CN116227940B - Enterprise fund flow anomaly detection method based on fund flow diagram - Google Patents
Enterprise fund flow anomaly detection method based on fund flow diagram Download PDFInfo
- Publication number
- CN116227940B CN116227940B CN202310490256.4A CN202310490256A CN116227940B CN 116227940 B CN116227940 B CN 116227940B CN 202310490256 A CN202310490256 A CN 202310490256A CN 116227940 B CN116227940 B CN 116227940B
- Authority
- CN
- China
- Prior art keywords
- fund
- node
- fund flow
- funds
- representing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/12—Accounting
- G06Q40/125—Finance or payroll
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Educational Administration (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Technology Law (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to an enterprise fund flow anomaly detection method based on a fund flow diagram, which comprises the following steps: acquiring financial fund flow data of enterprises and converting the characteristic value of the type of the financial fund flow data; splicing the converted data characteristics to obtain a fund flow direction data set; converting the fund flow direction data set into a topological graph by adopting a self-organizing feature mapping network to obtain a fund flow direction graph; and carrying out fund anomaly self-circulation detection and fund anomaly convergence detection according to the fund flow chart. The method can help enterprises to discover and correct abnormal funds flowing conditions in time, reduce financial risks of the enterprises and improve financial management level of the enterprises.
Description
Technical Field
The application relates to the technical field of enterprise wind control, in particular to an enterprise fund flow anomaly detection method based on a fund flow chart.
Background
Abnormal self circulation of the enterprise funds flow means that funds operation is continuously carried out by means of fictitious transaction and the like, so that the funds flow is continuously circulated, and a normal and actually abnormal funds flow is formed, and benefits are obtained. Abnormal convergence of enterprise funds flows refers to large amounts of funds that are surmounted from different sources into the same account or location over a period of time, with prominent anomalies relative to normal funds flow, possibly implying the existence of abnormal activity. Abnormal self-circulation and convergence of the fund flows can cause financial data distortion of enterprises, financial conditions of the enterprises cannot be truly reflected, risks are brought to investors and creditors, reputation and reputation of the enterprises are further destroyed, and long-term losses are brought to the enterprises.
The traditional enterprise fund flow supervision method mainly relies on financial statement, audit and other means to carry out manual detection, and is long in time consumption, and loopholes and false detection exist. Therefore, an intelligent detection method is needed to replace manual operation to efficiently detect abnormal enterprise funds flow.
Disclosure of Invention
Based on this, there is a need to provide a method for detecting abnormal funds flow of enterprises based on funds flow diagrams, the method comprising:
s1: acquiring financial fund flow data of enterprises and converting the characteristic value of the type of the financial fund flow data; splicing the converted data characteristics to obtain a fund flow direction data set;
s2: converting the fund flow direction data set into a topological graph by adopting a self-organizing feature mapping network to obtain a fund flow direction graph;
s3: carrying out fund anomaly self-circulation detection and fund anomaly convergence detection according to the fund flow diagram;
representing the fund flow graph as a adjacency matrix; calculating a normalized Laplace matrix based on the adjacency matrix, and taking the normalized Laplace matrix as an input of the diffusion map neural network; calculating self-circulation weights of all first nodes of the fund flow chart based on the feature matrix extracted by the last layer of the diffusion chart neural network; determining abnormal fund self-circulation conditions in the fund flow diagram according to the self-circulation weights;
converting the fund flow diagram into a directed diagram, and dividing all second nodes into a source node and a sink node according to the fund flow of each second node in the directed diagram; setting a super source node and a super sink node; connecting all source nodes to the super source node, and connecting all sink nodes to the super sink node; solving the maximum flow from the super source node to the super sink node; and determining abnormal fund convergence conditions in the fund flow diagram according to the maximum flow.
Preferably, in S1, the financial funds flow data of the enterprise includes a balance statement, a journal of running, and a cash journal; the types of the enterprise financial funds stream data include time information, amount information, account information, transaction type.
Preferably, in S1, the process of performing the eigenvalue conversion on the type of the financial funds stream data of the enterprise includes:
converting the time information into a 'year-month-day-week' form;
the amount information comprises an amount sample and an amount characteristic, and Min-Max normalization is carried out on the amount characteristic to obtain normalized firstiSample of the amount of money in the first placejThe value on the individual monetary value characteristic, the calculation formula is:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,representing normalized firstiSample of the amount of money in the first placejA value on the individual monetary feature;x ij represent the firstiSample of the amount of money in the first placejA value on the individual monetary feature;X j represent the firstjA value collection of the individual monetary features;
converting account information into category characteristics by adopting One-Hot coding;
the transaction type is coded as 0 or 1 using binary coding.
Preferably, in S2, further comprising obtaining a fund flow cluster:
step 1: initializing a weight vector of the self-organizing feature mapping network, wherein the weight vector is recorded asThe method comprises the steps of carrying out a first treatment on the surface of the Wherein the method comprises the steps ofmRepresenting the number of nodes of the ad hoc feature mapping network,pthe length of the weight vector of each node is represented;
step 2: for each sample of the fund flow direction in the fund flow direction dataset, calculating its Euclidean distance from a node of the respective organizational feature mapping network; and find the winning node, record the winning node asJThe method comprises the steps of carrying out a first treatment on the surface of the The calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,krepresent the firstkThe number of nodes in the network is,Z a represent the firstaA sample of the individual funds flow directions;W k represent the firstkA weight vector of each node;represents an L2 norm;
step 3: updating the weight vector, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,W k (t+1) represents the updated weight vector;trepresenting the current iteration round number;α(t) Representing a learning rate;h kJ (t) Representing a gaussian function centered on the winning node;
step 4: iterating the step 2 and the step 3 until the self-organizing feature mapping network converges; after the network converges, taking each neuron of the self-organizing feature mapping network as a fund flow cluster; the fund flow cluster is used to construct the fund flow graph.
Preferably, in S2, the process of obtaining the fund flow chart is:
calculating Euclidean distance between each fund flow direction sample and each fund flow direction cluster; dividing each fund flow direction sample into the fund flow direction clusters closest to the fund flow direction sample based on the calculated Euclidean distance, and constructing the fund flow direction diagram by taking the fund flow direction sample as a first node; the fund flow chart is recorded asG=(V,E) The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the liquid crystal display device comprises a liquid crystal display device,Vrepresenting the set of all funds flows to the first node contained in the cluster,Erepresenting a set of edges between any two first nodes;
the fund flow diagram is a weighted directed diagram, and the weight of the edge is the Euclidean distance between the fund flow clusters corresponding to the first node.
Preferably, in S3, the fund flow graph is represented as an adjacency matrix, and the adjacency matrix is expressed as:Athe method comprises the steps of carrying out a first treatment on the surface of the And calculating a normalized Laplace matrix based on the adjacency matrix, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,Hrepresenting a normalized Laplace matrix;Irepresenting the identity matrix;Da degree matrix representing the adjacency matrix;
taking the normalized Laplace matrix as the input of the diffusion map neural network, and extracting the characteristics by updating the information propagation of multiple layers in the diffusion map neural network; the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,H l(+1) represent the firstlA feature matrix of +1 layer;σ(. Cndot.) represents the activation function,representing the sum of the adjacency matrix and the identity matrix; />Is a diagonal matrix;H l() represent the firstlA feature matrix of the layer;W l() represent the firstlA weight matrix of the layer;；L-1 represents the number of layers of the diffusion map neural network;
calculating self-circulation weights of all nodes of the fund flow chart based on the characteristics extracted from the last layer of the diffusion chart neural network; the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,represent the firstqSelf-circulating weights of the first nodes;nrepresenting a total number of first nodes; />Representing the diffusion map neural network at the firstLIn the feature matrix of the layerqLine 1pElements of a column; />Represent the firstqElements in the self-adjacency matrix of the individual nodes; />Representing the diffusion map neural network at the firstLIn the feature matrix of the layerpLine 1qColumn elements.
Preferably, in S3, each of the second nodes in the directed graph is a cluster of funds flow directions, and the weight value of each edge in the directed graph represents the funds flow from one second node to another second node; taking the second node into which the funds flow as a sink node, and taking the second node out of which the funds flow as a source node;
solving the maximum flow from the super source node to the super sink node by adopting a maximum flow algorithm, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,F max representing the maximum traffic from the supersource node to the supersink node,maxFlow() Representing a maximum flow algorithm;G 1 representing a directed graph comprising super source nodes and super sink nodes;s source representing a super source node;s sink representing a super sink node.
Preferably, in S3, a first threshold value and a second threshold value are further set; the first threshold is used for determining abnormal self-circulation conditions of funds in the fund flow diagram; the second threshold is used for determining abnormal fund convergence conditions in the fund flow diagram;
when the self-circulation weight exceeds the first threshold value, judging that a first node corresponding to the self-circulation weight has a fund abnormal self-circulation condition; otherwise, it does not exist;
when the maximum flow is smaller than the second threshold, judging that abnormal fund convergence exists in the fund flow chart; otherwise, it does not exist.
Preferably, the method further comprises the step of calculating the risk level of the enterprise fund flow according to the detected abnormal self-circulation of the funds, the abnormal convergence quantity of the funds and the fund scale, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,Riskrepresenting an enterprise funds flow risk level value;N loop representing the number of abnormal self-circulation of funds;S u represent the firstuThe fund scale of the abnormal self-circulation of the individual funds;N con representing the number of abnormal funds collections;T v represent the firstvThe size of the funds which are abnormally converged.
Preferably, in S3, the method further includes:
acquiring a real weight label; calculating a loss value between the real weight label and the self-circulation weight by adopting a mean square error loss function; and optimizing network parameters of the diffusion map neural network by adopting a gradient descent back propagation algorithm based on the loss value.
The beneficial effects are that: the method can help enterprises to discover and correct abnormal funds flowing conditions in time, reduce financial risks of the enterprises and improve financial management level of the enterprises.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of an enterprise funds flow anomaly detection method according to an embodiment of the present application.
Detailed Description
In order to make the above objects, features and advantages of the present application more comprehensible, embodiments accompanied with figures are described in detail below. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. This application is, however, susceptible of embodiment in many other forms than those described herein and similar modifications can be made by those skilled in the art without departing from the spirit of the application, and therefore the application is not to be limited to the specific embodiments disclosed below.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "plurality" is at least two, such as two, three, etc., unless explicitly defined otherwise.
As shown in fig. 1, the present embodiment provides a method for detecting abnormal funds flow of an enterprise based on a funds flow diagram, which includes:
s1: acquiring financial fund flow data of enterprises and converting the characteristic value of the type of the financial fund flow data; splicing the converted data characteristics to obtain a fund flow direction data set;
in this embodiment, the enterprise financial funds flow data includes a balance statement, a journal of running, a cash journal; the types of the enterprise financial funds stream data include time information, amount information, account information, transaction type.
Specifically, the process of performing eigenvalue conversion on the type of the financial funds flow data of the enterprise includes:
converting the time information into a 'year-month-day-week' form;
the money information comprises money samples and money characteristics, and the money characteristics are subjected to Min-Max normalization to obtain normalized firstiSample of the amount of money in the first placejThe value on the individual monetary value characteristic, the calculation formula is:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,representing normalized firstiSample of the amount of money in the first placejA value on the individual monetary feature;x ij represent the firstiSample of the amount of money in the first placejA value on the individual monetary feature;X j represent the firstjA value collection of the individual monetary features;
converting account information into category characteristics by adopting One-Hot coding;
the transaction type is coded as 0 or 1 using binary coding.
S2: converting the fund flow direction data set into a topological graph by adopting a self-organizing feature mapping network to obtain a fund flow direction graph;
in this embodiment, the step further includes obtaining a fund flow cluster:
step 1: initializing a weight vector of the self-organizing feature mapping network, wherein the weight vector is recorded asThe method comprises the steps of carrying out a first treatment on the surface of the Wherein the method comprises the steps ofmRepresenting the number of nodes of the ad hoc feature mapping network,pthe length of the weight vector of each node is represented;
step 2: for each sample of the fund flow direction in the fund flow direction dataset, calculating its Euclidean distance from a node of the respective organizational feature mapping network; and find the winning node, record the winning node asJThe method comprises the steps of carrying out a first treatment on the surface of the The calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,krepresent the firstkThe number of nodes in the network is,Z a represent the firstaA sample of the individual funds flow directions;W k represent the firstkA weight vector of each node;represents an L2 norm;
step 3: updating the weight vector, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,W k (t+1) represents the updated weight vector;trepresenting the current iteration round number;α(t) Representing a learning rate;h kJ (t) Representing a Gaussian function centered at the winning node, the value of which is atJThe vicinity is largest and gradually decreases with increasing distance. Updated weight vectorW k (t+1) will be closer toZ a 。
Step 4: iterating the step 2 and the step 3 until the self-organizing feature mapping network converges; after the network converges, taking each neuron of the self-organizing feature mapping network as a fund flow cluster; the fund flow cluster is used to construct the fund flow graph.
The specific process of obtaining the fund flow chart is as follows:
calculating Euclidean distance between each fund flow direction sample and each fund flow direction cluster; dividing each fund flow direction sample into the fund flow direction clusters closest to the fund flow direction sample based on the calculated Euclidean distance, and constructing the fund flow direction diagram by taking the fund flow direction sample as a first node; the fund flow chart is recorded asG=(V,E) The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the liquid crystal display device comprises a liquid crystal display device,Vrepresenting the set of all funds flows to the first node contained in the cluster,Erepresenting a set of edges between any two first nodes;
the fund flow diagram is a weighted directed diagram, and the weight of the edge is the Euclidean distance between the fund flow clusters corresponding to the first node; the edge weight between first nodes in the same cluster of funds flows is 0.
S3: carrying out fund anomaly self-circulation detection and fund anomaly convergence detection according to the fund flow diagram;
specifically, the fund flow diagram is expressed as an adjacency matrix; the adjacency matrix is noted as:Athe method comprises the steps of carrying out a first treatment on the surface of the And calculating a normalized Laplace matrix based on the adjacency matrix, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,Hrepresenting a normalized Laplace matrix;Irepresenting the identity matrix;Da degree matrix representing the adjacency matrix;
taking the normalized Laplace matrix as the input of the diffusion map neural network, and extracting the characteristics by updating the information propagation of multiple layers in the diffusion map neural network; the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,H l(+1) represent the firstlA feature matrix of +1 layer;σ(. Cndot.) represents the activation function,representing the sum of the adjacency matrix and the identity matrix; />Is a diagonal matrix;H l() represent the firstlA feature matrix of the layer;W l() represent the firstlA weight matrix of the layer;；L-1 represents the number of layers of the diffusion map neural network;
calculating self-circulation weights of all nodes of the fund flow chart based on the characteristics extracted from the last layer of the diffusion chart neural network; the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,represent the firstqSelf-circulating weights of the first nodes;nrepresenting a total number of first nodes; />Representing the diffusion map neural network at the firstLIn the feature matrix of the layerqLine 1pElements of a column; />Represent the firstqElements in the self-adjacency matrix of the individual nodes; />Representing the diffusion map neural network at the firstLIn the feature matrix of the layerpLine 1qColumn elements.
Further, the method also comprises the steps of obtaining a real weight label; calculating a loss value between the real weight label and the self-circulation weight by adopting a mean square error loss function; and optimizing network parameters of the diffusion map neural network by adopting a gradient descent back propagation algorithm based on the loss value.
Converting the fund flow diagram into a directed diagram, wherein each second node in the directed diagram is a fund flow diagram cluster, and the weight of each edge in the directed diagram represents the fund flow from one second node to another second node; taking the second node into which the funds flow as a sink node, and taking the second node out of which the funds flow as a source node;
solving the maximum flow from the super source node to the super sink node by adopting a maximum flow algorithm, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,F max representing the maximum traffic from the supersource node to the supersink node,maxFlow() Representing a maximum flow algorithm;G 1 representation comprising super source nodes and super sink nodesA directed graph of points;s source representing a super source node;s sink representing a super sink node.
In the present embodiment, a first threshold value and a second threshold value are set; the first threshold is used for determining abnormal self-circulation conditions of funds in the fund flow diagram; the second threshold is used for determining abnormal fund convergence conditions in the fund flow diagram;
when the self-circulation weight exceeds the first threshold value, judging that a first node corresponding to the self-circulation weight has a fund abnormal self-circulation condition; otherwise, it does not exist;
when the maximum flow is smaller than the second threshold, the preset flow cannot be transmitted between the source node and the sink node, which means that the flow of the edges on some paths exceeds the second threshold, namely, the abnormal convergence condition of funds in the fund flow diagram is judged; otherwise, it does not exist.
The method provided by the embodiment further comprises the steps of calculating the risk level of the enterprise fund flow according to the detected abnormal self-circulation of the funds, the abnormal convergence quantity of the funds and the fund scale, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,Riskrepresenting an enterprise funds flow risk level value;N loop representing the number of abnormal self-circulation of funds;S u represent the firstuThe fund scale of the abnormal self-circulation of the individual funds;N con representing the number of abnormal funds collections;T v represent the firstvThe size of the funds which are abnormally converged. The first term represents the sum of the fund scales of the abnormal fund self-circulation, and the second term represents the combination of the fund scales of the abnormal fund convergence, and the risks brought to enterprises by the abnormal fund self-circulation and the abnormal fund convergence are respectively reflected. And calculating the risk level value of the enterprise fund flow according to the formula, and making corresponding measures to reduce the risk. For example: risk review of transactions suspected of abnormal funds flow, or for businessesThe fund flow is regulated and improved to reduce the probability of abnormal self-circulation of funds and abnormal convergence of funds.
The technical features of the above-described embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above-described embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the claims. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.
Claims (8)
1. The enterprise fund flow anomaly detection method based on the fund flow diagram is characterized by comprising the following steps of:
s1: acquiring financial fund flow data of enterprises and converting the characteristic value of the type of the financial fund flow data; splicing the converted data characteristics to obtain a fund flow direction data set;
s2: converting the fund flow direction data set into a topological graph by adopting a self-organizing feature mapping network to obtain a fund flow direction graph;
s3: carrying out fund anomaly self-circulation detection and fund anomaly convergence detection according to the fund flow diagram;
representing the fund flow graph as a adjacency matrix; calculating a normalized Laplace matrix based on the adjacency matrix, and taking the normalized Laplace matrix as an input of the diffusion map neural network; calculating self-circulation weights of all first nodes of the fund flow chart based on the feature matrix extracted by the last layer of the diffusion chart neural network; determining abnormal fund self-circulation conditions in the fund flow diagram according to the self-circulation weights;
converting the fund flow diagram into a directed diagram, and dividing all second nodes into a source node and a sink node according to the fund flow of each second node in the directed diagram; setting a super source node and a super sink node; connecting all source nodes to the super source node, and connecting all sink nodes to the super sink node; solving the maximum flow from the super source node to the super sink node; determining abnormal fund convergence conditions in a fund flow chart according to the maximum flow;
each second node in the directed graph is a fund flow direction cluster, and the weight of each edge in the directed graph represents the fund flow from one second node to another second node; taking the second node into which the funds flow as a sink node, and taking the second node out of which the funds flow as a source node;
solving the maximum flow from the super source node to the super sink node by adopting a maximum flow algorithm, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,F max representing the maximum traffic from the supersource node to the supersink node,maxFlow() Representing a maximum flow algorithm;G 1 representing a directed graph comprising super source nodes and super sink nodes;s source representing a super source node;s sink representing a super sink node;
a first threshold value and a second threshold value are set; the first threshold is used for determining abnormal self-circulation conditions of funds in the fund flow diagram; the second threshold is used for determining abnormal fund convergence conditions in the fund flow diagram;
when the self-circulation weight exceeds the first threshold value, judging that a first node corresponding to the self-circulation weight has a fund abnormal self-circulation condition; otherwise, it does not exist;
when the maximum flow is smaller than the second threshold, judging that abnormal fund convergence exists in the fund flow chart; otherwise, it does not exist.
2. The method for detecting abnormal flow of funds in enterprises according to claim 1, wherein in S1, the financial flow data of finance in enterprises comprises balance details, journal of line, cash journal; the types of the enterprise financial funds stream data include time information, amount information, account information, transaction type.
3. The method for detecting abnormal flow of funds in enterprises according to claim 2, wherein in S1, the process of converting the characteristic value of the type of the financial flow data of enterprises comprises:
converting the time information into a 'year-month-day-week' form;
the amount information comprises an amount sample and an amount characteristic, and Min-Max normalization is carried out on the amount characteristic to obtain normalized firstiSample of the amount of money in the first placejThe value on the individual monetary value characteristic, the calculation formula is:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,representing normalized firstiSample of the amount of money in the first placejA value on the individual monetary feature;x ij represent the firstiSample of the amount of money in the first placejA value on the individual monetary feature;X j represent the firstjA value collection of the individual monetary features;
converting account information into category characteristics by adopting One-Hot coding;
the transaction type is coded as 0 or 1 using binary coding.
4. The method for detecting abnormal funds flow of enterprises according to claim 1, wherein in S2, further comprising obtaining a funds flow cluster:
step 1: initializing a weight vector of the self-organizing feature mapping network, wherein the weight vector is recorded asThe method comprises the steps of carrying out a first treatment on the surface of the Wherein the method comprises the steps ofmRepresentation ad hocThe number of nodes of the texture feature map network,pthe length of the weight vector of each node is represented;
step 2: for each sample of the fund flow direction in the fund flow direction dataset, calculating its Euclidean distance from a node of the respective organizational feature mapping network; and find the winning node, record the winning node asJThe method comprises the steps of carrying out a first treatment on the surface of the The calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,krepresent the firstkThe number of nodes in the network is,Z a represent the firstaA sample of the individual funds flow directions;W k represent the firstkA weight vector of each node;represents an L2 norm;
step 3: updating the weight vector, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,W k (t+1) represents the updated weight vector;trepresenting the current iteration round number;α(t) Representing a learning rate;h kJ (t) Representing a gaussian function centered on the winning node;
step 4: iterating the step 2 and the step 3 until the self-organizing feature mapping network converges; after the network converges, taking each neuron of the self-organizing feature mapping network as a fund flow cluster; the fund flow cluster is used to construct the fund flow graph.
5. The method for detecting abnormal funds flow of enterprises according to claim 4, wherein in S2, the process of obtaining the funds flow chart is as follows:
calculating Euclidean distance between each fund flow direction sample and each fund flow direction cluster; and based on the calculated Euclidean distance, each of theDividing the fund flow direction sample into the fund flow direction clusters closest to the fund flow direction sample, and constructing the fund flow direction diagram by taking the fund flow direction sample as a first node; the fund flow chart is recorded asG=(V,E) The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the liquid crystal display device comprises a liquid crystal display device,Vrepresenting the set of all funds flows to the first node contained in the cluster,Erepresenting a set of edges between any two first nodes;
the fund flow diagram is a weighted directed diagram, and the weight of the edge is the Euclidean distance between the fund flow clusters corresponding to the first node.
6. The method of claim 5, wherein in S3, the fund flow graph is represented as a adjacency matrix, and the adjacency matrix is expressed as:Athe method comprises the steps of carrying out a first treatment on the surface of the And calculating a normalized Laplace matrix based on the adjacency matrix, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,Hrepresenting a normalized Laplace matrix;Irepresenting the identity matrix;Da degree matrix representing the adjacency matrix;
taking the normalized Laplace matrix as the input of the diffusion map neural network, and extracting the characteristics by updating the information propagation of multiple layers in the diffusion map neural network; the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,H l(+1) represent the firstlA feature matrix of +1 layer;σ(. Cndot.) represents the activation function,representing the sum of the adjacency matrix and the identity matrix; />Is a diagonal matrix;H l() represent the firstlA feature matrix of the layer;W l() represent the firstlA weight matrix of the layer;；L-1 represents the number of layers of the diffusion map neural network;
calculating self-circulation weights of all nodes of the fund flow chart based on the characteristics extracted from the last layer of the diffusion chart neural network; the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,represent the firstqSelf-circulating weights of the first nodes;nrepresenting a total number of first nodes; />Representing the diffusion map neural network at the firstLIn the feature matrix of the layerqLine 1pElements of a column; />Represent the firstqElements in the self-adjacency matrix of the individual nodes; />Representing the diffusion map neural network at the firstLIn the feature matrix of the layerpLine 1qColumn elements.
7. The method for detecting abnormal flow of funds in enterprises according to claim 1, further comprising calculating the risk level of the flow of funds in enterprises according to the detected abnormal self-circulation of funds and the number and the scale of funds of abnormal convergence, wherein the calculation formula is as follows:
；
wherein, the liquid crystal display device comprises a liquid crystal display device,Riskrepresenting an enterprise funds flow risk level value;N loop representing the number of abnormal self-circulation of funds;S u represent the firstuThe fund scale of the abnormal self-circulation of the individual funds;N con representing the number of abnormal funds collections;T v represent the firstvThe size of the funds which are abnormally converged.
8. The method for detecting abnormal flow of funds in an enterprise according to claim 6, wherein in S3, further comprising:
acquiring a real weight label; calculating a loss value between the real weight label and the self-circulation weight by adopting a mean square error loss function; and optimizing network parameters of the diffusion map neural network by adopting a gradient descent back propagation algorithm based on the loss value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310490256.4A CN116227940B (en) | 2023-05-04 | 2023-05-04 | Enterprise fund flow anomaly detection method based on fund flow diagram |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310490256.4A CN116227940B (en) | 2023-05-04 | 2023-05-04 | Enterprise fund flow anomaly detection method based on fund flow diagram |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116227940A CN116227940A (en) | 2023-06-06 |
| CN116227940B true CN116227940B (en) | 2023-07-25 |
Family
ID=86587547
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310490256.4A Active CN116227940B (en) | 2023-05-04 | 2023-05-04 | Enterprise fund flow anomaly detection method based on fund flow diagram |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116227940B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117520441B (en) * | 2024-01-03 | 2024-03-15 | 国网浙江省电力有限公司金华供电公司 | Method, device and equipment for detecting abnormity of fund flow data |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022651A (en) * | 2016-06-14 | 2016-10-12 | 深圳市迪博企业风险管理技术有限公司 | Risk early warning method based on business attribute and index system |
| CN108280755A (en) * | 2018-02-28 | 2018-07-13 | 阿里巴巴集团控股有限公司 | The recognition methods of suspicious money laundering clique and identification device |
| CN109146669A (en) * | 2018-08-24 | 2019-01-04 | 阿里巴巴集团控股有限公司 | Detection method, device and the server of abnormal funds transfer plan |
| CN109345221A (en) * | 2018-09-25 | 2019-02-15 | 阿里巴巴集团控股有限公司 | The checking method and device of resource circulation |
| CN110033360A (en) * | 2019-03-11 | 2019-07-19 | 阿里巴巴集团控股有限公司 | Method for detecting abnormality, device and the equipment of fund flow data |
| TWM598995U (en) * | 2020-04-16 | 2020-07-21 | 彰化商業銀行股份有限公司 | Cash flow account analysis and warning system |
| CN113935832A (en) * | 2021-09-29 | 2022-01-14 | 光大科技有限公司 | Abnormal behavior detection processing method and device |
| CN113962614A (en) * | 2021-12-21 | 2022-01-21 | 深圳市迪博企业风险管理技术有限公司 | Intelligent examination method and device for business abnormity of listed company |
| CN113988724A (en) * | 2021-12-28 | 2022-01-28 | 深圳市迪博企业风险管理技术有限公司 | Risk analysis method for financial activity knowledge graph of listed company |
| CN114581233A (en) * | 2022-03-02 | 2022-06-03 | 中国工商银行股份有限公司 | Money laundering account determining method, apparatus, device, storage medium and program product |
| CN114581230A (en) * | 2022-01-14 | 2022-06-03 | 中国科学院计算技术研究所 | Money laundering behavior detection method, device and medium in flow chart |
| CN115099825A (en) * | 2022-06-28 | 2022-09-23 | 中国工商银行股份有限公司 | Abnormal fund flow account identification method and device |
| CN115170319A (en) * | 2022-07-25 | 2022-10-11 | 中国工商银行股份有限公司 | Abnormal account detection method, and method and device for constructing graph neural network model |
| CN115271733A (en) * | 2022-09-28 | 2022-11-01 | 深圳市迪博企业风险管理技术有限公司 | Privacy-protecting block chain transaction data anomaly detection method and equipment |
| CN115456078A (en) * | 2022-09-15 | 2022-12-09 | 中国工商银行股份有限公司 | Capital flow direction classification method, device, equipment, medium and product |
| CN115905427A (en) * | 2022-12-16 | 2023-04-04 | 中国建设银行股份有限公司 | Capital flow direction abnormity identification method, device, equipment and computer storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111149122B (en) * | 2018-05-30 | 2023-10-10 | 重庆小雨点小额贷款有限公司 | Block chain-based security management method, related device and storage medium |
-
2023
- 2023-05-04 CN CN202310490256.4A patent/CN116227940B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022651A (en) * | 2016-06-14 | 2016-10-12 | 深圳市迪博企业风险管理技术有限公司 | Risk early warning method based on business attribute and index system |
| CN108280755A (en) * | 2018-02-28 | 2018-07-13 | 阿里巴巴集团控股有限公司 | The recognition methods of suspicious money laundering clique and identification device |
| CN109146669A (en) * | 2018-08-24 | 2019-01-04 | 阿里巴巴集团控股有限公司 | Detection method, device and the server of abnormal funds transfer plan |
| CN109345221A (en) * | 2018-09-25 | 2019-02-15 | 阿里巴巴集团控股有限公司 | The checking method and device of resource circulation |
| CN110033360A (en) * | 2019-03-11 | 2019-07-19 | 阿里巴巴集团控股有限公司 | Method for detecting abnormality, device and the equipment of fund flow data |
| TWM598995U (en) * | 2020-04-16 | 2020-07-21 | 彰化商業銀行股份有限公司 | Cash flow account analysis and warning system |
| CN113935832A (en) * | 2021-09-29 | 2022-01-14 | 光大科技有限公司 | Abnormal behavior detection processing method and device |
| CN113962614A (en) * | 2021-12-21 | 2022-01-21 | 深圳市迪博企业风险管理技术有限公司 | Intelligent examination method and device for business abnormity of listed company |
| CN113988724A (en) * | 2021-12-28 | 2022-01-28 | 深圳市迪博企业风险管理技术有限公司 | Risk analysis method for financial activity knowledge graph of listed company |
| CN114581230A (en) * | 2022-01-14 | 2022-06-03 | 中国科学院计算技术研究所 | Money laundering behavior detection method, device and medium in flow chart |
| CN114581233A (en) * | 2022-03-02 | 2022-06-03 | 中国工商银行股份有限公司 | Money laundering account determining method, apparatus, device, storage medium and program product |
| CN115099825A (en) * | 2022-06-28 | 2022-09-23 | 中国工商银行股份有限公司 | Abnormal fund flow account identification method and device |
| CN115170319A (en) * | 2022-07-25 | 2022-10-11 | 中国工商银行股份有限公司 | Abnormal account detection method, and method and device for constructing graph neural network model |
| CN115456078A (en) * | 2022-09-15 | 2022-12-09 | 中国工商银行股份有限公司 | Capital flow direction classification method, device, equipment, medium and product |
| CN115271733A (en) * | 2022-09-28 | 2022-11-01 | 深圳市迪博企业风险管理技术有限公司 | Privacy-protecting block chain transaction data anomaly detection method and equipment |
| CN115905427A (en) * | 2022-12-16 | 2023-04-04 | 中国建设银行股份有限公司 | Capital flow direction abnormity identification method, device, equipment and computer storage medium |
Non-Patent Citations (2)
| Title |
|---|
| "基于信息披露的内部控制指数研究";林斌等;《会计研究》;第12-20页 * |
| "基于行为金融视角的金融供应链风险管理研究";宋佳栋;《中国优秀硕士学位论文全文数据库 经济与管理科学辑》;J145-421 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116227940A (en) | 2023-06-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bahnsen et al. | Example-dependent cost-sensitive logistic regression for credit scoring | |
| US8296205B2 (en) | Connecting decisions through customer transaction profiles | |
| US11263644B2 (en) | Systems and methods for detecting unauthorized or suspicious financial activity | |
| CN112633426B (en) | Method and device for processing data class imbalance, electronic equipment and storage medium | |
| Wang et al. | Multi-classification assessment of bank personal credit risk based on multi-source information fusion | |
| Zelenkov | Example-dependent cost-sensitive adaptive boosting | |
| CN112700324A (en) | User loan default prediction method based on combination of Catboost and restricted Boltzmann machine | |
| CN116227940B (en) | Enterprise fund flow anomaly detection method based on fund flow diagram | |
| AU2019101189A4 (en) | A financial mining method for credit prediction | |
| CN114187112A (en) | Training method of account risk model and determination method of risk user group | |
| Lei et al. | Research on supply chain financial risk prevention based on machine learning | |
| Huang et al. | Enterprise risk assessment based on machine learning | |
| CN111563187A (en) | Relationship determination method, device and system and electronic equipment | |
| Mukherjee et al. | Detection of defaulters in P2P lending platforms using unsupervised learning | |
| CN112329862A (en) | Decision tree-based anti-money laundering method and system | |
| CN112419045A (en) | Unbalanced credit user classification method based on oversampling and random forest | |
| CN116862658A (en) | Credit evaluation method, apparatus, electronic device, medium and program product | |
| CN116227939A (en) | Enterprise credit rating method and device based on graph convolution neural network and EM algorithm | |
| CN116485406A (en) | Account detection method and device, storage medium and electronic equipment | |
| CN114818999B (en) | Account identification method and system based on self-encoder and generation countermeasure network | |
| Naik | Predicting credit risk for unsecured lending: A machine learning approach | |
| CN115496205A (en) | Detection model training method, data detection method, device, equipment and storage medium | |
| Datkhile et al. | Statistical modelling on loan default prediction using different models | |
| CN115271442A (en) | Modeling method and system for evaluating enterprise growth based on natural language | |
| CN115204457A (en) | Loan default risk prediction method based on graph attention network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |