Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a low-security fund supervision method and system based on supervised data association analysis, so that the effect of well discovering the low-security field and low-cost problems is achieved on the basis of solving the problems of irregular subsidy release, unobvious common characteristics of data and the like in massive low-security release data, and the efficiency of supervision work is improved.
In order to achieve the purpose, the invention adopts the technical scheme that:
a low-insurance fund supervision method based on supervision data association analysis comprises the following steps:
step 1, acquiring low-security user information data and one-card data of subsidy issuing, importing the low-security user information data and the one-card data of subsidy issuing into an Oracle database for data persistence processing, extracting low-security fund issuing data from the one-card data, removing interference information in low-security issuing from the low-security fund issuing data containing various subsidy items by using a preprocessing method facing the low-security user information data and the low-security fund issuing data, and extracting key fields in the low-security fund issuing data; firstly filtering data missing from the main user identity number and the member identity number in the low-security user information data, then obtaining a low-security user information data table and a low-security fund issuing data table, and counting the total amount and the total times of each subsidy received by each low-security user in a period of time;
the step is mainly used for extracting useful information related to low-security data in the mass civil fund data so as to ensure the calculation efficiency and the analysis precision of the subsequent analysis process.
Step 2, comparing the beneficiary identity number in the low-security fund distribution data table with the household main identity number and the household member identity number in the low-security fund information data table through data collision and decision analysis among association tables to obtain problem data that a part of beneficiaries receive the low-security fund without being in the low-security fund information data table, and extracting data of the real low-security fund received by the low-security household;
the step is mainly used for ensuring that the data of subsequent analysis are all true low-security users registered by related departments and finding out the possible phenomenon that non-low-security users receive low-security funds.
And 3, calculating the total amount and total times of earning of each low-security user in one year in the data of the actual low-security fund earning of the low-security user, and respectively calculating a data clustering center of the total amount and the total times of earning to obtain problem data which is separated from the clustering center, namely annual earning amount abnormal data or annual earning time abnormal data.
The step is mainly used for accurately clustering low-security issuing data with the same characteristics and accurately classifying behaviors of illegally getting low-security capital within a period of time.
In the step 1, for the acquired massive low-security-user information data and low-security-fund dispensing data, according to the low-security-fund dispensing rule, a preprocessing method facing the low-security-user information data and the low-security-fund dispensing data is used for screening out irrelevant interference information, and a method for extracting key fields comprises the following steps:
1) uniformly converting the low-security user information data and the one-card data for issuing subsidies into a CSV format, and performing data persistence processing by using an Oracle database;
2) extracting low-security fund issuing data from the one-card data, and extracting key field complete data including the amount of money received, the issuing date, the beneficiary identity card number, the city, the county, the district and the village group from the low-security fund issuing data; extracting complete data of the identity card numbers of the householder and the householder members from the low-security information data;
3) and screening the data processed in the two steps by using SQL language in an Oracle database according to the release year, the subsidy project code, the city and county fields to obtain a data table for receiving a certain subsidy project from a certain county of a certain city in a certain year, and counting the total sum and the total times of receiving the subsidy by each low insured user in a period of time.
3. The low-insurance fund supervision method based on supervised data association analysis as recited in claim 1, wherein in step 3, normal data common features are found through clustering, and problem data departing from a clustering center is found by using an unsupervised classification method based on clustering analysis.
The method for discovering the abnormal data of the annual pickup amount comprises the following steps:
the method comprises the following steps: there is a low-security fund release data set L, L ═ L1,l2,…,li,…,lnWhere {1,2, …, i, …, n } is the sequence number of the data, n is the number of data in the data set L,/iIndicates the ith data, lnRepresents the nth piece of data; randomly selecting K values from the annual total amount of money to be collected, { eta1,η2,…,ηi,…ηKK initial clustering centers respectively representing annual total money pickup, {1, …, i, …, K } indicating the serial number of data,. eta.iDenotes the ith initial clustering center, ηKRepresenting the Kth initial clustering center, defining the initial clustering center { eta1,η2,…,ηi,…ηKNeighborhood of { U }1,U2,…,Ui,…,UKValue of { lambda }1,λ2,…,λi,…λK}, initial clustering center ηiNeighborhood U ofiThe distance between the sample set L and the eta i is not more than lambdaiA subsample set of, i.e. Ui={li∈L|dis(li,ηi)≤λi},UKRepresenting the initial clustering center ηKIs a neighborhood ofiDenotes the ith value, λKRepresents the Kth value; { eta [. eta. ]1,η2,…,ηi,…ηKAnd the corresponding { lambda }1,λ2,…,λi,…λKForm a class cluster { Z }1,Z2,…,Zi,…,ZKWherein the ith class cluster Zi={ηi±λi}, the Kth class cluster ZK={ηK±λK}; for each value remaining in the annual total amount, calculate each value and { η [ ]1,η2,…,ηi,…ηKA distance, assigning the value to the nearest cluster; then calculating the mean value of each cluster, namely the clustering center, and continuously and repeatedly calculating the clustering centers of the clusters until the clustering centers of the clusters do not change;
step two: calculating L ═ L
1,l
2,…,l
i,…,l
nSet A of all data not in K clusters
Amount of moneyWherein
Denotes an empty set, A
Amount of moneyThere is a problem set for annual collection of amounts;
the method for discovering the abnormal data of the annual picking times comprises the following steps:
the method comprises the following steps: randomly selecting K values in the total annual picking times, { eta1',η2',…,ηi',…ηK' } K initial cluster centers for respectively taking total times of year, {1, …, i, …, K } denotes the number of data,. eta.i' denotes the ith initial clustering center, ηK' denotes the Kth initial clustering center, defining the initial clustering center { η1',η2',…,ηi',…ηK' } neighborhood U1',U2',…,Ui',…,UKA value of' } λ1',λ2',…,λi',…λK' } initial clustering center ηi' neighborhood Ui' includes the sum of the sample set L and ηi' is not more than λi' A subsample set, Ui'={li∈L|dis(li,ηi')≤λi'},UK' representing initial clustering center etaK' neighborhood, λi' denotes the ith value, λK' represents a Kth numerical value; { eta [. eta. ]1',η2',…,ηi',…ηK' } and corresponding λ1',λ2',…,λi',…λK' } construction of class clusters Z1',Z2',…,Zi',…,ZK' }, wherein the ith class cluster Zi'={ηi'±λi' } K class cluster ZK'={ηK'±λK' }; for each value remaining in the total annual count, calculate each value and { η }1',η2',…,ηi',…ηK' } assigning the value to the nearest cluster; then calculating the mean value of each cluster, namely the clustering center, and continuously and repeatedly calculating the clustering centers of the clusters until the clustering centers of the clusters are not changed;
step two: calculating L ═ L
1,l
2,…,l
i,…,l
nSet A of all data not in K clusters
Number of timesWherein
A
Number of timesThere is a problem set for annual pick-up times.
The mean calculation distance formula of the clusters is as follows:
where dis (l)
i,l
k) Is the ith data l
iAnd the k-th data l
kDistance, x
iAnd x
kAre each l
iAnd l
kThe numerical value of (c).
The invention also provides a low-security fund supervision system based on the supervised data association analysis, which comprises a data preprocessing module, an associated data collision and decision analysis module and an unsupervised classification module, wherein the step 1, the step 2 and the step 3 are respectively and correspondingly executed.
Compared with the prior art, the method has the advantages that aiming at the phenomenon that non-low-security users possibly receive low-security funds in the mass low-security fund issuing data and the characteristic that issuing standards of different cities and different years are inconsistent when the low-security fund issuing data is input in a basic department, a preprocessing method facing to the low-security user information data and the low-security fund issuing data is adopted, interference information in low-security issuing is eliminated, and key fields in the low-security fund issuing data are extracted; and obtaining a part of problem data which is not in the low-security-user information data table and receives the low-security-user funds by adopting data collision and decision analysis among the association tables, and extracting the data of the real low-security-user receiving the low-security-user funds; and extracting the suspected problem data of illegally drawing the low-security fund in one stage in the low-security issuing data by adopting a problem data analysis method in a period of time based on cluster analysis. Therefore, the problems that subsidy issuing is not unified in standard, issuing time is not fixed, temporary subsidies are completely filed as low insurance and the like in the existing low-insurance fund issuing data are well solved, and the efficiency of the supervision department for finding the problems in the field of civil fund issuing is improved.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail by embodiments with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Fig. 1 presents a schematic flow chart diagram in accordance with an embodiment of the present invention. In summary, the method comprises:
step 1: acquiring low-insurance-user information data and one-card data for issuing subsidies, uniformly converting the format into a CSV format, and performing data persistence processing by using an Oracle database; the method comprises the steps of extracting low-security fund issuing data from one-card data by using SQL language in an Oracle database, removing interference information in low-security issuing from the low-security fund issuing data containing multiple types (300 types) of subsidy items by using a preprocessing method facing the low-security user information data and the low-security fund issuing data, and extracting key fields in the low-security fund issuing data. Specifically, data of missing fields of a receiving amount, a release date, a beneficiary number, a city and a county are removed, complete data of key fields of a receiving amount, a release date, a beneficiary number (namely, a beneficiary identity number), a city, a county, a district, a village group and the like in low-security fund release data are extracted, data of missing identity numbers of a householder and a member of the householder are firstly filtered in low-security user information data, and complete data of identity numbers of the householder and the member of the householder are extracted, so that a low-security user information data table and a low-security fund release data table are obtained. Then, SQL language is used in Oracle database to screen the processed data according to the release year, subsidy project code, city, county and other fields, so as to obtain a data table for getting some subsidy project from some county in some city in some year. And (3) processing the data obtained through the steps by using an SQL language in an Oracle database, and counting the total sum and the total times of getting various subsidies by a certain low security holder in a period of time.
Generally, the information data of the low-security deposit can be provided by a civil department, the one-card data for issuing subsidies can be provided by a financial department, and the one-card data comprises low-security fund issuing data. The two data formats of the low-security fund issuing data and the low-security user information data are greatly different, direct correlation analysis is difficult to perform, and the data characteristics are targeted. Therefore, the invention uniformly converts the low-security fund data format from each department into the CSV format and uses the Oracle database for data persistence processing.
Step 2: minimum life support fund release rules, the beneficiary in the low-protection fund release data is a member of the low-protection family. For the possible phenomenon that non-low-security users receive low-security funds, the beneficiary identity number in the low-security fund issuing data table is compared with the household main identity number and the household member identity number in the low-security user information data table through data collision and decision analysis among the association tables to obtain problem data that a part of beneficiaries do not receive the low-security funds in the low-security user information data table, and data of really low-security users receiving the low-security funds are extracted.
And 3, when the low-security fund issuing data is recorded in the basic department, two special conditions exist. The first is that some subsidy items are issued without proper filing categories in the database, and all the subsidy items are filed as the minimum life guarantee when the subsidy items are recorded in the database by the basic department. The second is temporary subsidies such as price escalation subsidies and winter heating subsidies which are temporarily and irregularly issued by local governments and civil administration departments, and all the subsidies are filed as the lowest life guarantee when the subsidies are recorded in a database, so that the problem that whether data exist cannot be judged by using low-security fund issuance standards is caused. Aiming at the characteristic of issuing low-security funds, in the data of the actual low-security fund collection of each low-security household, the total collection amount and the total collection times of each low-security household in one year are calculated, and the data clustering centers of the total collection amount and the total collection times are respectively calculated to obtain problem data which is separated from the clustering centers, namely annual collection amount abnormal data or annual collection times abnormal data.
Specifically, common features of normal data are found through clustering, and problem data which are separated from a clustering center are found by using an unsupervised classification method based on clustering analysis.
The method for discovering the abnormal data of the annual pickup amount comprises the following steps:
the method comprises the following steps: there is a low-security fund release data set L, L ═ L1,l2,…,li,…,lnWhere {1,2, …, i, …, n } is the sequence number of the data, n is the number of data in the data set L,/iIndicates the ith data, lnRepresents the nth piece of data; randomly selecting K values from the annual total amount of money to be collected, { eta1,η2,…,ηi,…ηKK initial clustering centers respectively representing annual total money pickup, {1, …, i, …, K } indicating the serial number of data,. eta.iDenotes the ith initial clustering center, ηKRepresenting the Kth initial clustering center, defining the initial clustering center { eta1,η2,…,ηi,…ηKNeighborhood of { U }1,U2,…,Ui,…,UKValue of { lambda }1,λ2,…,λi,…λK}, initial clustering center ηiNeighborhood U ofiIncludes the sum of L and η of the sample setiIs not more than lambdaiA subsample set of, i.e. Ui={li∈L|dis(li,ηi)≤λi},UKRepresenting the initial clustering center ηKIs a neighborhood ofiDenotes the ith value, λKRepresents the Kth value; { eta [. eta. ]1,η2,…,ηi,…ηKAnd the corresponding { lambda }1,λ2,…,λi,…λKForm a class cluster { Z }1,Z2,…,Zi,…,ZKWherein the ith class cluster Zi={ηi±λi}, the Kth class cluster ZK={ηK±λK}; for each value remaining in the annual total amount, calculate each value and { η [ ]1,η2,…,ηi,…ηKA distance, assigning the value to the nearest cluster; then calculating the mean value of each cluster, namely the clustering center, and continuously and repeatedly calculating the clustering centers of the clusters until the clustering centers of the clusters do not change;
step two: calculating L ═ L
1,l
2,…,l
i,…,l
nSet A of all data not in K clusters
Amount of moneyWherein
Denotes an empty set, A
Amount of moneyThere is a problem set for annual collection of amounts;
the method for discovering the abnormal data of the annual picking times comprises the following steps:
the method comprises the following steps: randomly selecting K values in the total annual picking times, { eta1',η2',…,ηi',…ηK' } K initial cluster centers for respectively taking total times of year, {1, …, i, …, K } denotes the number of data,. eta.i' denotes the ith initial clustering center, ηK' denotes the Kth initial clustering center, defining the initial clustering center { η1',η2',…,ηi',…ηK' } neighborhood U1',U2',…,Ui',…,UKA value of' } λ1',λ2',…,λi',…λK' } initial clustering center ηi' neighborhood Ui' includes the sum of the sample set L and ηi' is not more than λi' A subsample set, Ui'={li∈L|dis(li,ηi')≤λi'},UK' representing initial clustering center etaK' neighborhood, λi' denotes the ith value, λK' represents a Kth numerical value; { eta [. eta. ]1',η2',…,ηi',…ηK' } and corresponding λ1',λ2',…,λi',…λK' } construction of class clusters Z1',Z2',…,Zi',…,ZK' }, wherein the ith class cluster Zi'={ηi'±λi' } K class cluster ZK'={ηK'±λK' }; for each value remaining in the total annual count, calculate each value and { η }1',η2',…,ηi',…ηK' } assigning the value to the nearest cluster; then calculating the mean value of each cluster, namely the clustering center, and continuously and repeatedly calculating the clustering centers of the clusters until the clustering centers of the clusters are not changed;
step two: calculating L ═ L
1,l
2,…,l
i,…,l
nSet A of all data not in K clusters
Number of timesWherein
A
Number of timesThere is a problem set for annual pick-up times.
In the invention, the cluster mean value calculation distance formula is as follows:
where dis (l)
i,l
k) For the ith data l
iAnd the k-th data l
kDistance, x
iAnd x
kAre each l
iAnd l
kThe numerical value of (c).
The invention also provides a low-security fund supervision system based on the supervised data association analysis, which comprises a data preprocessing module, an associated data collision and decision analysis module and an unsupervised classification module, wherein the step 1, the step 2 and the step 3 are respectively and correspondingly executed.
The overall implementation of the above method is illustrated by a specific example.
The embodiment is established on a cloud computing platform, the platform consists of 15 servers, comprises Vmware Esxi 5,20T disk arrays and 1000M network switches, and is provided with a Hadoop cluster. In this embodiment, 2018-year low-security fund issuance data table is obtained by using 2018-year town minimum life support fund issuance data as an analysis object. For convenience of presentation, the step of extracting the low-guarantee fund issuing data from the one-card fund issuing data is omitted, and the step of screening according to city and county in the steps is omitted. And selecting the small amount of low-security fund issuance data in 2018 as a display in the following table, and deleting the data of fields such as the shortage of issuance amount, issuance date, city, county, subsidy project code and the like. The low-security funds release data includes 12 fields, respectively: { city, county, town street, village, group, subsidy object code, subsidy object, beneficiary code, subsidy project name, issuance amount, issuance date }.
The method comprises the following steps: performing correlation analysis and data collision on the 2018 low-security fund issuing data table and the low-security user information table, and comparing the beneficiary codes in the 2018 low-security fund issuing data table with the user member identity card account numbers in the low-security user information table to obtain a part of problem data which is not in the low-security user information data table and is used for picking up low-security funds; and extracts data of the actual low-security fund collected by the low-security user in the 2018 low-security fund release data table.
Step two: calculating the total times and the total amount of all low-security households for receiving the low-security funds in 2018; clustering the annual total amount of money, setting the number K of clustering centers to be 3, and setting the initial clustering center eta to be 31=6000, η2=12000,η317000, calculate the distance of all data from the three initial cluster centers, assign data to Z according to distance1、Z2、Z3In the three initial clusters, then calculating the mean value in the cluster, namely the final cluster center, and respectively eta ″1=6300,η″2=12225,η″316751 sets η ″1、η″2、η″3Neighborhood U of1、U2、U3A value of (A)1、 λ2、λ32000, 1500, 1000 respectively. Three clusters were obtained to obtain three Z ″)1=6300±2000、 Z″2=12225±1500、Z″316751 +/-1000, and then calculating the data which is not in the three clusters, namely acquiring the suspected problem data of the low-security fund in 2018. Similar methods can calculate data with problems in annual access to the total number of low-security times, and this embodiment is not described herein again. In the embodiment, in 2018, 762708 pieces of data are preprocessed, and a possible non-low-security consumer receives 22 consumers with low-security; and the suspicious problem data 411 of the total amount of the low-security fund picked in 2018 of the county are found to be similar to the suspicion of the total annual picking times of the low-security fund2501 pieces of problem data, 384 pieces of problem data exist in the total amount of the low-security fund taken annually and the total times of annual taking.
Through the implementation steps of the invention, it can be seen that 762708 pieces of civil fund issuing data in the one-card data for issuing subsidies in the embodiment are subjected to screening out useless information, extracting key fields, performing association table collision analysis and clustering analysis methods, so that the data dimension is reduced, and the problems that the subsidies issuing in the existing low-security fund issuing data is not unified in standard, the issuing time is not fixed, all temporary subsidies are filed as the lowest life guarantee and the like are solved; and finds that there may be non-low-security households claiming 22 households with low-security impressions; and finding that 411 pieces of suspected problem data of the total amount of the low-security fund in 2018 of the county, 2501 pieces of suspected problem data of the total annual times of the low-security fund, 384 pieces of problem data of the total amount of the low-security fund taken annually and the total annual times of the low-security fund taken annually exist. The minimum life insurance fund supervision method based on the supervision data association analysis comprehensively considers the characteristics of low insurance fund issuing rules and low insurance fund data. In the practical example, the method and the system process mass low-security fund release data, and the efficiency of finding suspected problems is improved by nearly 1000 times compared with the efficiency of a traditional manual method.
In conclusion, the method analyzes the common characteristics of the data and finds the behavior of taking the low-security fund in a suspected violation in one stage on the basis of massive low-security data preprocessing and associated data collision and decision analysis, thereby realizing the minimum life support fund supervision of the high-efficiency supervision data association analysis.
Although the present invention has been described by way of preferred embodiments, the present invention is not limited to the embodiments described herein, and various changes and modifications may be made without departing from the scope of the present invention.