in the business event data into a node in a directed graph. The node's attribute vector includes: system, operation time, associated budget item, amount percentage, approval status, and number of process interruptions. Connect the nodes in the directed graph according to preset connection rules to obtain the edges in the directed graph, where the directed graph G=(V,E), the node set V is the set of nodes of business events and nodes of budget anomalies, and the edge set E represents the business impact path between the two events. Calculate the weight of each edge based on the rhythm deviation of the budget item, the time distance between nodes, and the business driving factor to construct a directed graph containing the nodes and edges. Step 4: In the directed graph, using the node with the budget anomaly as the endpoint of the path, backtrack upstream to search for all paths that satisfy the business constraints, obtaining a path set; calculate the path score for each path using the following formula: ； in: For path Path scoring; Indicates the number of hops in the path; The weight of the edge; It is a semantic attention factor; Step 5: Calculate the priority score for each path based on the path score, obtain the main path based on the priority score, and generate response suggestions based on the main path.

2. The method for risk early warning management of financial business data based on artificial intelligence according to claim 1, characterized in that, The budget execution data includes time, budget items, actual execution amount, and corresponding budget amount.

3. The method for risk early warning management of financial business data based on artificial intelligence according to claim 1, characterized in that, Determining whether a budget item is abnormal based on anomaly confidence scoring includes: Determine whether the anomaly confidence score is greater than the set anomaly confidence score threshold. If yes, the budget item is considered an anomaly; otherwise, the budget item is not considered an anomaly.

4. The method for risk early warning management of financial business data based on artificial intelligence according to claim 1, characterized in that, The main path is obtained based on priority scoring, including: The path with the highest priority score among all paths in the path set is selected as the primary path.

5. The method for risk early warning management of financial business data based on artificial intelligence according to claim 1, characterized in that, Response suggestions are generated based on the main path, including: The system populates a pre-defined response suggestion template with metadata corresponding to the main path and generates a response suggestion.

6. A financial business data risk early warning management system based on artificial intelligence, characterized in that: It includes a first acquisition module, a second acquisition module, a construction module, a third acquisition module, and a generation module; The first acquisition module is used to collect budget execution data from the enterprise's budget management system and enterprise resource planning system; the actual execution amount in the budget execution data is converted into a time series, and the time series is decomposed into trend-seasonal-residual using the STL method to obtain residual items. Based on the residual items and their historical residual standard deviations, the anomaly score of the budget items is calculated, as follows: ； in, This indicates the abnormal score of budget item c at time t; This is the residual term at that point in time; This represents the standard deviation of the residuals for the corresponding budget item c in the historical time series. The second acquisition module is used to obtain the anomaly confidence score of the budget item based on the anomaly score, and to determine whether the budget item is an anomaly budget item based on the anomaly confidence score, including: The historical tempo median is calculated based on anomaly scores, using the following formula: ； in, This represents the historical median of budget item c; These are abnormal scores for the same month in the previous i years, obtained through... A three-year baseline is constructed cumulatively; median indicates that the median within the parentheses is retrieved. The rhythm deviation of budget items is calculated based on the historical rhythm median, using the following formula: ； in: This indicates the degree of deviation of the budget item from the rhythm at time t; yes The median absolute deviation corresponding to historical rhythms; The outlier confidence score for each budget item is calculated based on its rhythm deviation, using the following formula: ； in, This indicates the abnormal confidence score for the budget item. Indicates business driving factors; As a regulatory factor; The construction module encodes each business event in the business event data into a node in a directed graph. The node's attribute vector includes: system, operation time, associated budget item, amount percentage, approval status, and number of process interruptions. It connects the nodes in the directed graph according to preset connection rules to obtain the edges in the directed graph, where the directed graph G=(V,E), the node set V is the set of nodes of business events and nodes of budget anomalies, and the edge set E represents the business impact path between the two events. It then calculates the weight of each edge based on the rhythm deviation of the budget item, the time distance between nodes, and business driving factors, constructing a directed graph containing the nodes and edges. The third acquisition module is used in the directed graph to backtrack upstream to search for all paths that meet the business constraints, using the node with the budget anomaly as the endpoint of the path, to obtain a path set; and to calculate the path score for each path. The formula is as follows: ; in: For path Path scoring; Indicates the number of hops in the path; The weight of the edge; It is a semantic attention factor; The generation module is used to calculate the priority score for each path based on the path score, obtain the main path based on the priority score, and generate response suggestions based on the main path.

Images