Search

CN-121481736-B - Cross-border fund flow direction data analysis method and system based on depth feature fusion

CN121481736BCN 121481736 BCN121481736 BCN 121481736BCN-121481736-B

Abstract

The invention discloses a cross-border fund flow direction data analysis method and a system based on depth feature fusion, which relate to the technical field of digital transaction analysis and comprise the steps of extracting time sequence feature vectors and map feature vectors according to initial analysis granularity and fusing to generate fusion feature vectors; the method comprises the steps of establishing a risk time sequence data neighborhood model taking a preset ideal point as a center, mapping a fusion feature vector into the model to obtain a position coordinate, correcting an initial analysis granularity according to a difference value to obtain a first analysis granularity, re-executing feature extraction and risk assessment until the re-execution times reach a preset upper limit or Euclidean distance between two adjacent position coordinates is smaller than a preset second threshold value to obtain a final risk score, and sending out a second early warning and outputting a visual fund flow chart if the final risk score is larger than the preset first threshold value. The system improves the accuracy and the self-adaptive capacity of abnormal transaction detection through feature fusion and dynamic granularity adjustment.

Inventors

  • XU JING

Assignees

  • 北京物资学院

Dates

Publication Date
20260508
Application Date
20251111

Claims (9)

  1. 1. The cross-border fund flow direction data analysis method based on depth feature fusion is characterized by comprising the following steps of: s1, acquiring financial transaction data and blockchain transaction data, performing feature extraction according to initial analysis granularity, obtaining a time sequence feature vector and a map feature vector, performing feature fusion, and generating a fusion feature vector; S2, constructing a risk time sequence data neighborhood model, wherein a preset ideal point is used as a neighborhood center, and a risk distance function is constructed and obtained, and the risk distance function is obtained through calculation according to a risk offset and a time attenuation; s3, mapping the fusion feature vector into a risk time sequence data neighborhood model, acquiring a corresponding position coordinate of the fusion feature vector, calculating to obtain an abnormal transaction risk score according to the position coordinate through a risk distance function, judging whether the abnormal transaction risk score is larger than a preset first threshold, if so, sending out primary early warning, otherwise, entering S4; S4, correcting the initial analysis granularity according to the difference value between the abnormal transaction risk score and a preset first threshold value to obtain a first analysis granularity, re-executing S1 to S3 according to the first analysis granularity, and recording the re-executing times until the re-executing times reach a preset upper limit or the Euclidean distance of the position coordinates corresponding to the abnormal transaction risk scores of two adjacent times is smaller than a preset second threshold value to obtain the abnormal transaction risk score after the execution is finished; S5, judging whether the risk score of the abnormal transaction after the execution is finished is greater than a preset first threshold value, if so, sending out a secondary early warning and outputting a visual fund flow chart; The specific process of mapping the fusion feature vector to the risk time sequence data neighborhood model and obtaining the corresponding position coordinate comprises the following steps: The risk time sequence data neighborhood model is a risk assessment space model containing time attenuation factors, and is specifically constructed through a two-dimensional coordinate system, wherein the horizontal axis represents risk offset, the vertical axis represents time attenuation and the time-space characteristic is used for comprehensively assessing risks; Calculating the position coordinates of the fusion feature vector in the two-dimensional coordinate system by using the preset ideal point as a coordinate origin through a coordinate transformation algorithm, wherein the specific formula of the coordinate transformation algorithm is as follows: ; ; Wherein, the The abscissa representing the position coordinates, The ordinate representing the position coordinates, Representing the amount of risk offset that is to be used, The amount of time decay is indicated and, Representing the coordinate rotation angle; The coordinate rotation angle is calculated by the steps of: according to the variance distribution of the fusion feature vector in the preset first dimension and the preset second dimension, obtaining the variance of the fusion feature vector in the preset first dimension and the variance of the fusion feature vector in the preset second dimension, and calculating the coordinate rotation angle according to the variance distribution, wherein the calculation formula of the coordinate rotation angle is as follows: , wherein, Representing the variance of the fused feature vector over a preset first dimension, And representing the variance of the fusion feature vector in a preset second dimension.
  2. 2. The depth feature fusion-based cross-border fund flow direction data analysis method of claim 1, wherein the generating the fusion feature vector specifically comprises: Acquiring financial transaction data and blockchain transaction data; Based on a preset initial analysis granularity, processing financial transaction data by using a time sequence analysis model, extracting transaction frequency trend characteristics, amount fluctuation characteristics and time distribution characteristics, and generating a time sequence characteristic vector; Based on a preset initial analysis granularity, processing blockchain transaction data by using a graph analysis model, extracting address association features, transaction path features and network topology features, and generating a graph feature vector; And carrying out feature fusion processing on the time sequence feature vector and the map feature vector to generate a fusion feature vector.
  3. 3. The depth feature fusion-based cross-border fund flow direction data analysis method of claim 1, wherein the specific process of constructing and obtaining the risk distance function is as follows: Extracting a preset attenuation coefficient from a cross-border fund flow database; obtaining a risk offset by calculating Euclidean distance between the fusion feature vector and a preset unit vector of the preset ideal point in a feature space; Based on the transaction time stamp and the current time extracted from the financial transaction data and the blockchain transaction data, calculating to obtain a time interval, and calculating the time attenuation through an exponential decay function, wherein the specific calculation formula of the time attenuation is as follows: , wherein, The amount of time decay is indicated and, Represents a natural constant of the natural product, Representing the preset attenuation coefficient of the optical fiber, Representing a time interval; And calculating a risk distance by weighted summation according to the risk offset and the time attenuation, wherein a risk distance function is specifically as follows: , wherein, Representing the distance of risk, Representing the amount of risk offset that is to be used, Indicating that a first weight is preset, Indicating that a second weight is preset, 。
  4. 4. The depth feature fusion-based cross-border fund flow direction data analysis method according to claim 1, wherein the specific process of obtaining the abnormal transaction risk score through risk distance function calculation according to the position coordinates is as follows: calculating a normalized distance from the position coordinates to the preset ideal point based on the position coordinates; the calculation formula of the normalized distance is as follows: , wherein, Indicating a preset maximum normalized distance that is to be found, Representing the normalized distance; inputting the normalized distance into a preset risk score conversion function, and calculating to obtain an abnormal transaction risk score, wherein the risk score conversion function is as follows: , wherein, Representing an abnormal transaction risk score, Represents a natural constant of the natural product, Representing a preset coefficient of sensitivity.
  5. 5. The depth feature fusion-based cross-border fund flow direction data analysis method according to claim 1, wherein the specific process of obtaining the first analysis granularity and re-executing S1 to S3 and recording the re-execution times according to the first analysis granularity is as follows: calculating a difference value between the abnormal transaction risk score and a preset first threshold value; inquiring corresponding analysis granularity parameters from a preset granularity configuration table based on the difference value; updating the initial analysis granularity according to the queried analysis granularity parameters to obtain a first analysis granularity; and re-executing the steps S1 to S3 based on the first analysis granularity, generating updated fusion characteristic vectors and re-calculating abnormal transaction risk scores.
  6. 6. The depth feature fusion-based cross-border fund flow data analysis method of claim 5, wherein the configuration rules of the granularity configuration table comprise: when the difference value is smaller than or equal to a preset first difference value threshold value, configuring a corresponding time sequence analysis window as a first duration, and enabling the spectrum analysis depth to be in secondary correlation; When the difference value is larger than a preset first difference value threshold value and smaller than or equal to a preset second difference value threshold value, configuring a corresponding time sequence analysis window as a second duration, and enabling the spectrum analysis depth to be three-degree correlation; When the difference value is larger than a preset second difference value threshold value, configuring a corresponding time sequence analysis window as a third duration, and enabling the spectrum analysis depth to be in full-path correlation; The first time period is longer than the second time period, and the second time period is longer than the third time period.
  7. 7. The depth feature fusion-based cross-border fund flow direction data analysis method according to claim 1, wherein the obtaining the abnormal transaction risk score after the execution is finished specifically comprises: Initializing an execution counter and setting a preset maximum execution number; After each re-execution, the execution counter is increased, and the abnormal transaction risk score calculated under the current times and the corresponding position coordinates thereof are recorded; calculating Euclidean distance between position coordinates obtained by two adjacent execution steps, wherein the calculation formula of the Euclidean distance is as follows: , wherein, The euclidean distance is represented as, 、 Represent the first The position coordinates of the next execution are calculated, 、 Representing the position coordinates of the i-1 st execution; And stopping the re-execution process when the execution counter reaches the preset maximum execution times or the Euclidean distance is smaller than a preset second threshold value, and obtaining the abnormal transaction risk score after the execution is finished.
  8. 8. The depth feature fusion-based cross-border fund flow data analysis method of claim 1, wherein the specific process of outputting the visual fund flow graph comprises: extracting relevant transaction path data from the corresponding financial transaction data and blockchain transaction data based on the finally determined abnormal transaction risk score; constructing an overlay map data structure comprising a traditional financial transaction path and a blockchain transaction path; according to the magnitude of the abnormal transaction risk score, corresponding risk grade identifiers are distributed for corresponding different transaction paths; Three-dimensional visualization chart data of the transaction path including the risk level identification is generated.
  9. 9. Cross-border fund flow direction data analysis system based on depth feature fusion is characterized by comprising: The data acquisition module is used for acquiring financial transaction data and blockchain transaction data, extracting features according to initial analysis granularity, acquiring time sequence feature vectors and atlas feature vectors, carrying out feature fusion, and generating fusion feature vectors; The data construction module is used for constructing a risk time sequence data neighborhood model, wherein a preset ideal point is used as a neighborhood center, and a risk distance function is constructed and obtained, and the risk distance function is obtained through calculation according to a risk offset and a time attenuation; The data judging first module is used for mapping the fusion feature vector into a risk time sequence data neighborhood model, acquiring corresponding position coordinates, calculating according to the position coordinates through a risk distance function to obtain abnormal transaction risk scores, judging whether the abnormal transaction risk scores are larger than a preset first threshold, if so, sending out primary early warning, otherwise, entering the data judging second module; the data judging second module is used for correcting the initial analysis granularity according to the difference value between the abnormal transaction risk score and a preset first threshold value to obtain the first analysis granularity, re-executing the data obtaining module to the data judging first module according to the first analysis granularity, and recording the re-executing times until the re-executing times reach a preset upper limit or the Euclidean distance of the position coordinates corresponding to the abnormal transaction risk score of two adjacent times is smaller than a preset second threshold value to obtain the abnormal transaction risk score after the execution is finished; the data output module is used for judging whether the abnormal transaction risk score after the execution is finished is larger than a preset first threshold value, if so, sending out a secondary early warning and outputting a visual fund flow chart: The specific process of mapping the fusion feature vector to the risk time sequence data neighborhood model and obtaining the corresponding position coordinate comprises the following steps: The risk time sequence data neighborhood model is a risk assessment space model containing time attenuation factors, and is specifically constructed through a two-dimensional coordinate system, wherein the horizontal axis represents risk offset, the vertical axis represents time attenuation and the time-space characteristic is used for comprehensively assessing risks; Calculating the position coordinates of the fusion feature vector in the two-dimensional coordinate system by using the preset ideal point as a coordinate origin through a coordinate transformation algorithm, wherein the specific formula of the coordinate transformation algorithm is as follows: ; ; Wherein, the The abscissa representing the position coordinates, The ordinate representing the position coordinates, Representing the amount of risk offset that is to be used, The amount of time decay is indicated and, Representing the coordinate rotation angle; The coordinate rotation angle is calculated by the steps of: according to the variance distribution of the fusion feature vector in the preset first dimension and the preset second dimension, obtaining the variance of the fusion feature vector in the preset first dimension and the variance of the fusion feature vector in the preset second dimension, and calculating the coordinate rotation angle according to the variance distribution, wherein the calculation formula of the coordinate rotation angle is as follows: , wherein, Representing the variance of the fused feature vector over a preset first dimension, And representing the variance of the fusion feature vector in a preset second dimension.

Description

Cross-border fund flow direction data analysis method and system based on depth feature fusion Technical Field The application relates to the technical field of digital transaction analysis, in particular to a cross-border fund flow direction data analysis method and system based on depth feature fusion. Background With the acceleration of global economic integration processes and the rapid development of digital payment technologies, cross-border funds flow scales continue to expand, and funds flow complexity increases significantly. Traditional money laundering monitoring systems are based primarily on rules engines and static thresholds, with obvious limitations in dealing with new money laundering approaches. In current technical practice, cross-border fund flow analysis is faced with the following technical challenges: The existing system generally adopts an isolated data processing mode. The traditional financial transaction data and the blockchain transaction data are respectively processed by independent systems to form an information island. Transaction monitoring systems of financial institutions such as banks mainly analyze structured transaction data, while blockchain data analysis systems concentrate on-chain transaction patterns, and an effective data association and feature fusion mechanism is lacked between the two systems. This split approach results in an inability to track the path of funds through the traditional financial system and blockchain network, leaving a blind area of supervision for money laundering. Existing risk assessment methods lack the ability to perform spatiotemporal joint analysis. Conventional risk scoring models are typically based on static analysis of transaction characteristics that fail to adequately account for the impact of time factors on risk determinations. In practical applications, transaction behaviors with the same characteristics may have completely different risk meanings at different time points, and the existing system cannot dynamically adjust the risk assessment strategy. Particularly in a cross-border money laundering scene, the time sensitivity of money transfer is very high, and the money laundering mode with time regularity is difficult to capture in time by the traditional system. Third, the prior art lacks an adaptive analysis granularity adjustment mechanism. Most monitoring systems employ fixed analysis windows and associated depths, either too coarse to report risk or too fine to produce a large number of false positives. For example, when analyzing the cross-border funds flow, a fixed use of a 30-day time window may not detect a fast funds transfer completed only within hours, and employing depth map analysis all the time may bring about huge computational overhead, affecting the system real-time. Fourth, the visual display capability is insufficient. The risk reports generated by the existing system are mostly presented in a two-dimensional chart form, complex circulation paths of funds in the traditional financial channel and blockchain network are difficult to clearly display, association relations of multiple dimensions such as time, amount and risk cannot be intuitively embodied, and the understanding difficulty and decision time of an analyst are increased. In particular, in the context of cross-border payments, money laundering often exhibits characteristics of decentralised transaction amounts, abnormally high transaction frequency, cross-link complexity of transaction paths, and increased speed of funds circulation. These features make it more difficult for conventional monitoring techniques to discover and pre-warn of risks in time. For example, blockchain transactions through a coinage service, or traditional bank transfers through multiple empty companies, may escape detection by existing systems. Therefore, a cross-border fund flow direction analysis scheme capable of integrating multi-source data, dynamically adjusting analysis strategies and considering timeliness and accuracy is needed to cope with increasingly complex backwash money supervision demands. Disclosure of Invention In order to solve the technical problems, the technical scheme solves the problems in the background technology by providing a cross-border fund flow direction data analysis method and system based on depth feature fusion. The embodiment of the application provides a cross-border fund flow direction data analysis method based on depth feature fusion, which is characterized by comprising the following steps of S1, obtaining financial transaction data and blockchain transaction data, carrying out feature extraction according to initial analysis granularity, obtaining a time sequence feature vector and a map feature vector, carrying out feature fusion, generating a fusion feature vector, S2, constructing a risk time sequence data neighborhood model, wherein a preset ideal point is used as a neighborhood center, constructing and obtaining a risk distanc