CN112445844B - Big data platform financial data management control system - Google Patents

Abstract

The invention proposes a financial data management and control system on a big data platform, comprising: a query and extraction module, which is used to obtain financial data through a cloud database, log in abnormal data in the financial data, and then start a preliminary query. The data is subjected to real-time query, verification and extraction; the abnormality judgment module is used to set the judgment interval of abnormal data after the real-time query, verification and extraction, and form standardized data in the judgment interval; the screening and scoring module is used to measure the deviation of the standardized data. Analysis, after the analysis, the abnormal data is screened by the screening model, and the characteristic score of the abnormal data after screening is carried out. Judgment output.

Description

Big data platform financial data management control system

technical field

The invention relates to the field of big data analysis, in particular to a financial data management control system of a big data platform.

Background technique

With the rapid development of informatization and intelligence, in the process of financial data management, due to the increase in the number of transactions, innumerable transaction history data is formed. How much of the transaction history data is compliant operation or normal transaction behavior? For financial data managers, traditional inspection methods can no longer meet the changing and complex abnormal transaction behaviors of the current society.

Especially in schools, government agencies or large chain enterprises, the total amount of transactions and the number of transactions are even more difficult to count. Traditional computer accumulation methods or statistical principles cannot quickly and accurately obtain transaction behaviors that include transaction risks, and even if some are used. Extraction algorithm, its control and inspection process of abnormal financial data is not accurate. This requires those skilled in the art to solve the corresponding technical problems.

SUMMARY OF THE INVENTION

The present invention aims to at least solve the technical problems existing in the prior art, and particularly innovatively proposes a financial data management control system on a big data platform.

In order to achieve the above-mentioned purpose of the present invention, the present invention provides a financial data management control system of a big data platform, including:

The query extraction module is used to obtain financial data through the cloud database, log in the abnormal data in the financial data, and then start a preliminary query, and perform real-time query verification and extraction of invalid data during the query process;

The abnormality judgment module is used to set the judgment interval of abnormal data after real-time query, verification and extraction, and form standardized data in the judgment interval;

The screening and scoring module is used to analyze the deviation degree of the standardized data. After the analysis, the abnormal data is screened by the screening model, and the characteristic score of the abnormal data after screening is performed.

The comprehensive judgment module is used to judge and output the risk degree of abnormal data in the financial data through the comprehensive risk judgment model after characteristic scoring.

Preferably, the query extraction module includes:

The financial data is retrieved from the cloud database, and abnormal data is obtained from the financial data. The abnormal data extraction process performs data balance through the preliminary query process to dynamically request the financial data of the cloud database, and uses the dynamic configuration method to set the acquisition threshold of abnormal data. According to the security control mechanism and authority management requirements of different financial data, extract different abnormal data for login operation,

During the initial query process, the cloud database stores financial data authentication and functional access authority information in the local database, and performs unified financial data authentication and functional authority control; for financial data, abnormal data is logically isolated and stored in an independent database; During the data login process, the user's identity is verified, and the abnormal data set that the user has access to is constructed according to the abnormal data access authority information in the financial data, and the authentication access is performed through the identity authentication process of the cloud database; if the access fails, the abnormal data access failure information is returned. ; If the access is successful, the login is successful; establish an independent channel between the user and the application server instance dynamically allocated by the system.

Preferably, the query extraction module includes:

The process of accessing and using abnormal data is to form abnormal data relationship nodes according to multiple abnormal data, search for PaaS platform resources to convert them into tree nodes, generate a list of abnormal data tree nodes, and use the empty abnormal data node set as the For the current node set, perform a traversal operation on the current abnormal data tree node set, so as to determine whether the abnormal data parent resource information list of the node set of the current traversal operation is equal to the preset abnormal data root node information list, if it is equal to the preset abnormal data root node information list If it is not equal to the preset abnormal data root node information list, the node set of the current traversal operation is the root node of the current abnormal data permission tree. The resource of the parent resource information list of the node of the traversal operation is marked as the abnormal data parent node of the node of the current traversal operation.

Preferably, the query extraction module includes:

For whether the abnormal data node is equal to the parent resource information list of the currently traversed node, it is judged whether the current tree node list has been traversed; if the traversal is completed, the abnormal data parent node information list is detected. The data parent node information list is used as the root node of the current tree node, and the abnormal data service query tree is constructed recursively by marking; multiple abnormal data query requests allocated on a certain abnormal data node are reassigned to a certain abnormal data computing node and Backup so that each of the certain compute node and the backup compute node is assigned only one subquery.

Preferably, the abnormality judgment module includes:

After the query and verification, the abnormal data is divided into judgment intervals, and the similarity of the abnormal data is calculated to generate the judgment _interval . The conversion value is u′ _i , the conversion value of transaction data vi of abnormally large amount of scattered funds is transferred in and fast centralized transfer out is v′ _i , the conversion value of transaction data _xi at abnormal time point is x′ _{i ,} abnormal transaction data of the same amount The conversion value of y _i is y′ _i , and the conversion value of abnormal over-limit transaction data _zi is z′ _k ;

Substitute the transformed abnormal transaction data into the judgment model together with the time and date variables, and calculate the judgment value of the abnormal data at any time and date of the statistics:

Among them, B(t,d) is the judgment value of abnormal data at any time t and date d; f( _u′i ; t,d) is the time and date of an abnormally large amount of funds transferred in and out of transaction data quickly dispersed The judgment value of ; f(v′ _i ; t, d) is the judgment value of the time and date of the transaction data of an abnormally large amount of decentralized funds transferred into the fast centralized transfer; f(x′ _i ; t, d) is an abnormal The judgment value of the time and date of the transaction data at the time point; f(y′ _i ; t, d) is the judgment value of the time and date of an abnormal transaction data of the same amount; f (z′ _i ; t, d) is an abnormal Judgment value of over-quota transaction data; the maximum i is 60 is to ensure that every second of abnormal data within one minute is monitored and judged in real time.

Preferably, the abnormality judgment module includes:

Calculate the difference between the actual and judged values of each abnormal data on time and date, and perform a linear curve fitting process on the discrete abnormal data through the residual sum of squares, so as to judge the risk trend of abnormal data.

Among them, W is the residual sum of squares of each abnormal data; B ₀ (t,d) is the actual value of each abnormal data at this time and date; B(t, d) is the sum of each abnormal data at this time Judgment value of the date; M is the time with the largest time or the maximum number of days of the date.

Preferably, the abnormality judgment module includes:

Then calculate the deviation of abnormal data

Among them, F is the calculation constant, which is adjusted by the adjustment coefficient λ. Since W increases, the calculation constant F is larger; the value H _j is accurately obtained by increasing the abnormal data, and the accurately obtained values are accumulated for all N abnormal data. The deviation of the eigenvalue e converges, and β is the feature threshold.

Preferably, the screening scoring module includes:

和

计算得到：After the deviation degree of abnormal data is analyzed, statistical information is calculated in the abnormal data through the prior probability distribution; the prior conditional probability distribution of abnormal data is calculated, and the internal attributes of the first abnormal data set C and the second abnormal data set E are set, wherein the first abnormal data set C and the second abnormal data set E are set. An abnormal data set includes u _i and v _i , and the second abnormal data set includes _xi , _yi and _zi . By defining the time class attribute G and date class attribute I of the abnormal data, the conditional probability is calculated respectively under the condition of probability distribution

and

Calculated:

continue to derive,

表示第一异常数据集合

和时间类属性G和日期类属性I联合概率分布，遍历第一异常数据集合

和时间类属性G的全部值得到其条件概率分布

以及第一异常数据集合

和日期类属性I的全部值得到其条件概率分布

时间类属性条件概率Q(G)，日期类属性条件概率Q(I)；in

Represents the first abnormal data set

Joint probability distribution with time class attribute G and date class attribute I, and traverse the first abnormal data set

and all the values of the time class attribute G to get its conditional probability distribution

and the first abnormal data set

and all values of the date class attribute I get its conditional probability distribution

Time class attribute conditional probability Q(G), date class attribute conditional probability Q(I);

Then calculate:

continue to derive,

表示第二异常数据集合

和时间类属性G和日期类属性I联合概率分布，遍历第二异常数据集合

和时间类属性G的全部值得到其条件概率分布

以及第二异常数据集合

和日期类属性I的全部值得到其条件概率分布

in

Represents the second abnormal data set

Joint probability distribution with time class attribute G and date class attribute I, and traverse the second abnormal data set

and all the values of the time class attribute G to get its conditional probability distribution

and the second set of abnormal data

and all values of the date class attribute I get its conditional probability distribution

Preferably, the screening scoring module includes:

The joint probability distribution values of the condition information of the time attribute and date attribute of each abnormal data node in the first abnormal data set C and the US abnormal data nodes in the second abnormal data set E are as follows;

Select the class attribute J of the abnormal data and put it into the big data platform; take the class attribute J as the parent node of the internal attributes of the first abnormal data set C and the second abnormal data set E, and construct a Naive Bayesian network;

则将

放入网络作为其父节点；；若第二异常数据集合E中

则将

放入网络作为其父节点；从而得到用于异常数据等级筛选排序的贝叶斯网络。Put the nodes in the first abnormal data set C and the second abnormal data set E into the Bayesian network one by one; if the first abnormal data set C

will

Put it into the network as its parent node; if the second abnormal data set E is

will

Put it into the network as its parent node; thus get a Bayesian network for rank screening and sorting of abnormal data.

Preferably, the comprehensive judgment module includes:

Combined with the calculation of risk weight, calculate the transaction data _ui of abnormally large amount of funds transferred in and out quickly and dispersedly:

Among them, T _total is the total reference time; p _ui is the dynamic change component of the transaction data weight of the abnormally large amount of funds transferred in and out quickly and dispersedly; V _total is the total reference date, U is the transaction data detection time component; K is the transaction data detection date component;

Calculate the risk weight for the transaction data v _i of the abnormally large amount of decentralized funds transferred in and out quickly and in a centralized manner,

为异常大量分散资金转入快速集中转出的交易数据v_i权重的动态变化分量；in,

It is the dynamic change component of the weight of transaction data v _i for the transfer of an abnormally large amount of decentralized funds in and out of the rapid centralized transfer;

Calculate the risk weight for the transaction data x _i at the abnormal time point,

Among them, h _xi is the dynamic change component of transaction data _xi at abnormal time point;

Calculate the risk weight for the abnormal transaction data _yi of the same amount,

为异常相同数额交易数据y_i的动态变化分量；in,

is the dynamic change component of abnormal transaction data _yi of the same amount;

Calculate the risk weight for the abnormal over-limit transaction data _zi ,

The definition of comprehensive risk judgment model:

为异常大量资金转入快速分散转出的交易数据预测值；

为异常大量资金转入快速分散转出的交易数据的判断阈值，

为异常大量分散资金转入快速集中转出的交易数据预测值；

为异常大量分散资金转入快速集中转出的交易数据的判断阈值，

为异常时间点交易数据预测值；

为异常时间点交易数据的判断阈值，

为异常相同数额交易数据预测值；

为异常相同数额交易数据的判断阈值，

为异常超限额交易数据预测值；

为异常超限额交易数据的判断阈值，ε为判断校正系数。in,

Predicted value of transaction data for abnormally large amount of funds transferred in and out quickly and dispersedly;

It is the judgment threshold for the transaction data of abnormally large amount of funds transferred in and quickly dispersed and transferred out.

Predicted value of transaction data for the transfer of abnormally large amount of scattered funds in and out of rapid centralized transfer;

It is the judgment threshold for the transaction data of abnormally large amount of decentralized funds transferred into and out of rapid centralized transfer.

Predicted values for transaction data at abnormal time points;

is the judgment threshold of transaction data at abnormal time points,

Predicted values for abnormal transaction data of the same amount;

is the judgment threshold for abnormal transaction data of the same amount,

It is the predicted value of abnormal over-limit transaction data;

is the judgment threshold of abnormal over-limit transaction data, and ε is the judgment correction coefficient.

To sum up, due to the adoption of the above-mentioned technical solutions, the beneficial effects of the present invention are:

In schools, government agencies or large chain enterprises, the total amount of transactions and the number of transactions are even more difficult to count. Traditional computer accumulation methods or statistical principles cannot quickly and accurately obtain transaction behaviors containing transaction risks, and even if some extraction methods are used. Algorithms, counting in the total amount of financial transactions and the number of transactions, are inaccurate in the process of controlling and checking abnormal financial data. In particular, there are many unreasonable convergence thresholds and judgment conditions in the calculation process in the data screening process. This application uses a neural network learning algorithm to extract relatively accurate abnormal transaction behaviors, and uses the big data platform to screen, evaluate and evaluate abnormal behaviors. Risk control improves work efficiency and improves data security, predictability and transaction stability in the process of extracting and managing massive financial data.

Additional aspects and advantages of the present invention will be set forth, in part, from the following description, and in part will be apparent from the following description, or may be learned by practice of the invention.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of embodiments taken in conjunction with the accompanying drawings, wherein:

Figure 1 is an overall schematic diagram of the present invention.

Detailed ways

The following describes in detail the embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary, only used to explain the present invention, and should not be construed as a limitation of the present invention.

As shown in Figure 1, the present invention discloses a financial data management control system for a big data platform, comprising the following steps:

S1, obtain financial data through the cloud database, log in the abnormal data in the financial data, and then start a preliminary query, and perform real-time query, verification and extraction of invalid data during the query process;

S2, after real-time query verification and extraction, set a judgment interval for abnormal data, and form standardized data in the judgment interval;

S3, carry out deviation degree analysis on the standardized data, after the analysis, carry out the screening operation on the abnormal data through the screening model, and carry out the feature score for the abnormal data after screening,

S4, after feature scoring, the risk degree of abnormal data in the financial data is judged and output by a comprehensive risk judgment model.

The S1 includes:

S1-1, retrieve financial data from the cloud database, and obtain abnormal data from the financial data. The abnormal data extraction process performs data balance through the preliminary query process to dynamically request the financial data of the cloud database, and uses dynamic configuration to set abnormal data. According to the security control mechanism and authority management requirements of different financial data, different abnormal data are extracted for login operations.

S1-2, in the preliminary query process, the cloud database stores financial data authentication and functional access authority information in the local database, and performs unified financial data authentication and functional authority control; for financial data, logically isolates abnormal data and stores it in an independent In the database; the user's identity is verified during the financial data login process, and the abnormal data set that the user has access to is constructed according to the abnormal data access permission information in the financial data, and the access is authenticated through the identity authentication process of the cloud database; if the access fails, an exception is returned. Data access failure information; if the access is successful, the login is successful; establish an independent channel between the user and the application server instance dynamically allocated by the system,

S1-3, the process of accessing and using abnormal data is: forming abnormal data relationship nodes according to multiple abnormal data, searching for PaaS platform resources to convert them into tree nodes, generating a list of abnormal data tree nodes, and converting empty abnormal data The node set is used as the current node set to traverse the current abnormal data tree node set, so as to determine whether the abnormal data parent resource information list of the current traversing node set is equal to the preset abnormal data root node information list , if it is equal to the preset exception data root node information list, the node set of the current traversal operation is the root node of the current exception data permission tree, if not equal to the preset exception data root node information list, continue to traverse the exception The resource whose data identifier is equal to the parent resource information list of the node of the current traversal operation is marked as the abnormal data parent node of the node of the current traversal operation,

S1-4, for whether the abnormal data node is equal to the parent resource information list of the currently traversed node, it is judged whether the current tree node list is traversed; if the traversal is completed, the abnormal data parent node information list is detected. Then, the current abnormal data parent node information list is used as the root node of the current tree node, and the abnormal data business query tree is constructed recursively by marking; multiple abnormal data query requests allocated on a certain abnormal data node are reassigned to the abnormal data certain node. a computing node and backup, such that each of the certain computing node and the backup computing node is assigned only one subquery;

The query process of abnormal data through the node tree method plays the role of preliminary judgment of abnormal data. If further data extraction is performed, the data needs to be deeply excavated.

The S2 includes:

S2-1, after the query and verification, the abnormal data is divided into judgment intervals, and the similarity of the abnormal data is calculated, so as to generate the judgment interval, and the abnormal data is standardized through the scaling calculation, and the abnormally large amount of funds are transferred to the transaction that is quickly dispersed and transferred out. The conversion value of data ui is u′ _i , the conversion value of transaction data vi of abnormally large amount of scattered funds is transferred in and fast centralized transfer is v′ _{i , the conversion value of transaction data x i at} abnormal time point is x′ _{i ,} abnormal The conversion value of transaction data _{yi of the same amount is y′ i ,} and the conversion value of abnormal over-limit transaction data _zi is z′ _k ;

Substitute the transformed abnormal transaction data into the judgment model together with the time and date variables, and calculate the judgment value of the abnormal data at any time and date of the statistics:

Among them, B(t,d) is the judgment value of abnormal data at any time t and date d; f( _u′i ; t,d) is the time and date of an abnormally large amount of funds transferred in and out of transaction data quickly dispersed The judgment value of ; f(v′ _i ; t, d) is the judgment value of the time and date of the transaction data of an abnormally large amount of decentralized funds transferred into the fast centralized transfer; f(x′ _i ; t, d) is an abnormal The judgment value of the time and date of the transaction data at the time point; f(y′ _i ; t, d) is the judgment value of the time and date of an abnormal transaction data of the same amount; f (z′ _i ; t, d) is an abnormal Judgment value of over-limit transaction data; the maximum value of i is 60 to ensure that every second of abnormal data within one minute is monitored and judged in real time;

S2-2: Calculate the difference between the actual and judged values of each abnormal data on time and date, and perform a linear curve fitting process on the discrete abnormal data through the residual sum of squares, so as to judge the risk trend of the abnormal data.

Among them, W is the residual sum of squares of each abnormal data; B ₀ (t,d) is the actual value of each abnormal data at this time and date; B(t, d) is the sum of each abnormal data at this time Judgment value of the date; M is the time with the largest statistical time or the maximum number of days of the date;

S2-3, then calculate the deviation of abnormal data

Among them, F is the calculation constant, which is adjusted by the adjustment coefficient λ. Since W increases, the calculation constant F is larger; the value H _j is accurately obtained by increasing the abnormal data, and the accurately obtained values are accumulated for all N abnormal data. The deviation of the eigenvalue e converges, and β is the feature threshold.

The S3 includes:

和

计算得到：S3-1, after analyzing the deviation degree of the abnormal data, calculate the statistical information in the abnormal data through the prior probability distribution; attribute, wherein the first abnormal data set includes u _i and v _i , and the second abnormal data set includes _xi , _yi and z _i , by defining the time class attribute G and date class attribute I of the abnormal data, under the condition of probability distribution Calculate conditional probabilities separately

and

Calculated:

continue to derive,

表示第一异常数据集合

和时间类属性G和日期类属性I联合概率分布，遍历第一异常数据集合

和时间类属性G的全部值得到其条件概率分布

以及第一异常数据集合

和日期类属性I的全部值得到其条件概率分布

时间类属性条件概率Q(G)，日期类属性条件概率Q(I)；in

Represents the first abnormal data set

Joint probability distribution with time class attribute G and date class attribute I, and traverse the first abnormal data set

and all the values of the time class attribute G to get its conditional probability distribution

and the first abnormal data set

and all values of the date class attribute I get its conditional probability distribution

Time class attribute conditional probability Q(G), date class attribute conditional probability Q(I);

Then calculate:

continue to derive,

表示第二异常数据集合

和时间类属性G和日期类属性I联合概率分布，遍历第二异常数据集合

和时间类属性G的全部值得到其条件概率分布

以及第二异常数据集合

和日期类属性I的全部值得到其条件概率分布

in

Represents the second abnormal data set

Joint probability distribution with time class attribute G and date class attribute I, and traverse the second abnormal data set

and all the values of the time class attribute G to get its conditional probability distribution

and the second set of abnormal data

and all values of the date class attribute I get its conditional probability distribution

S3-2, the joint probability distribution value of the condition information of each abnormal data node in the first abnormal data set C and the time attribute and date attribute of the abnormal data nodes in the second abnormal data set E is as follows;

Select the class attribute J of the abnormal data and put it into the big data platform; take the class attribute J as the parent node of the internal attributes of the first abnormal data set C and the second abnormal data set E, and construct a Naive Bayesian network;

则将

放入网络作为其父节点；；若第二异常数据集合E中

则将

放入网络作为其父节点；从而得到用于异常数据等级筛选排序的贝叶斯网络；S3-3, put the nodes in the first abnormal data set C and the nodes in the second abnormal data set E into the Bayesian network one by one;

will

Put it into the network as its parent node; if the second abnormal data set E is

will

Put it into the network as its parent node; thus get a Bayesian network for the level screening and sorting of abnormal data;

得到异常数据中属性值最突出的概率分布；S3-4, calculate the probability mass function of class attribute J

Obtain the probability distribution with the most prominent attribute value in the abnormal data;

表示J关联的所有第一异常数据集合C中和第二异常数据集合E节点的条件概率的乘积；由大数据平台对财务异常数据根据概率分布的情况，向贝叶斯网络中第一异常数据集合C中和第二异常数据集合E的各个属性节点赋值；轮流将异常数据的基本属性代入贝叶斯网络中通过概率质量函数；按照计算数值从大到小的顺序对异常数据排列。in

Represents the product of the conditional probabilities of all the first abnormal data set C associated with J and the second abnormal data set E; the big data platform sends the financial abnormal data to the first abnormal data in the Bayesian network according to the probability distribution. Assign values to each attribute node in the set C and the second abnormal data set E; alternately substitute the basic attributes of the abnormal data into the Bayesian network to pass the probability mass function; arrange the abnormal data in descending order of the calculated values.

The S4 includes:

S4-1, combined with the calculation of the risk degree weight, calculate the transaction data _ui of the abnormally large amount of funds transferred in and out quickly and dispersedly:

Among them, T _total is the total reference time; p _ui is the dynamic change component of the transaction data weight of the abnormally large amount of funds transferred in and out quickly and dispersedly; V _total is the total reference date, U is the transaction data detection time component; K is the transaction data detection date component;

S4-2, calculate the risk weight for the transaction data v _i of the abnormally large amount of decentralized funds transferred in and quickly centralized transferred out,

为异常大量分散资金转入快速集中转出的交易数据v_i权重的动态变化分量；in,

It is the dynamic change component of the weight of transaction data v _i for the transfer of an abnormally large amount of decentralized funds in and out of the rapid centralized transfer;

S4-3, calculate the risk degree weight for the transaction data x _i at the abnormal time point,

为异常时间点交易数据x_i的动态变化分量；in,

is the dynamic change component of transaction data _xi at abnormal time point;

S4-4, calculate the risk weight for the abnormal transaction data _yi of the same amount,

为异常相同数额交易数据y_i的动态变化分量；in,

is the dynamic change component of abnormal transaction data _yi of the same amount;

S4-5, calculate the risk weight for the abnormal over-limit transaction data _zi ,

The definition of comprehensive risk judgment model:

为异常大量资金转入快速分散转出的交易数据预测值；

为异常大量资金转入快速分散转出的交易数据的判断阈值，

为异常大量分散资金转入快速集中转出的交易数据预测值；

为异常大量分散资金转入快速集中转出的交易数据的判断阈值，

为异常时间点交易数据预测值；

为异常时间点交易数据的判断阈值，

为异常相同数额交易数据预测值；

为异常相同数额交易数据的判断阈值，

为异常超限额交易数据预测值；

为异常超限额交易数据的判断阈值，ε为判断校正系数。in,

Predicted value of transaction data for abnormally large amount of funds transferred in and out quickly and dispersedly;

It is the judgment threshold for the transaction data of abnormally large amount of funds transferred in and quickly dispersed and transferred out.

Predicted value of transaction data for the transfer of abnormally large amount of scattered funds in and out of rapid centralized transfer;

It is the judgment threshold for the transaction data of abnormally large amount of decentralized funds transferred into and out of rapid centralized transfer.

Predicted values for transaction data at abnormal time points;

is the judgment threshold of transaction data at abnormal time points,

Predicted values for abnormal transaction data of the same amount;

is the judgment threshold for abnormal transaction data of the same amount,

It is the predicted value of abnormal over-limit transaction data;

is the judgment threshold of abnormal over-limit transaction data, and ε is the judgment correction coefficient.

Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, The scope of the invention is defined by the claims and their equivalents.

Claims (7)

1. a big data platform financial data management control system, is characterized in that, comprises: The query extraction module is used to obtain financial data through the cloud database, log in the abnormal data in the financial data, and then start a preliminary query, and perform real-time query verification and extraction of invalid data during the query process; The abnormality judgment module is used to set the judgment interval of abnormal data after real-time query, verification and extraction, and form standardized data in the judgment interval; The abnormality judgment module includes: After the query and verification, the abnormal data is divided into judgment intervals, and the similarity of the abnormal data is calculated to generate the judgment _interval . The conversion value is u′ _i , the conversion value of transaction data vi of abnormally large amount of scattered funds is transferred in and fast centralized transfer out is v′ _i , the conversion value of transaction data _xi at abnormal time point is x′ _{i ,} abnormal transaction data of the same amount The conversion value of y _i is y′ _i , and the conversion value of abnormal over-limit transaction data _zi is z′ _k ; Substitute the transformed abnormal transaction data into the judgment model together with the time and date variables, and calculate the judgment value of the abnormal data at any time and date of the statistics:

Among them, B(t,d) is the judgment value of abnormal data at any time t and date d; f( _u′i ; t,d) is the time and date of an abnormally large amount of funds transferred in and out of transaction data quickly dispersed The judgment value of ; f(v′ _i ; t, d) is the judgment value of the time and date of the transaction data of an abnormally large amount of decentralized funds transferred into the fast centralized transfer; f(x′ _i ; t, d) is an abnormal The judgment value of the time and date of the transaction data at the time point; f(y′ _i ; t, d) is the judgment value of the time and date of an abnormal transaction data of the same amount; f (z′ _i ; t, d) is an abnormal Judgment value of over-limit transaction data; the maximum value of i is 60 to ensure that every second of abnormal data within one minute is monitored and judged in real time; Calculate the difference between the actual and judged values of each abnormal data on time and date, and perform a linear curve fitting process on the discrete abnormal data through the residual sum of squares, so as to judge the risk trend of abnormal data.

Among them, W is the residual sum of squares of each abnormal data; B ₀ (t,d) is the actual value of each abnormal data at this time and date; B(t, d) is the sum of each abnormal data at this time Judgment value of the date; M is the time with the largest statistical time or the maximum number of days of the date; Calculate the deviation of abnormal data

Among them, F is the calculation constant, which is adjusted by the adjustment coefficient λ, and the value H _j is accurately obtained by adding abnormal data. After accumulating the accurately obtained values of all N abnormal data, the deviation convergence of the characteristic value e is performed, and β is the characteristic threshold; The screening and scoring module is used to analyze the deviation degree of the standardized data. After the analysis, the abnormal data is screened by the screening model, and the characteristic score of the abnormal data after screening is performed. The comprehensive judgment module is used to judge and output the risk degree of abnormal data in the financial data through the comprehensive risk judgment model after characteristic scoring. 2. The big data platform financial data management control system according to claim 1, wherein the query extraction module comprises: The financial data is retrieved from the cloud database, and abnormal data is obtained from the financial data. The abnormal data extraction process performs data balance through the preliminary query process to dynamically request the financial data of the cloud database, and uses the dynamic configuration method to set the acquisition threshold of abnormal data. According to the security control mechanism and authority management requirements of different financial data, extract different abnormal data for login operation, During the initial query process, the cloud database stores financial data authentication and functional access authority information in the local database, and performs unified financial data authentication and functional authority control; for financial data, abnormal data is logically isolated and stored in an independent database; During the data login process, the user's identity is verified, and the abnormal data set that the user has access to is constructed according to the abnormal data access authority information in the financial data, and the authentication access is performed through the identity authentication process of the cloud database; if the access fails, the abnormal data access failure information is returned. ; If the access is successful, the login is successful; establish an independent channel between the user and the application server instance dynamically allocated by the system. 3. The big data platform financial data management control system according to claim 1, wherein the query extraction module comprises: The process of accessing and using abnormal data is to form abnormal data relationship nodes according to multiple abnormal data, search for PaaS platform resources to convert them into tree nodes, generate a list of abnormal data tree nodes, and use the empty abnormal data node set as the For the current node set, perform a traversal operation on the current abnormal data tree node set, so as to determine whether the abnormal data parent resource information list of the node set of the current traversal operation is equal to the preset abnormal data root node information list, if it is equal to the preset abnormal data root node information list If it is not equal to the preset abnormal data root node information list, the node set of the current traversal operation is the root node of the current abnormal data permission tree. The resource of the parent resource information list of the node of the traversal operation is marked as the abnormal data parent node of the node of the current traversal operation. 4. The big data platform financial data management control system according to claim 1, wherein the query extraction module comprises: For whether the abnormal data node is equal to the parent resource information list of the currently traversed node, it is judged whether the current tree node list has been traversed; if the traversal is completed, the abnormal data parent node information list is detected. The data parent node information list is used as the root node of the current tree node, and the abnormal data service query tree is constructed recursively by marking; multiple abnormal data query requests allocated on a certain abnormal data node are reassigned to a certain abnormal data computing node and Backup so that each of the certain compute node and the backup compute node is assigned only one subquery. 5. The big data platform financial data management control system according to claim 1, wherein the screening and scoring module comprises: After the deviation degree of abnormal data is analyzed, statistical information is calculated in the abnormal data through the prior probability distribution; the prior conditional probability distribution of abnormal data is calculated, and the internal attributes of the first abnormal data set C and the second abnormal data set E are set, wherein the first abnormal data set C and the second abnormal data set E are set. An abnormal data set includes u _i and v _i , and the second abnormal data set includes _xi , _yi and _zi . By defining the time class attribute G and date class attribute I of the abnormal data, the conditional probability is calculated respectively under the condition of probability distribution

and

Calculated:

continue to derive,

in

Represents the first abnormal data set

Joint probability distribution with time class attribute G and date class attribute I, and traverse the first abnormal data set

and all the values of the time class attribute G to get its conditional probability distribution

and the first abnormal data set

and all values of the date class attribute I get its conditional probability distribution

Time class attribute conditional probability Q(G), date class attribute conditional probability Q(I); Then calculate:

continue to derive,

in

Represents the second abnormal data set

Joint probability distribution with time class attribute G and date class attribute I, and traverse the second abnormal data set

and all the values of the time class attribute G to get its conditional probability distribution

and the second set of abnormal data

and all values of the date class attribute I get its conditional probability distribution

6. The big data platform financial data management control system according to claim 1, wherein the screening and scoring module comprises: The joint probability distribution value of the condition information of the time attribute and date attribute of each abnormal data node in the first abnormal data set C and each abnormal data node in the second abnormal data set E is as follows;

Select the class attribute J of the abnormal data and put it into the big data platform; take the class attribute J as the parent node of the internal attributes of the first abnormal data set C and the second abnormal data set E, and construct a Naive Bayesian network; Put the nodes in the first abnormal data set C and the second abnormal data set E into the Bayesian network one by one; if the first abnormal data set C

will

Put it into the network as its parent node; if the second abnormal data set E is

will

Put it into the network as its parent node; thus get a Bayesian network for rank screening and sorting of abnormal data. 7. The big data platform financial data management control system according to claim 1, wherein the comprehensive judgment module comprises: Combined with the calculation of risk weight, calculate the transaction data _ui of abnormally large amount of funds transferred in and out quickly and dispersedly:

Among them, T _total is the total reference time;

is the dynamic change component of the transaction data weight of abnormally large amount of funds transferred in and out quickly and dispersedly; V _total is the total base date, U is the transaction data detection time component; K is the transaction data detection date component; Calculate the risk weight for the transaction data v _i of the abnormally large amount of decentralized funds transferred in and out quickly and in a centralized manner,

in,

It is the dynamic change component of the weight of transaction data v _i for the transfer of an abnormally large amount of decentralized funds in and out of the fast centralized transfer; Calculate the risk weight for the transaction data x _i at the abnormal time point,

in,

is the dynamic change component of transaction data _xi at abnormal time point; Calculate the risk weight for the abnormal transaction data _yi of the same amount,

in,

is the dynamic change component of abnormal transaction data _yi of the same amount; Calculate the risk weight for the abnormal over-limit transaction data _zi ,

The definition of comprehensive risk judgment model:

in,

Predicted value of transaction data for abnormally large amount of funds transferred in and out quickly and dispersedly;

It is the judgment threshold for the transaction data of abnormally large amount of funds transferred in and out quickly and dispersedly.

Predicted value of transaction data for the transfer of abnormally large amount of decentralized funds into and out of rapid centralized transfer;

It is the judgment threshold for the transaction data of the abnormally large amount of decentralized funds transferred in and out of the rapid centralized transfer.

Predicted values for transaction data at abnormal time points;

is the judgment threshold of transaction data at abnormal time points,

Predicted values for abnormal transaction data of the same amount;

is the judgment threshold for abnormal transaction data of the same amount,

It is the predicted value of abnormal over-limit transaction data;

is the judgment threshold of abnormal over-limit transaction data, and ε is the judgment correction coefficient.

Images