CN-122022459-A - Multi-source data fusion intelligent fund supervision method and system for chain stores

CN122022459ACN 122022459 ACN122022459 ACN 122022459ACN-122022459-A

Abstract

The application discloses a multi-source data fusion intelligent fund supervision method and system for chain stores, and belongs to the technical field of artificial intelligence fund data supervision. The intelligent fund supervision method comprises the steps of obtaining multi-source heterogeneous fund data of interlocking stores, carrying out unified fusion processing on the multi-source heterogeneous fund data of each interlocking store, abstracting each interlocking store and a corresponding associated party into fund nodes according to the fused multi-source heterogeneous fund data, constructing a fund cooperation network taking the fund nodes as vertexes and taking a fund circulation relation as a directed edge, calculating the comprehensive health value of each target node in the fund cooperation network, determining the risk level matched with the target nodes according to the comprehensive health value, and applying automatic supervision operation matched with the risk level to the target nodes, so that the intelligent fund supervision effect of the interlocking stores is achieved.

Inventors

Bao Minliang
LUAN LISHENG

Assignees

杭州动态宇宙科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260119

Claims (10)

1. A multi-source data fusion intelligent fund supervision method for chain stores is characterized by comprising the steps of obtaining multi-source heterogeneous fund data of the chain stores, conducting unified fusion processing on the multi-source heterogeneous fund data of each chain store, abstracting each chain store and a corresponding associated party into fund nodes according to the fused multi-source heterogeneous fund data, constructing a fund cooperation network taking the fund nodes as peaks and taking a fund circulation relationship as a directed edge, calculating the comprehensive health value of each target node in the fund cooperation network, determining the risk level matched with the target node according to the comprehensive health value, and applying automatic supervision operation matched with the risk level to the target node.
2. The method of claim 1, wherein the step of calculating the composite health value for each target node in the fund collaboration network comprises the steps of overall network impact values, local collaboration bias values, and individual turnover health values, determining individual turnover health values for each target node in the fund collaboration network by a preset network assessment model, determining local collaboration bias values for a plurality of neighboring nodes, and determining overall network impact values for all target nodes.
3. The method of claim 2, wherein the steps of determining the individual turnover health value of each target node in the fund cooperative network through a preset network evaluation model, determining local cooperative deviation values corresponding to a plurality of adjacent nodes, and determining the overall network influence value corresponding to all target nodes comprise the steps of respectively performing feature extraction on historical operation running data of the target node, historical fund business data of all the adjacent nodes of the target node and historical data of all the nodes in the fund cooperative network to form a multidimensional feature vector of the target node, inputting the multidimensional feature vector of the target node and the adjacent relation thereof in the fund cooperative network into the preset network evaluation model, encoding the topological structure and the node relation of the fund cooperative network through a graph neural network layer in the network evaluation model, generating a structural feature vector of the node, mapping and calculating the structural feature vector through a full connection layer in the network evaluation model, and synchronously outputting the individual turnover health value, the local cooperative deviation value and the overall cooperative network influence value corresponding to the target node.
4. The method of claim 2, wherein before the step of performing the fund circulation analysis on each target node in the fund cooperation network through a preset network evaluation model, the method further comprises the steps of extracting a fund inflow sequence, a fund outflow sequence and proportional distributions of transactions with different correspondents of each target node in the fund cooperation network as time sequences and structural features, establishing a model to be trained comprising a graph neural network layer and a full-connection layer, wherein the graph neural network layer is used for encoding a topological structure and a node relation of the fund cooperation network, the full-connection layer is used for performing mapping and calculation of node health values based on the encoded features, performing iterative training on the model to be trained by using a historical dataset comprising fund data of the target node and corresponding risk labels, fitting a correlation degree of a fund circulation result and a risk grade of the target node as an optimization target, and optimizing model parameters by a gradient descent algorithm to obtain a final deployable network evaluation model.
5. The method of claim 2, wherein after the step of applying an automated policing operation to the target node that matches the risk level, the method further comprises collecting a manual review conclusion of historical automated policing operations as a feedback signal, and adjusting the network assessment model to calculate the overall network impact value, the local collaborative bias value, and the individual turn-around health value, based on the feedback signal, the indicator weights and risk decision thresholds involved.
6. The method of claim 1, wherein determining the risk level to which the target node is matched based on the integrated health value comprises determining a network hierarchy scale consisting of different numbers of target nodes, converting the global network impact value, the local collaboration bias value, and the individual turnover health value for each target node to bias values for each of the network hierarchy scales, weighting and fusing the bias values for each hierarchy to obtain a risk bias score, and determining the risk level to which the target node is matched based on the risk bias scores.
7. The method of claim 6, wherein the step of determining the risk level matched by the target node according to the risk deviation score includes performing risk level matching from a preset overall network risk level table, a preset local network risk level table and a preset individual risk level table according to the risk deviation score, determining the risk level matched by the target node as low risk if the matching result of all risk level tables is low risk, determining the risk level matched by the target node as high risk if the matching result of any risk level table is high risk, and determining the risk level matched by the target node as medium risk if the matching result does not meet any of the above conditions.
8. The method of claim 7, wherein the step of performing risk level matching from a preset overall network risk level table, a preset local network risk level table and a preset individual risk level table according to the risk deviation score comprises the steps of obtaining historical data of all nodes in the fund collaborative network, determining network centrality indexes and network flow rates of all target nodes according to the historical data, dividing the network centrality indexes and the network flow rates into a plurality of continuous intervals, combining and mapping an overall network risk level for each interval, constructing an overall network risk level table, obtaining historical fund current data of the target nodes and adjacent nodes in the fund collaborative network, determining local connection closeness and fund flow rates of all the target nodes according to the historical fund current data, dividing the connection degree and the flow rates into a plurality of continuous intervals, combining and mapping a local network risk level for each interval, constructing a local network risk level table, obtaining the historical fund current data of all the target nodes and the adjacent nodes, dividing the current operation level into a plurality of the current operation levels, and settling the individual risk level, and calculating the historical operation level deviation from the historical operation level of all the target nodes, and the current operation level is divided into a plurality of the continuous intervals, and the individual operation level is determined according to the historical state of the current state of the target nodes.
9. The method of claim 7, wherein the step of applying an automated policing operation to the target node that matches the risk level includes marking the target node as a regular monitoring state if the risk level of the target node is low, archiving cash flow records generated by the target node, performing an anomaly marking on large amounts of cash flow subsequently generated by the target node if the risk level of the target node is medium risk, and generating a corresponding report based on the result of the anomaly marking, and suspending some or all of the cash flow operations of the target node if the risk level of the target node is high risk.
10. The intelligent funds supervision system of the chain stores based on the multi-source data fusion of the method of any one of claims 1-9, which is characterized by comprising an acquisition module, a construction module, a calculation module and a supervision module, wherein the acquisition module is used for acquiring multi-source heterogeneous funds data of the chain stores, carrying out unified fusion processing on the multi-source heterogeneous funds data of each chain store, the construction module is used for abstracting each chain store and a corresponding associated party into funds nodes according to the fused multi-source heterogeneous funds data, constructing a funds cooperation network taking the funds nodes as peaks and taking funds circulation relations as directed edges, the calculation module is used for calculating comprehensive health values of each target node in the funds cooperation network, and the supervision module is used for determining risk grades matched with the target nodes according to the comprehensive health values and applying automatic supervision operations matched with the risk grades to the target nodes.

Description

Multi-source data fusion intelligent fund supervision method and system for chain stores Technical Field The application relates to the technical field of fund data supervision of artificial intelligence, in particular to a chain store intelligent fund supervision method and system with multi-source data fusion. Background In the chain operation field, with the expansion of the door shop rule model and the complexity of the business model, the fund management becomes the core and the difficulty of the management and control of the brand side operation increasingly. Each store generates a huge amount of transaction flow every day, has frequent settlement with a plurality of suppliers, and simultaneously relates to various funds such as member storage value, expense reimbursement and the like. The fund activities are scattered in different stores, different bank accounts and different business systems, heterogeneous data with various sources and different formats are formed, and therefore real-time perception of the overall fund flow direction by a headquarter, accurate identification of abnormal risks and rapid intervention face great challenges. Currently, common fund supervision schemes rely mainly on two types of technical paths. Firstly, the post audit based on traditional enterprise resource planning or financial software, namely, the manual audit is carried out on the summarized financial statement in a fixed period (such as month), and the attention is paid to checking the compliance and accuracy of the accounting process. And secondly, an independent and rule-based risk early warning system is deployed, and a trigger type alarm is carried out by setting a simple financial index threshold (such as cash lower limit and large transaction limit) for a single store. The methods realize the monitoring of dominant and static risks to a certain extent. However, the above conventional solutions have significant drawbacks, such as static and punctual analysis of isolated shops or single transactions, lack of view of the whole federation system as a dynamic organic network, and are unable to describe systematic risks (such as collusion across shops and abnormal circulation of funds among related parties) implied by complex circulation of funds among multiple nodes of shops, suppliers, headquarters and the like, thereby resulting in poor supervision effect. Disclosure of Invention The application mainly aims to provide a multi-source data fusion intelligent fund supervision method and system for a chain store, and aims to solve the technical problem that the conventional common fund supervision scheme is poor in supervision effect. In order to achieve the above purpose, the application provides a multi-source data fusion intelligent fund supervision method for a chain store, which comprises the following steps: Acquiring multi-source heterogeneous fund data of the interlocking store, and carrying out unified fusion processing on the multi-source heterogeneous fund data of each interlocking store; according to the fused multi-source heterogeneous fund data, abstracting each interlocking store and a corresponding associated party as fund nodes, and constructing a fund cooperation network which takes the fund nodes as vertexes and takes a fund circulation relationship as a directed edge; calculating a comprehensive health value of each target node in the fund collaboration network; and determining the risk level matched with the target node according to the comprehensive health value, and applying automatic supervision operation matched with the risk level to the target node. In an embodiment, the step of calculating a composite health value for each target node in the funding collaboration network comprises: the comprehensive health value comprises an overall network influence value, a local cooperation deviation value and an individual turnover health value; and determining an individual turnover health value of each target node in the fund cooperation network through a preset network evaluation model, determining local cooperation deviation values corresponding to a plurality of adjacent nodes, and determining an overall network influence value corresponding to all the target nodes. In an embodiment, the step of determining, through a preset network evaluation model, an individual turnover health value of each target node in the fund collaboration network, determining local collaboration deviation values corresponding to a plurality of neighboring nodes, and determining overall network influence values corresponding to all target nodes includes: Respectively extracting characteristics of historical operation flow data of the target node, historical fund business data of all the target node and all adjacent nodes thereof, and historical data of all nodes in the fund cooperation network to form a multidimensional characteristic vector of the target node; inputting the multi-dimensional feature vector of the target