CN115952240A - Financial data compliance examination method and device based on block chain - Google Patents
Financial data compliance examination method and device based on block chain Download PDFInfo
- Publication number
- CN115952240A CN115952240A CN202310248208.4A CN202310248208A CN115952240A CN 115952240 A CN115952240 A CN 115952240A CN 202310248208 A CN202310248208 A CN 202310248208A CN 115952240 A CN115952240 A CN 115952240A
- Authority
- CN
- China
- Prior art keywords
- financial
- financial data
- compliance
- data
- correlation
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 230000008859 change Effects 0.000 claims abstract description 111
- 238000012550 audit Methods 0.000 claims abstract description 95
- 238000013528 artificial neural network Methods 0.000 claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 32
- 230000007246 mechanism Effects 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 41
- 238000011176 pooling Methods 0.000 claims description 20
- 238000007689 inspection Methods 0.000 claims description 17
- 238000012552 review Methods 0.000 claims description 17
- 230000000977 initiatory effect Effects 0.000 claims description 15
- 238000010276 construction Methods 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 230000001902 propagating effect Effects 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 6
- 230000004075 alteration Effects 0.000 claims description 5
- 230000004044 response Effects 0.000 claims 1
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of block chains and compliance internal control, and discloses a financial data compliance examination method and device based on block chains, wherein the method comprises the following steps: constructing a multi-department financial block chain, and chaining financial change data blocks based on a consensus mechanism; acquiring financial data indexes and physical addresses of financial change data blocks, and constructing a query index table; deploying all nodes in a block chain, and acquiring target financial data from all financial data change blocks by using a query index table; generating a financial data checking association cube model according to the target financial data; constructing a three-dimensional time sequence convolution neural network, and training on a preset training set to obtain a financial data compliance audit model; and after the financial data auditing association cube model is expanded, inputting a financial data compliance audit model to obtain a compliance audit result of the target financial data. The invention realizes the automatic, high-efficiency and high-accuracy compliance examination of the dynamic financial data in different time periods.
Description
Technical Field
The invention relates to the technical field of block chains and internal control compliance, in particular to a financial data compliance examination method and device based on a block chain.
Background
Financial compliance is one of the important links of control compliance in the enterprise, and financial compliance has very important meaning to management and control enterprise operation risk. The traditional financial data compliance examination method is generally used for carrying out compliance examination on annual financial data, and due to the fact that the data volume of the annual financial data is large, the compliance examination time is long, and the compliance examination results are all the same, so that the traditional method based on static financial data is difficult to effectively monitor and carry out compliance examination on the dynamic changing process of financial data when the financial data are changed frequently and carefully and the financial data are changed mistakenly or are not trusted.
To ensure the validity of the financial data, block-chain techniques may be employed to ensure that the financial detail data is not tamper-able and non-repudiatable. However, the current general blockchain technology does not support block query indexing, and needs to search block data in a full traversal manner, so that it is difficult to implement efficient on-chain query for frequently-changing financial detail data.
Disclosure of Invention
Based on this, the present invention is directed to solve at least one technical problem in the background art, and thereby provides a block chain-based financial data compliance review method and apparatus.
In order to solve the above problem, an embodiment of the present invention provides a financial data compliance examination method based on a block chain, including:
constructing a multi-department financial block chain, and chaining financial change data blocks based on a consensus mechanism;
acquiring financial data indexes and physical addresses of the financial change data blocks to construct a query index table;
deploying all nodes in the multi-department financial block chain, and acquiring target financial data from all the financial data change blocks by using the query index table;
generating a financial data checking association cube model according to the target financial data;
constructing a three-dimensional time sequence convolution neural network, and training the three-dimensional time sequence convolution neural network on a preset training set to obtain a financial data compliance examination model;
and after expanding the financial data auditing association cube model, inputting the financial data compliance audit model to obtain a compliance audit result of the target financial data.
Optionally, the constructing a three-dimensional time series convolutional neural network, and training the three-dimensional time series convolutional neural network on a preset training set to obtain a financial data compliance audit model, includes:
constructing a three-dimensional time sequence convolution neural network comprising an input layer, a three-dimensional convolution layer, a pooling layer, a full-connection layer and an output layer;
obtaining a checking correlation cube model and a compliance audit real label of a financial data sample, and constructing a training set;
expanding the checking association cube model of the financial data sample to obtain a plurality of checking association matrixes in different time periods;
inputting a plurality of audit incidence matrixes into the three-dimensional time sequence convolution neural network, extracting the independent characteristic of each audit incidence matrix and the correlation characteristic between the audit incidence matrixes through the three-dimensional time sequence convolution neural network, and obtaining the audit compliance probability value of the financial data sample according to the independent characteristic and the correlation characteristic;
obtaining a loss function according to the review compliance probability value and the compliance review true label;
acquiring the gradient between a full connection layer and an output layer in the three-dimensional time sequence convolution neural network according to the loss function;
and reversely propagating the gradient by a gradient descent method, optimizing weight parameters between the full connection layer and the output layer, and obtaining a financial data compliance review model.
Optionally, the three-dimensional convolutional layer comprises first convolutional kernels with the size of 3 × 3 and second convolutional kernels with the size of 1 × 10, the number of the first convolutional kernels and the number of the second convolutional kernels are both 10, and the pooling layer comprises maximal pooling kernels with the size of 2 × 2;
the extracting of the independent feature of each audit incidence matrix and the correlation feature between the audit incidence matrixes through the three-dimensional time sequence convolution neural network and the obtaining of the audit compliance probability value of the financial data sample according to the independent feature and the correlation feature comprise:
acquiring a plurality of checking correlation matrixes through an input layer of the three-dimensional time sequence convolution neural network;
performing convolution operation on each check correlation matrix through a first convolution core of the three-dimensional convolution layer to extract independent features, and performing convolution operation on numerical values of the same data element in the check correlation matrix in different time periods through a second convolution core of the three-dimensional convolution layer to extract correlation features;
performing dimension reduction processing on the independent features and the correlation features through a maximum pooling core of the pooling layer;
and performing network parameter fitting through the ReLU activation function and the Softmax activation function of the full connection layer, and outputting an inspection compliance probability value through the output layer.
Optionally, the loss function is a mean square error loss function, specifically:
wherein ,
for a loss of mean square function>
An audit compliance probability value output for a three dimensional time sequential convolutional neural network>
Reviewing the authenticity label for compliance;
the gradient between the fully-connected layer and the output layer is as follows:
wherein ,
is a step size, is->
Is the input of the full connection layer.
Optionally, the chaining financial change data blocks based on a consensus mechanism includes:
each department node in the multi-department financial block chain detects whether the department financial data is changed;
when the department financial data is changed, the department node with the financial data change generates a financial change data block containing a timestamp, a hash abstract, a digital signature, financial data change content and a financial data index, and generates an uplink request containing the financial change data block;
and the multi-department financial block chain receives and responds to the cochain request, identifies the financial change data blocks together, and adds and stores the financial change data blocks after the identification is successful.
Optionally, the multi-department financial block chain responds to the uplink request, identifies the financial change data block in common, and adds and stores the financial change data block after the identification is completed, including:
respectively determining department nodes and other department nodes with financial data change in the multi-department financial block chain as change initiating nodes and consensus nodes;
each consensus node acquires a chain link request containing the financial change data block, performs identity verification on the change initiating node according to a digital signature in the financial change data block, and votes for the change initiating node when the identity verification is passed;
and the multi-department financial block chain acquires a voting result, and adds the financial change data block to each consensus node when the voting result meets a preset consensus success condition.
Optionally, the deploying all nodes in the multi-department financial block chain and acquiring target financial data from all the financial data change blocks by using the query index table includes:
performing asymmetric encryption on the financial data indexes of the financial change data blocks on the multi-department financial block chain by adopting a public key, and deploying full nodes with private keys in the multi-department financial block chain;
and when the all-node receives a compliance examination request containing index information, acquiring a financial data index matched with the index information from the query index table, and accessing a physical address of the financial change data block according to the matched financial data index to acquire target financial data.
Optionally, the generating a financial data audit association cube model according to the target financial data includes:
determining a row-column index of the checking incidence matrix according to field items in the target financial data;
obtaining a correlation coefficient between every two field items corresponding to the row and column index; the correlation coefficient comprises a linear correlation coefficient and a pearson correlation coefficient;
filling the correlation coefficient into the corresponding position of the row and column index of the audit incidence matrix to complete the construction of the audit incidence matrix;
and acquiring the audit correlation matrixes of a plurality of time segments, and combining to obtain the financial data audit correlation cube model.
In addition, an embodiment of the present invention further provides a financial data compliance examining device based on a block chain, including:
the block chain building module is used for building a multi-department financial block chain and chaining financial change data blocks based on a consensus mechanism;
the index table building module is used for obtaining the financial data index and the physical address of the financial change data block so as to build a query index table;
the financial data query module is used for deploying all nodes in the multi-department financial block chain and acquiring target financial data from all the financial data change blocks by utilizing the query index table;
the correlation model generation module is used for generating a financial data checking correlation cube model according to the target financial data;
the compliance audit model building module is used for building a three-dimensional time sequence convolution neural network and training the three-dimensional time sequence convolution neural network on a preset training set to obtain a financial data compliance audit model;
and the compliance examination module is used for expanding the financial data checking association cube model and then inputting the financial data compliance examination model to obtain a compliance examination result of the target financial data.
Optionally, the compliance review model building module includes:
the neural network construction submodule is used for constructing a three-dimensional time sequence convolution neural network comprising an input layer, a three-dimensional convolution layer, a pooling layer, a full-connection layer and an output layer;
the training set construction submodule is used for acquiring the checking correlation cube model and the compliance audit real label of the financial data sample and constructing a training set;
the correlation model processing submodule is used for expanding the checking correlation cube model of the financial data sample to obtain a plurality of checking correlation matrixes in different time periods;
the neural network training submodule is used for inputting a plurality of checking correlation matrixes into the three-dimensional time sequence convolution neural network, extracting correlation characteristics between the independent characteristics of each checking correlation matrix and the checking correlation matrixes through the three-dimensional time sequence convolution neural network, and obtaining an examination compliance probability value of the financial data sample according to the independent characteristics and the correlation characteristics; obtaining a loss function according to the inspection compliance probability value and the compliance inspection real label; acquiring the gradient between a full connection layer and an output layer in the three-dimensional time sequence convolution neural network according to the loss function; and reversely propagating the gradient by a gradient descent method, and optimizing weight parameters between the full-connection layer and the output layer to obtain a financial data compliance examination model.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
1) The financial data change is dynamically recorded through a multi-department financial block chain, so that the financial data can be ensured not to be falsified or forged;
2) The financial change data blocks of all department nodes on the chain are accessed through the query index table to obtain target financial data, so that the query traceability efficiency of the financial data on the chain can be improved;
3) By generating the financial data auditing association cube model, the correlation among field items in the financial data can be represented, so that the mining of non-compliant financial data modes is facilitated;
4) The financial data compliance audit model of the three-dimensional time sequence convolution neural network structure is adopted for compliance audit, and the automated, high-efficiency and high-accuracy compliance audit on the dynamic financial data in different time periods can be realized.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a block chain-based financial data compliance audit method according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating step S50 of a block chain-based financial data compliance audit method according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram illustrating a block chain-based financial data compliance audit device according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram illustrating a compliance audit model building module in a financial data compliance audit device based on a block chain according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention, as detailed in the appended claims.
As shown in fig. 1, a flowchart of a block chain-based financial data compliance audit method according to an embodiment of the present invention includes the following steps:
and S10, constructing a multi-department financial block chain, and chaining the financial change data blocks based on a consensus mechanism.
In step S10, a multi-department financial blockchain is constructed based on the federation chain, the multi-department financial blockchain comprising a plurality of department nodes, which may be represented as
，/>
A set of department nodes representing a multi-department financial blockchain, device for selecting or keeping>
A set of connected edges representing a multi-department financial blockchain. For each department node in the multi-department financial block chain, checking financial data recorded in the ledger for ^ s>
Whether or not a change operation such as addition, modification, deletion or the like occurs>
If a change operation occurs, then>
If so, the originating department node of the financial data change generates a financial change data block ≧>
Applies for uplink and performs consensus on other department nodes on the link based on the Raft protocol>
The financial block chain of the multiple departments is newly added and stores the financial change data block ^ after the consensus is successful>
。
Preferably, the uplink of the financial change data block based on the consensus mechanism in step S10 includes the following steps:
and S101, detecting whether the department financial data are changed or not by each department node in the multi-department financial block chain.
S102, when the department financial data is changed, the department node with the financial data change generates a financial change data block containing a timestamp, a hash abstract, a digital signature, financial data change contents and a financial data index, and generates a chain loading request containing the financial change data block.
In this embodiment, for any department node in the multi-department financial block chain, if it is detected that the financial data is changed, the data is generated to include a timestamp
Hash digest->
Digital signature
And the financial data change content->
Financial data index->
Based on the financial change data block->
And generates a block of data comprising financial changes &>
The uplink request is sent to a multi-department financial block chain. Wherein the time stamp->
For recording financial data change times; hash digest->
The method is characterized in that a characteristic code obtained after the SHA-1 (Secure Hash Algorithm 1 ) is adopted to carry out Hash operation on the whole financial change data block is used for ensuring that the financial change data block is not tampered; digital signature
Adopting private key of department for adopting RSA algorithm (public key encryption algorithm) to uniquely identify department
The encrypted ciphertext mark can be judged by other department nodes on the chain by using the public key->
After decryption, verifying the authenticity of the department node generating the financial data change block (namely the department node generating the financial data change); financial data change content->
All content for recording financial data changes; financial data index +>
Altering content for financial data>
The keyword(s).
And S103, the multi-department financial block chain receives and responds to the chain loading request, identifies the financial change data blocks, and adds and stores the financial change data blocks after successful identification.
In this embodiment, the step S103 may include the following steps:
and S1031, respectively determining the department nodes and other department nodes with financial data change in the multi-department financial block chain as change initiating nodes and consensus nodes.
S1032, each consensus node acquires a chain link request containing the financial change data block, performs identity verification on the change initiating node according to the digital signature in the financial change data block, and votes for the change initiating node when the identity verification passes.
And S1033, the multi-department financial block chain acquires a voting result, and when the voting result meets a preset consensus success condition, the financial change data blocks are added to the consensus nodes.
In step S1031, the department node where the financial data change occurs is a department node that generates a financial change data block or a department node that generates a uplink request.
In step S1032, after receiving the uplink request of the department node (i.e., change initiating node) that has undergone the financial data change, the other department nodes (i.e., consensus nodes) in the chain analyze the uplink request to obtain the financial change data block, analyze the financial change data block to obtain the digital signature, decrypt the digital signature with the public common key to obtain the department unique identifier of the change initiating node, and compare the department unique identifier of the change initiating node with the identification information recorded in advance by the consensus node to verify the identity authenticity of the change initiating node, if the comparison results are the same, determine that the identity verification passes, vote for the change initiating node, otherwise determine that the identity verification fails, and vote not for the change initiating node.
In step S1033, the multi-department financial block chain counts the voting number to obtain a voting result, and when it is detected that the voting number exceeds half of the node number, it is determined that the voting result satisfies a consensus success condition, and the financial change data block is newly added to each consensus node. It can be understood that the multi-department financial block chain constructed in the embodiment can dynamically record the financial data change, and ensure that the financial data cannot be falsified and counterfeited.
And S20, acquiring the financial data index and the physical address of the financial change data block to construct a query index table.
In step S20, a financial data index is first obtained by parsing from the financial data change data block, where the financial data index may be set according to the keywords of the financial data change content, and the keywords of the financial data change content may be obtained through an existing keyword extraction tool. Then, the financial data index is associated with the physical address of the financial change data block, and a query index table is constructed
And storing the data into a database of the multi-department financial block chain. It can be understood that the problem of low efficiency of querying the financial data on the block chain due to complicated and complicated financial data on the block chain can be solved by constructing the query index table in the embodiment.
And S30, deploying all nodes in the multi-department financial block chain, and acquiring target financial data from all the financial data change blocks by using the query index table.
In step S30, the financial data of the financial changed data blocks on the chain is changed because the financial data of a plurality of departments is contained on the block chain, which relates to the problem that the financial data may not have the relevant access rightIndex
With the public key public->
Performing RSA encryption (i.e., asymmetric encryption) <' > on>
And deploy a key on the block chain based on the private key +>
Has a private key->
The global node of (2) may decrypt the financial data index of the financial altered data block and then access the permissions of all financial altered data blocks on the chain. When performing compliance review, the full node may query and normalize all financial alteration data blocks using the query index table.
Preferably, the step S30 includes the steps of:
s301, performing asymmetric encryption on the financial data indexes of the financial change data blocks on the multi-department financial block chain by adopting a public key, and deploying full nodes with private keys in the multi-department financial block chain.
S302, when the whole node receives a compliance audit request containing index information, acquiring a financial data index matched with the index information from the query index table, and accessing a physical address of the financial change data block according to the matched financial data index to acquire target financial data.
In this embodiment, the multi-department financial blockchain includes a plurality of department nodes that store financial data and a full node that queries and reviews financial data. After receiving a compliance examination request containing index information, all nodes firstly analyze the compliance examination request to obtain the index information, wherein the index information can be a keyword or an index name of a financial data index; and finally, accessing the financial change data block of the corresponding department node on the link by using the target financial data index, and analyzing the financial data change content from the financial change data block to be used as the target financial data.
It will be appreciated that all nodes of the multi-department financial block chain may access financial alteration data blocks of any department node in the chain to obtain at least one set of target financial data in accordance with a compliance audit request.
And S40, generating a financial data checking association cube model according to the target financial data.
In step S40, based on the non-falsifiable and non-falsifiable target financial data obtained from each department node in the chain, generating a financial data checking association cube model, which is checked by checking association matrix
Is greater than or equal to>
Composition, which can be expressed as->
, wherein />
Indicates the fifth->
A checking correlation matrix for each time slice, and>
. For the audit incidence matrix, the row index and the column index of the audit incidence matrix both represent field items in the target financial data in a data table format, the numerical values of the corresponding positions of the rows and the columns represent correlation coefficients between every two field items, and the correlation coefficients comprise linear correlation and Pearson correlation.
Preferably, the step S40 specifically includes the following steps:
s401, determining a row and column index of the checking incidence matrix according to the field items in the target financial data.
S402, obtaining a correlation coefficient between every two field items corresponding to the row and column index.
S403, filling the correlation coefficient into the corresponding position of the row and column index of the audit incidence matrix to complete the construction of the audit incidence matrix.
S404, obtaining the checking correlation matrix of a plurality of time segments, and combining according to the time sequence to obtain the financial data checking correlation cube model.
In step S401, the field entries include, but are not limited to, expenses and profits of different enterprise operation activities, and the like.
In step S404, the time slice is set according to the change frequency of the financial data, and the higher the change frequency of the financial data is, the shorter the time slice is.
Further, the correlation coefficient includes a linear correlation coefficient and a pearson correlation coefficient, in this case, the step S302 specifically includes the following steps:
s3021, calculating a linear correlation coefficient according to the value of each row and each column field, where the linear correlation coefficient is calculated according to the following formula:
wherein ,
is the first->
The value of the row field entry, <' >>
Is the first->
The value of the column field entry, <' >>
、/>
Is constant and is->
Is a linear correlation coefficient.
S3022, calculating a pearson correlation coefficient according to the value of each row and each column of field entries and the average value of the values of each row and each column of field entries, where the pearson correlation coefficient is calculated according to the following formula:
wherein ,
、/>
are respectively the fifth->
Row and/or column>
Value of a column field entry>
、/>
Are respectively first>
Row and/or column>
The value of the column field entry is->
Average value in (d).
And S3023, determining the linear correlation coefficient and the Pearson correlation coefficient as correlation coefficients between the two field items.
Specifically, in the construction process of the checking incidence matrix, firstly, the target financial data is determined according to the data in the target financial data
Individual field entries determine the ∑ of an audit correlation matrix>
Individual row index and->
Column index and initialize->
Checking the incidence matrix to make the value of each element in the incidence matrix as a preset initial value; then, obtaining a correlation coefficient between a field item corresponding to the row index and a field item corresponding to the column index through a calculation formula of the linear correlation coefficient and a calculation formula of the Pearson correlation coefficient, and replacing a preset initial value of an element corresponding to the row index and the column index in the initialized colluding correlation matrix with the correlation coefficient; and finally, obtaining a final audit incidence matrix after the numerical values of all elements in the initialized audit incidence matrix are replaced.
It can be understood that when multiple sets of target financial data are acquired according to the compliance audit request, a corresponding audit incidence matrix can be constructed for each set of target financial data and associated with the department unique identifier.
And S50, constructing a three-dimensional time sequence convolution neural network, and training the three-dimensional time sequence convolution neural network on a preset training set to obtain a financial data compliance examination model.
In step S50, the predetermined training set includes the audit-related cube model of the financial data sample and the compliance audit authenticity label of the financial data sample. The financial data sample is historical financial data recorded in the account book by each department node on the chain. The checking of the financial data sample with the associative cube model may refer to the generation process of the financial data checking associative cube model, which specifically includes the steps S401 to S404. The compliance audit real label of the financial data sample is obtained by a manual marking mode.
It should be noted that, the construction of the financial data compliance audit model only needs to be performed before the financial data compliance audit model is applied for compliance audit, that is, step S50 may be performed before any step of step S10 to step S40, and fig. 1 is a flowchart of only one alternative embodiment.
Preferably, as shown in fig. 2, the step S50 specifically includes the following steps:
s501, constructing a three-dimensional time sequence convolution neural network comprising an input layer, a three-dimensional convolution layer, a pooling layer, a full-link layer and an output layer.
S502, obtaining a checking correlation cube model and a compliance audit real label of the financial data sample, and constructing a training set.
S503, expanding the checking association cube model of the financial data sample to obtain a plurality of checking association matrixes in different time periods.
S504, inputting the plurality of checking incidence matrixes into the three-dimensional time sequence convolution neural network, extracting the correlation characteristics between the independent characteristics of each checking incidence matrix and the checking incidence matrixes through the three-dimensional time sequence convolution neural network, and obtaining the checking compliance probability value of the financial data sample according to the independent characteristics and the correlation characteristics.
S505, obtaining a loss function according to the inspection compliance probability value and the compliance inspection real label; the loss function is a mean square error loss function and can be represented as:
wherein ,
is a loss of mean square function->
An audit compliance probability value output for a three dimensional time sequential convolutional neural network>
The authentic tag is reviewed for compliance.
S506, acquiring the gradient between the full connection layer and the output layer in the three-dimensional time sequence convolution neural network according to the loss function; the gradient between the fully-connected layer and the output layer can be expressed as:
wherein ,
for step size +>
Is input for the full connection layer.
And S507, reversely propagating the gradient by a gradient descent method, and optimizing the weight parameters between the full connection layer and the output layer to obtain a financial data compliance audit model.
It should be noted that, constructing the three-dimensional time series convolutional neural network and constructing the training set may be performed simultaneously, or may be performed preferentially in a certain step, that is, step S501 and step S501 may be performed simultaneously, or may be performed preferentially in any step over another step, and fig. 2 is only a flowchart of an optional embodiment.
Further, the three-dimensional convolutional layer includes a first convolutional kernel with a size of 3 × 3 and a second convolutional kernel with a size of 1 × 10, the number of the first convolutional kernel and the number of the second convolutional kernel are both 10, and the pooled layer includes a maximum pooled kernel with a size of 2 × 2, in this case, the step S504 includes the following steps:
s5031, obtaining a plurality of colluding correlation matrices through an input layer of the three-dimensional time-series convolutional neural network.
S5032, performing convolution operation on each check correlation matrix through a first convolution kernel of the three-dimensional convolution layer to extract an independent feature, and performing convolution operation on values of the same data element in the check correlation matrix in different time periods through a second convolution kernel of the three-dimensional convolution layer to extract a correlation feature.
S5033, dimension reduction processing is carried out on the independent feature and the correlation feature through the maximum pooling core of the pooling layer.
And S5034, performing network parameter fitting through a ReLU activation function and a Softmax activation function of the full connection layer, and outputting an inspection compliance probability value through the output layer.
In this embodiment, the financial data compliance audit model adopting the 3D time sequence convolutional neural network structure includes an input layer, a 3D convolutional layer, a pooling layer, a full connection layer and an output layer, wherein the input layer is used for acquiring a plurality of audit correlation matrices expanded by the audit correlation cube model; the 3D convolutional layer (namely the three-dimensional convolutional layer) comprises two types of convolution kernels, wherein one type of convolution kernel is 10 randomly generated convolution kernels with the sizes of 3 x 3 and is used for performing convolution operation on the check correlation matrix in different time periods and extracting independent features, and the other type of convolution kernel is 10 randomly generated convolution kernels with the sizes of 1 x 10 and is used for performing convolution operation on numerical values of the same data element in the check correlation matrix in different time periods and extracting correlation features; performing dimensionality reduction treatment on the pooling layer by adopting a maximum pooling kernel with the size of 2 x 2 to retain the significant features; the full connection layer comprises a ReLU activation function and a Softmax activation function and is used for processing the feature information after dimension reduction; the output layer is to output an audit compliance probability value. It can be understood that, in the embodiment, the financial data compliance audit model of the 3D time sequence convolutional neural network structure is adopted, so that the independent features of the checking incidence matrix in different time periods and the correlation features of the checking incidence matrix in different time periods can be extracted, the non-compliance financial data mode can be automatically mined, the possibility that the financial data is not compliant can be further judged, and the automated, high-efficiency and high-accuracy compliance audit on the dynamic financial data in different time periods can be realized.
And S60, expanding the financial data auditing association cube model, and inputting the financial data compliance audit model to obtain a compliance audit result of the target financial data.
Specifically, when the compliance audit is performed by using the financial data compliance audit model, a plurality of audit checking association matrixes obtained by expanding the financial data audit association cube model are obtained through an input layer of the financial data compliance audit model, correlation characteristics between the independent characteristics and the audit checking association matrixes are extracted through a 3D convolutional layer, and then, after the independent characteristics and the correlation characteristics are subjected to dimension reduction through a pooling layer, the compliance audit probability value is output through a full connection layer and an output layer, so that the compliance audit result of the target financial data is obtained.
In summary, the financial data compliance review method based on the block chain provided by the embodiment has the following beneficial effects:
1) Financial data change is dynamically recorded through a multi-department financial block chain, so that the financial data can be ensured not to be falsified and forged;
2) The financial change data blocks of all department nodes on the chain are accessed through the query index table to obtain target financial data, so that the query traceability efficiency of the financial data on the chain can be improved;
3) By generating the financial data auditing association cube model, the correlation among field items in the financial data can be represented, so that the mining of non-compliant financial data modes is facilitated;
4) The financial data compliance audit model of the three-dimensional time sequence convolution neural network structure is adopted for compliance audit, and the automated, high-efficiency and high-accuracy compliance audit on the dynamic financial data in different time periods can be realized.
In addition, as shown in fig. 3, an embodiment of the present invention further provides a block chain-based financial data compliance audit device, including:
a block chain construction module 110, configured to construct a multi-department financial block chain, and perform chain linking on financial change data blocks based on a consensus mechanism;
an index table building module 120, configured to obtain the financial data index and the physical address of the financial change data block to build a query index table;
a financial data query module 130, configured to deploy all nodes in the multi-department financial block chain, and obtain target financial data from all the financial data change blocks by using the query index table;
the correlation model generation module 140 is used for generating a financial data checking correlation cube model according to the target financial data;
the compliance audit model building module 150 is used for building a three-dimensional time sequence convolution neural network and training the three-dimensional time sequence convolution neural network on a preset training set to obtain a financial data compliance audit model;
and the compliance audit module 160 is configured to expand the financial data auditing association cube model, and then input the financial data compliance audit model to obtain a compliance audit result of the target financial data.
In some alternative embodiments, as shown in FIG. 4, the compliance audit model building module 150 includes:
the neural network construction sub-module 151 is used for constructing a three-dimensional time sequence convolution neural network comprising an input layer, a three-dimensional convolution layer, a pooling layer, a full-link layer and an output layer;
a training set construction submodule 152 for obtaining a check correlation cube model and a compliance audit real label of the financial data sample, and constructing a training set;
the correlation model processing submodule 153 is used for expanding the checking correlation cube model of the financial data sample to obtain a plurality of checking correlation matrixes in different time periods;
the neural network training sub-module 154 is used for inputting a plurality of checking incidence matrixes into the three-dimensional time sequence convolution neural network, extracting the correlation characteristics between the independent characteristics of each checking incidence matrix and the checking incidence matrix through the three-dimensional time sequence convolution neural network, and obtaining the inspection compliance probability value of the financial data sample according to the independent characteristics and the correlation characteristics; obtaining a loss function according to the inspection compliance probability value and the compliance inspection real label; obtaining the gradient between a full connection layer and an output layer in the three-dimensional time sequence convolution neural network according to the loss function; and reversely propagating the gradient by a gradient descent method, and optimizing weight parameters between the full-connection layer and the output layer to obtain a financial data compliance examination model.
It can be understood that the financial data compliance inspection device based on the blockchain provided in this embodiment is used to implement the financial data compliance inspection method of the blockchain in the foregoing embodiment, and has the beneficial effects of the corresponding method embodiment, which are not described herein again.
It should be noted that the terms "first," "second," and the like in the description of the present invention are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. Further, in the description of the present invention, the meaning of "a plurality" means at least two unless otherwise specified.
Although the present invention has been described with reference to the preferred embodiments, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (10)
1. A block chain-based financial data compliance review method is characterized by comprising the following steps:
constructing a multi-department financial block chain, and chaining financial change data blocks based on a consensus mechanism;
acquiring a financial data index and a physical address of the financial change data block to construct a query index table;
deploying all nodes in the multi-department financial block chain, and acquiring target financial data from all the financial data change blocks by using the query index table;
generating a financial data checking association cube model according to the target financial data;
constructing a three-dimensional time sequence convolution neural network, and training the three-dimensional time sequence convolution neural network on a preset training set to obtain a financial data compliance examination model;
and after expanding the financial data auditing association cube model, inputting the financial data compliance audit model to obtain a compliance audit result of the target financial data.
2. The method for financial data compliance review based on block chain according to claim 1, wherein the step of constructing a three-dimensional time sequence convolutional neural network and training the three-dimensional time sequence convolutional neural network on a preset training set to obtain a financial data compliance review model comprises the steps of:
constructing a three-dimensional time sequence convolution neural network comprising an input layer, a three-dimensional convolution layer, a pooling layer, a full-connection layer and an output layer;
obtaining a checking correlation cube model and a compliance audit real label of a financial data sample, and constructing a training set;
expanding the checking association cube model of the financial data sample to obtain a plurality of checking association matrixes in different time periods;
inputting a plurality of checking incidence matrixes into the three-dimensional time sequence convolution neural network, extracting the correlation characteristics between the independent characteristics of each checking incidence matrix and the checking incidence matrixes through the three-dimensional time sequence convolution neural network, and obtaining the inspection compliance probability value of the financial data sample according to the independent characteristics and the correlation characteristics;
obtaining a loss function according to the inspection compliance probability value and the compliance inspection real label;
acquiring the gradient between a full connection layer and an output layer in the three-dimensional time sequence convolution neural network according to the loss function;
and reversely propagating the gradient by a gradient descent method, and optimizing weight parameters between the full-connection layer and the output layer to obtain a financial data compliance examination model.
3. The blockchain-based financial data compliance review method of claim 2 wherein the three-dimensional convolution layer includes a first convolution kernel of size 3 x 3 and a second convolution kernel of size 1 x 10, the first and second convolution kernels each being 10 in number, the pooling layer including a maximum pooling kernel of size 2 x 2;
the extracting of the independent feature of each audit incidence matrix and the correlation feature between the audit incidence matrixes through the three-dimensional time sequence convolution neural network and the obtaining of the audit compliance probability value of the financial data sample according to the independent feature and the correlation feature comprise:
acquiring a plurality of colluding correlation matrixes through an input layer of the three-dimensional time sequence convolution neural network;
performing convolution operation on each check correlation matrix through a first convolution core of the three-dimensional convolution layer to extract independent features, and performing convolution operation on numerical values of the same data element in the check correlation matrix in different time periods through a second convolution core of the three-dimensional convolution layer to extract correlation features;
performing dimension reduction processing on the independent features and the correlation features through a maximum pooling core of the pooling layer;
and performing network parameter fitting through the ReLU activation function and the Softmax activation function of the full connection layer, and outputting an inspection compliance probability value through the output layer.
4. The block chain-based financial data compliance review method according to claim 2, wherein the loss function is a mean square error loss function, and specifically comprises:
wherein ,
is a loss of mean square function->
An audit compliance probability value output for a three dimensional time sequential convolutional neural network>
Reviewing the authenticity label for compliance;
the gradient between the fully-connected layer and the output layer is as follows:
wherein ,
is a step size, is->
Is the input of the full connection layer.
5. The method of claim 1, wherein the co-recognition based mechanism for uplink of financial changed data blocks comprises:
each department node in the multi-department financial block chain detects whether the department financial data is changed;
when the department financial data is changed, the department node with the financial data change generates a financial change data block containing a timestamp, a hash abstract, a digital signature, financial data change content and a financial data index, and generates a chain linking request containing the financial change data block;
and the multi-department financial block chain receives and responds to the cochain request, identifies the financial change data blocks together, and adds and stores the financial change data blocks after the identification is successful.
6. The block chain-based financial data compliance review method of claim 5 wherein the multi-department financial block chain agrees to the financial alteration data block in response to the uplink request and adds and stores the financial alteration data block after the agreement is completed, comprising:
respectively determining department nodes and other department nodes with financial data change in the multi-department financial block chain as change initiating nodes and consensus nodes;
each consensus node acquires a chain link request containing the financial change data block, performs identity verification on the change initiating node according to a digital signature in the financial change data block, and votes for the change initiating node when the identity verification is passed;
and the multi-department financial block chain acquires a voting result and adds the financial change data block to each consensus node when the voting result meets a preset consensus success condition.
7. The blockchain-based financial data compliance review method of claim 1 wherein said deploying full nodes in the multi-department financial blockchain and using the lookup index table to obtain target financial data from all of the financial data change blocks comprises:
performing asymmetric encryption on financial data indexes of the financial change data blocks on the multi-department financial block chain by adopting a public key, and deploying all nodes with private keys in the multi-department financial block chain;
and when the full node receives a compliance audit request containing index information, acquiring a financial data index matched with the index information from the query index table, and accessing a physical address of the financial change data block according to the matched financial data index to acquire target financial data.
8. The blockchain-based financial data compliance review method of claim 1 wherein generating a financial data audit association cube model based on the target financial data comprises:
determining a row-column index of the checking incidence matrix according to the field items in the target financial data;
obtaining a correlation coefficient between every two field items corresponding to the row and column index; the correlation coefficient comprises a linear correlation coefficient and a pearson correlation coefficient;
filling the correlation coefficient into the corresponding position of the row and column index of the check incidence matrix to complete the construction of the check incidence matrix;
and acquiring the audit correlation matrixes of a plurality of time segments, and combining to obtain the financial data audit correlation cube model.
9. A block chain-based financial data compliance review device, comprising:
the block chain building module is used for building a multi-department financial block chain and chaining financial change data blocks based on a consensus mechanism;
the index table building module is used for obtaining the financial data index and the physical address of the financial change data block so as to build a query index table;
the financial data query module is used for deploying all nodes in the multi-department financial block chain and acquiring target financial data from all the financial data change blocks by utilizing the query index table;
the correlation model generation module is used for generating a financial data checking correlation cube model according to the target financial data;
the compliance audit model building module is used for building a three-dimensional time sequence convolution neural network and training the three-dimensional time sequence convolution neural network on a preset training set to obtain a financial data compliance audit model;
and the compliance examination module is used for expanding the financial data checking association cube model and inputting the financial data compliance examination model to obtain a compliance examination result of the target financial data.
10. The blockchain-based financial data compliance audit device according to claim 9 wherein the compliance audit model building module includes:
the neural network construction submodule is used for constructing a three-dimensional time sequence convolution neural network comprising an input layer, a three-dimensional convolution layer, a pooling layer, a full-connection layer and an output layer;
the training set construction submodule is used for acquiring the checking correlation cube model and the compliance audit real label of the financial data sample and constructing a training set;
the correlation model processing submodule is used for expanding the checking correlation cube model of the financial data sample to obtain a plurality of checking correlation matrixes in different time periods;
the neural network training submodule is used for inputting a plurality of checking incidence matrixes into the three-dimensional time sequence convolution neural network, extracting correlation characteristics between the independent characteristics of each checking incidence matrix and the checking incidence matrixes through the three-dimensional time sequence convolution neural network, and obtaining examination compliance probability values of financial data samples according to the independent characteristics and the correlation characteristics; obtaining a loss function according to the inspection compliance probability value and the compliance inspection real label; acquiring the gradient between a full connection layer and an output layer in the three-dimensional time sequence convolution neural network according to the loss function; and reversely propagating the gradient by a gradient descent method, and optimizing weight parameters between the full-connection layer and the output layer to obtain a financial data compliance examination model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310248208.4A CN115952240B (en) | 2023-03-15 | 2023-03-15 | Financial data compliance examination method and device based on blockchain |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310248208.4A CN115952240B (en) | 2023-03-15 | 2023-03-15 | Financial data compliance examination method and device based on blockchain |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115952240A true CN115952240A (en) | 2023-04-11 |
| CN115952240B CN115952240B (en) | 2023-06-02 |
Family
ID=85891444
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310248208.4A Active CN115952240B (en) | 2023-03-15 | 2023-03-15 | Financial data compliance examination method and device based on blockchain |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115952240B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108959228A (en) * | 2018-07-13 | 2018-12-07 | 众安信息技术服务有限公司 | Based on block chain creation, retrieval, the method and readable storage medium storing program for executing for editing data |
| CN109241192A (en) * | 2018-09-18 | 2019-01-18 | 百度在线网络技术（北京）有限公司 | The data modification and block verification method, device, equipment and medium of block chain |
| US20200293549A1 (en) * | 2018-06-29 | 2020-09-17 | Alibaba Group Holding Limited | Blockchain-based data processing method and device |
| CN113255496A (en) * | 2021-05-17 | 2021-08-13 | 杭州西语科技有限公司 | Financial expense reimbursement management method based on block chain technology |
| CN115357623A (en) * | 2022-08-31 | 2022-11-18 | 普联软件股份有限公司 | Intelligent organization method, system and medium for multidimensional cube data |
-
2023
- 2023-03-15 CN CN202310248208.4A patent/CN115952240B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200293549A1 (en) * | 2018-06-29 | 2020-09-17 | Alibaba Group Holding Limited | Blockchain-based data processing method and device |
| CN108959228A (en) * | 2018-07-13 | 2018-12-07 | 众安信息技术服务有限公司 | Based on block chain creation, retrieval, the method and readable storage medium storing program for executing for editing data |
| CN109241192A (en) * | 2018-09-18 | 2019-01-18 | 百度在线网络技术（北京）有限公司 | The data modification and block verification method, device, equipment and medium of block chain |
| CN113255496A (en) * | 2021-05-17 | 2021-08-13 | 杭州西语科技有限公司 | Financial expense reimbursement management method based on block chain technology |
| CN115357623A (en) * | 2022-08-31 | 2022-11-18 | 普联软件股份有限公司 | Intelligent organization method, system and medium for multidimensional cube data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115952240B (en) | 2023-06-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443274B (en) | Abnormality detection method, abnormality detection device, computer device, and storage medium | |
| Raghavan | Digital forensic research: current state of the art | |
| US20160283920A1 (en) | Authentication and verification of digital data utilizing blockchain technology | |
| CN103685308B (en) | Detection method and system of phishing web pages, client and server | |
| CN106650799B (en) | A kind of electronic evidence classification extracting method and system | |
| US20210109917A1 (en) | System and Method for Processing a Database Query | |
| CN109255056B (en) | Data reference processing method, device, equipment and storage medium of block chain | |
| Poisel et al. | A comprehensive literature review of file carving | |
| CN111353180A (en) | Block chain evidence storing method, evidence obtaining method and system | |
| CN103201747A (en) | A method for attesting a plurality of data processing systems | |
| CN109801151B (en) | Financial falsification risk monitoring method, device, computer equipment and storage medium | |
| US20190095595A1 (en) | Data Watermarking and Fingerprinting System and Method | |
| US20220006638A1 (en) | Block chain proof for identification | |
| US20080243751A1 (en) | Method and Apparatus for Recording Associations with Logs | |
| CN111222181B (en) | AI model supervision method, system, server and storage medium | |
| US9223784B2 (en) | Method and apparatus for archiving media using a log | |
| US9465858B2 (en) | Systems and methods for authenticating and aiding in indexing of and searching for electronic files | |
| CN115952240A (en) | Financial data compliance examination method and device based on block chain | |
| CN116910820A (en) | Data report processing method, device, computer equipment and storage medium | |
| CN111917861A (en) | Knowledge storage method and system based on block chain and knowledge graph and application thereof | |
| CN112667661B (en) | Tracing information correlation query method and device | |
| CN115618394B (en) | Four-way detection method based on cloud archive integrated platform | |
| US20240184747A1 (en) | Method and system for blockchain-based data management | |
| CN114004456B (en) | Data tag calculation method, device, computer equipment and storage medium | |
| CN112560055B (en) | Credible electronic license system based on PKI technology and working method |
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 |