Search

CN-122022977-A - Multi-dimensional fund flow dynamic prediction analysis method

CN122022977ACN 122022977 ACN122022977 ACN 122022977ACN-122022977-A

Abstract

The invention provides a multi-dimensional fund flow dynamic prediction analysis method, which relates to the technical field of data processing, and comprises the steps of performing time tag standardization and delay correction processing on original data and aligning asynchronous data with different time granularities; the method comprises the steps of adaptively calculating the fluctuation rate, the periodic change rate and the cross correlation degree of each dimension data, extracting an influence factor, calculating time sequence dependence coefficients and interaction intensity between the dimensions, establishing a time sequence coupling mapping matrix, generating fund flow associated topology data, calculating the transmission intensity of any dimension data along a topology path when abnormal fluctuation occurs, calculating a time attenuation coefficient for correcting the transmission intensity in the topology path, dynamically adjusting the topology structure, detecting a high-sensitivity path, identifying a risk transmission source, tracing and positioning an abnormal chain, and generating fund flow risk trend prediction data.

Inventors

  • YU WEIBIN

Assignees

  • 杭州华奕智联科技有限公司

Dates

Publication Date
20260512
Application Date
20260129

Claims (10)

  1. 1. A method for multi-dimensional fund flow dynamic predictive analysis, the method comprising: Acquiring historical transaction data, industry economic indicators, policy event information, supply chain settlement records and market emotion data of a target account group as original data; Performing time tag standardization and delay correction processing on the original data, and generating time sequence matrix data after aligning asynchronous data with different time granularities; according to the time sequence matrix data, adaptively calculating the fluctuation rate, the periodic change rate and the cross correlation degree of each dimension data, extracting an influence factor capable of reflecting the multi-dimension dynamic characteristics, and generating multi-dimension influence factor data; according to the multidimensional influence factor data, calculating time sequence dependency coefficients and interaction strength among the dimensions, establishing a time sequence coupling mapping matrix, and generating fund flow association topology data; According to the fund flow associated topology data, when abnormal fluctuation occurs in any one dimension of data, calculating the conduction intensity along a topology path, and according to response delay between nodes, evaluating a time lag effect to obtain a time attenuation coefficient for correcting the conduction intensity in the topology path, dynamically adjusting the topology structure, and generating dynamic propagation data; And detecting a high-sensitivity path and identifying a risk conduction source according to the dynamic propagation data, tracing and positioning an abnormal chain to generate fund flow risk trend prediction data, and using the data to output a dynamic prediction result and a risk early warning signal of future fund flows of a target account group.
  2. 2. The method of claim 1, wherein adaptively calculating the fluctuation rate, the periodic change rate, and the cross correlation of each dimension data based on time series matrix data, extracting an influence factor capable of reflecting the dynamic characteristics of the multi-dimensions, and generating multi-dimension influence factor data, comprises: performing data smoothing and abnormal elimination processing on the time sequence matrix data to obtain time sequence characteristic data; calculating the variation amplitude of adjacent time periods in a time sliding window according to the time sequence characteristic data, and determining the fluctuation rate index of each dimension; According to the fluctuation rate index, carrying out pattern recognition on trend directions of each dimension in a plurality of period windows to generate periodic change rate data; Cross matching is carried out on the periodic change rate data and sequences with different time sequence dimensions, and the response degree under the time offset is analyzed to obtain cross-dimension cross correlation degree data; And weighting and fusing the fluctuation rate index, the periodic change rate data and the cross correlation degree data according to the dimension importance to generate multi-dimension influence factor data.
  3. 3. The method of claim 1, wherein calculating a time-series dependency coefficient and an interaction strength between dimensions according to the multi-dimensional influence factor data, and creating a time-series coupling mapping matrix, and generating the fund-flow association topology data, comprises: calculating the variation direction and amplitude difference of each dimension between adjacent time points according to the multidimensional influence factor data to obtain time sequence variation basic information; according to the time sequence change basic information, calculating the response intensity of each dimension to other dimension changes, and determining a time sequence dependence coefficient under time delay; Summarizing the time sequence dependence coefficients according to a dimension pairing mode, and calculating interaction duration and direction consistency among the dimensions to obtain interaction strength information; and establishing a time sequence coupling mapping matrix according to the interaction strength information and the time sequence dependence coefficient, removing a low-correlation path through a weight screening mechanism, and generating fund flow association topology data.
  4. 4. The method of claim 1, wherein according to the topology data associated with the fund flow, when any one dimension of data has abnormal fluctuation, calculating the conduction intensity along the topology path, and according to the response delay between nodes, evaluating the time lag effect to obtain a time attenuation coefficient for correcting the conduction intensity in the topology path, dynamically adjusting the topology structure, and generating dynamic propagation data, comprising: according to the fund flow associated topology data, monitoring the real-time change rate of the dimension corresponding to each node to form node state monitoring information; triggering abnormal propagation analysis according to the node state monitoring information when the change of any node exceeds a preset fluctuation threshold value, and extracting adjacent paths of the abnormal nodes to form path candidate information; According to the path candidate information, combining the variation amplitude of the abnormal node, the path weight and the response amplitude of the target node, and calculating the conduction intensity of the abnormal signal on each path; comparing the time difference of the state change of the adjacent nodes on the path according to the path candidate information to obtain response delay, and determining a time attenuation coefficient according to the ratio of the response delay to a preset time window; performing time attenuation correction on the conduction intensity of each node according to the time attenuation coefficient to form path propagation analysis information; Performing topology updating on paths with the transmission intensity exceeding a preset first transmission intensity threshold value in the path propagation analysis information, and adjusting the path weight and direction to form an updated topology structure; and generating dynamic propagation data according to the change trend among paths in the updated topological structure.
  5. 5. The method of claim 1, wherein detecting a highly sensitive path and identifying a risk conduction source based on dynamic propagation data, tracing an abnormal chain to locate, generating risk trend prediction data of the funds flow, comprises: screening paths with the conduction intensity higher than a preset second conduction intensity threshold value according to the dynamic transmission data to form a high-sensitivity path set; according to the time continuity and the direction consistency in the high-sensitivity path set, identifying chain nodes with continuous triggering abnormality to form risk chain information; Performing backward tracking on the risk chain information, determining the source node with the earliest abnormality, and generating risk conduction source positioning information; According to the risk conduction source positioning information, analyzing the fund flow diffusion rate and the influence range of the source node to the downstream node to form risk diffusion information; fitting the risk diffusion information with the historical time series trend to generate fund flow risk trend prediction data of a future time period.
  6. 6. The method of claim 2, wherein the step of weighting and fusing the volatility index, the periodic change rate data, and the cross correlation data according to the dimension importance to generate the multi-dimensional influence factor data comprises: Extracting initial fusion input data reflecting time sequence characteristics of different dimensions according to the fluctuation rate index, the periodic change rate data and the cross correlation degree data of each dimension; Performing sensitivity evaluation on the initial fusion input data, and determining an initial weight coefficient according to fluctuation activity degree and current change strength of each dimension in a historical time sequence to generate weighted input data; According to the weighted input data, performing dynamic adjustment on the weight coefficient, and when the fluctuation rate or the periodic variation mode of any dimension is enhanced, increasing the corresponding weight of the fluctuation rate or the periodic variation mode, and generating dynamic weight adjustment data; According to the dynamic weight adjustment data, redundant detection and correction are carried out on the cross correlation data, repeated contribution among high correlation dimensions is reduced, and redundant correction data are generated; performing time synchronization and fusion summarization on the redundant correction data to generate multi-dimensional influence factor vector data; and carrying out normalization and smoothing on the multidimensional influence factor vector data, eliminating feature scale differences among dimensions, and generating multidimensional influence factor data.
  7. 7. The method of claim 3, wherein establishing a time sequence coupling mapping matrix according to the interaction strength information and the time sequence dependency coefficient, and removing the low correlation path by a weight screening mechanism, generating the fund flow association topology data comprises: According to interaction strength information and time sequence dependency coefficients, layering mapping is carried out on the dependency coefficients and the interaction strength according to different attribute categories of macroscopic indexes, industry fund trends and enterprise behaviors, and multi-level mapping data are generated; According to the multi-level mapping data, calculating a master conduction weight between a macroscopic view and an industry layer and a slave conduction weight between the industry layer and an enterprise layer to form multi-level weight mapping matrix data; According to the multi-layer weight mapping matrix data, evaluating the time sequence stability of each path in a preset time window, and generating path stability data; according to the path stability data, performing dynamic adjustment on the path weight to form dynamic coupling mapping matrix data; and performing weight screening on the dynamic coupling mapping matrix data, removing the low-correlation path, optimizing the direction consistency, and generating the fund flow association topology data.
  8. 8. The method according to claim 4, wherein calculating the conduction intensity of the abnormal signal on each path according to the path candidate information by combining the variation amplitude of the abnormal node, the path weight and the response amplitude of the target node comprises: acquiring state change data of an initial node and a target node on a path according to the path candidate information to form path state input data; calculating the variation amplitude of the initial node according to the path state input data so as to reflect the initial energy of the abnormal signal and generate variation amplitude data; extracting corresponding path weights according to the path topological structure to represent basic conduction relations among nodes, and generating path weight data; determining the response amplitude of the target node according to the state change rate of the target node, and generating target response data; and carrying out multi-factor weighted fusion on the variation amplitude data, the path weight data and the target response data to generate path conduction intensity data.
  9. 9. The method of claim 4, wherein comparing the time differences of the state changes of the neighboring nodes on the path according to the path candidate information to obtain the response delay, and determining the time attenuation coefficient according to the ratio of the response delay to the preset time window comprises: Extracting state change time points of adjacent nodes on the path according to the path candidate information to form node time sequence data; Calculating state change time intervals between adjacent nodes according to the node time sequence data, and generating node response delay data; comparing the node response delay data with a preset time window, determining a delay proportion, and generating time delay ratio data; And according to the time delay ratio data, evaluating the time attenuation degree of the conduction signals on the paths, and calculating a time attenuation coefficient to form time attenuation coefficient data.
  10. 10. The multi-dimensional fund flow dynamic prediction analysis method of claim 5, wherein fitting risk spread information to historical time series trends generates fund flow risk trend prediction data for a future time period, comprising: extracting risk propagation rate data of a risk conduction source node and a downstream node thereof according to the risk diffusion information, and generating risk propagation input data; Extracting the change trend of the fund flow of each node in the historical time sequence according to the risk transmission input data to generate historical trend data; Performing time alignment and amplitude normalization on risk propagation input data and historical trend data to form comparable data; According to the comparable data, performing multi-period trend comparison in a preset time window, calculating the deviation degree between the short-term abnormal propagation trend and the long-term stable trend, and generating trend deviation data; and predicting the propagation direction and intensity of the risk in a future time period according to the trend deviation data, and generating fund flow risk trend prediction data.

Description

Multi-dimensional fund flow dynamic prediction analysis method Technical Field The invention relates to the technical field of data processing, in particular to a multidimensional fund flow dynamic prediction analysis method. Background Currently, the prior art performs predictive analysis of the funds flow based primarily on time series statistical models (e.g., ARIMA, VAR, or GARCH models) and partial machine learning algorithms (e.g., random forest, support vector regression). Such methods typically rely on historical transaction data of a single dimension as the primary input, such as account balance records, market volume or asset price fluctuations, and the like. The prediction process is based on fixed window historical data, and future fund flow changes are predicted by fitting time series trends. However, when the method processes multi-source and multi-dimensional coupling relation of fund flows in a complex economic system, dynamic modeling capability is often lacking, and nonlinear association among multi-dimensional factors such as macro economic indexes, industry fund trend, public opinion fluctuation, policy change and the like cannot be comprehensively reflected, so that the prediction result has insufficient precision under a multi-factor interaction scene. In practical application scenarios, for example, when business banks monitor mobility risk of enterprise account groups, the prior art often only performs short-term prediction based on historical account balances and transaction flows. Conventional models fail to dynamically capture the timing linkage of these cross-dimensional information when market incidents (such as industry credit policy adjustments or upstream supply chain fund breaks) cause abnormal fluctuations in some types of enterprise fund flows. In particular, the bank may still show that the enterprise funds flow is stable in the forecast results, but in practice, the enterprise funds chain may have entered a high risk state due to the superposition of asynchronous factors such as upstream settlement delays, downstream return delays, etc. Disclosure of Invention The invention aims to provide a multidimensional fund flow dynamic prediction analysis method, which aims to solve the problems in the background technology. In order to solve the technical problems, the technical scheme of the invention is as follows: A method of multi-dimensional fund flow dynamic predictive analysis, the method comprising: Acquiring historical transaction data, industry economic indicators, policy event information, supply chain settlement records and market emotion data of a target account group as original data; Performing time tag standardization and delay correction processing on the original data, and generating time sequence matrix data after aligning asynchronous data with different time granularities; according to the time sequence matrix data, adaptively calculating the fluctuation rate, the periodic change rate and the cross correlation degree of each dimension data, extracting an influence factor capable of reflecting the multi-dimension dynamic characteristics, and generating multi-dimension influence factor data; according to the multidimensional influence factor data, calculating time sequence dependency coefficients and interaction strength among the dimensions, establishing a time sequence coupling mapping matrix, and generating fund flow association topology data; According to the fund flow associated topology data, when abnormal fluctuation occurs in any one dimension of data, calculating the conduction intensity along a topology path, and according to response delay between nodes, evaluating a time lag effect to obtain a time attenuation coefficient for correcting the conduction intensity in the topology path, dynamically adjusting the topology structure, and generating dynamic propagation data; And detecting a high-sensitivity path and identifying a risk conduction source according to the dynamic propagation data, tracing and positioning an abnormal chain to generate fund flow risk trend prediction data, and using the data to output a dynamic prediction result and a risk early warning signal of future fund flows of a target account group. The scheme of the invention at least comprises the following beneficial effects: Firstly, the limitation of the prior art that the prior art depends on single historical data is broken through by constructing a multidimensional input system comprising historical transaction data, industry economic indexes, policy event information, supply chain settlement records and market emotion data. Through time tag standardization and delay correction, data with different time granularity are synchronized, and multisource dynamic characteristics of fund flows can be expressed uniformly on macroscopic, industrial and microscopic levels. Secondly, by adaptively calculating the fluctuation rate, the periodic change rate and the cross correlation degree of