Fund supervision method, client, medium and system

Abstract

The invention provides a fund supervision method and system, a client and a medium. According to the method, screening processing, cleaning processing, integration processing and conversion processing are carried out on local business data, basic data are obtained, and the basic data comprise historical contract information, historical budget information and historical fund transfer information; performing feature alignment processing on the basic data by adopting an asymmetric encryption protocol to obtain aligned basic data; and the initial federated learning model is trained by adopting the aligned basic data, and the final federated learning model is determined until the initial federated learning model converges, so that compared with the existing scheme, the artificial intelligence model can be trained in a safer and effective output mode for the multi-source data; therefore, the security of the data and the accuracy of subsequent model prediction are improved, and the problem of how to ensure the data privacy and security of the fund supervision system while using the multi-data-source data to perform artificial intelligence model training is solved.

Description

Technical Field

[0001] The present application relates to the technical field of fund supervision, and specifically, to a fund supervision method, client, medium and system. Background Art

[0002] The supervision method of most existing fund supervision systems is: the fund user stores the fund usage data in the supervision system, and the supervisor manually checks and inspects the fund user's fund operation data. This method cannot perform correlation analysis through system algorithms, and the system cannot automatically calculate and determine whether the business data involves abnormal information, abnormal fund flow, and other scenarios, and cannot automatically trigger system warnings. If the abnormal fund destination is missed manually, the actual cost of recovery is very high, and multiple data sources cannot guarantee the accuracy of the model.

[0003] How to ensure the data privacy and security of the fund supervision system while using data from multiple data sources to train artificial intelligence models has become an urgent problem to be solved. Summary of the invention

[0004] The main purpose of this application is to provide a fund supervision method, client, medium and system to at least solve the problem of how to ensure the data privacy and security of the fund supervision system while using data from multiple data sources for artificial intelligence model training.

[0005] In order to achieve the above object, according to one aspect of the present application, a fund supervision method is provided, which is applied to each client in a fund supervision system, and the method includes:

[0006] Preprocessing the local business data to obtain basic data, wherein the preprocessing includes screening, cleaning, integration and conversion, wherein the integration represents collating and merging the local business data according to the association relationship, and the basic data includes historical contract information, historical budget information and historical fund disbursement information;

[0007] Performing feature alignment processing on the basic data using an asymmetric encryption protocol to obtain aligned basic data; and using the aligned basic data to train an initial federated learning model until a final federated learning model is determined when the initial federated learning model converges;

[0008] The final federated learning model is used to predict the real-time data to determine whether the current fund flow is abnormal, and if it is determined that the current fund flow is abnormal, an early warning message is generated to indicate that the current fund flow is abnormal.

[0009] Optionally, the fund supervision system further includes a central server, which trains the initial federated learning model using the aligned basic data until the final federated learning model is determined when the initial federated learning model converges, including: a first receiving step: receiving the initial encryption parameters sent by the central server, where the initial encryption parameters represent the encrypted data of the initial parameters of the initial federated learning model; a first processing step: locally training the initial federated learning model using the aligned basic data and the initial encryption parameters, calculating the training gradient and loss value of the initial federated learning model to update the initial federated learning model; a second processing step: uploading the parameters of the trained initial federated learning model to the central server using encryption or differential privacy technology, so that the central server performs weighted averaging on the parameters of the trained initial federated learning model of each client to obtain corrected parameters and send the corrected parameters to each client; a second receiving step: receiving the corrected parameters sent by the central server, updating the parameters of the trained initial federated learning model using the corrected parameters, and repeating the first receiving step, the first processing step, and the second processing step until the final federated learning model is determined when the initial federated learning model converges.

[0010] Optionally, after determining the final federated learning model, the method further includes: every preset period, based on the newly added associated business feature data and warning situations within the preset period, performing the first receiving step, the first processing step, the second processing step, and the second receiving step to update the final federated learning model.

[0011] Optionally, preprocessing the local business data to obtain basic data includes: screening the local business data to obtain screened data to filter out the feature data for abnormal warning of fund flow in federated learning; cleaning the screened data to remove the incomplete data in the screened data to obtain cleaned data; organizing and merging the data from different businesses and different data table structures in the cleaned data according to the association relationship guided by the feature dimension of the data to obtain integrated processed data; performing conversion processing on the integrated processed data to obtain the basic data, so that the integrated processed data meets the data specifications of the initial federated learning model.

[0012] Optionally, use the final federated learning model to predict real-time data to determine whether the current fund flow is abnormal, including: using the final federated learning model to predict real-time data to obtain an abnormal confidence level; determining that the current fund flow is abnormal when the abnormal confidence level is greater than or equal to the confidence level threshold; and determining that the current fund flow is normal when the abnormal confidence level is less than the confidence level threshold.

[0013] Optionally, use an asymmetric encryption protocol to perform feature alignment processing on the basic data to obtain aligned basic data, including: based on the asymmetric encryption protocol, combining the feature values of the local business data and the generation time of the local business data, performing feature alignment processing on the basic data to obtain the aligned basic data.

[0014] Optionally, based on the asymmetric encryption protocol, combining the feature values of the local business data and the generation time of the local business data, performing feature alignment processing on the basic data to obtain the aligned basic data, including: extracting the first feature value and time feature to be aligned from the local business data, and using the time feature as the second feature value; using an asymmetric encryption algorithm to encrypt the first feature value and the second feature value to obtain first encrypted data and second encrypted data; aligning and matching the first encrypted data and the second encrypted data with the corresponding features in other data sources or data sets respectively to establish a corresponding relationship to obtain the matched data; and integrating the matched data to obtain the aligned basic data.

[0015] According to another aspect of the present application, a client is provided, and the client includes:

[0016] A first processing unit, configured to preprocess local business data to obtain basic data, where the preprocessing includes screening processing, cleaning processing, integration processing, and transformation processing, the integration processing represents organizing and merging the local business data according to the association relationship, and the basic data includes historical contract information, historical budget information, and historical fund appropriation information;

[0017] A second processing unit, configured to use an asymmetric encryption protocol to perform feature alignment processing on the basic data to obtain aligned basic data; and use the aligned basic data to train an initial federated learning model until the initial federated learning model converges to determine a final federated learning model;

[0018] A third processing unit, configured to use the final federated learning model to predict real-time data to determine whether the current fund flow is abnormal, and generate a warning message to prompt that the current fund flow is abnormal when it is determined that the current fund flow is abnormal.

[0019] According to another aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute any one of the above methods.

[0020] According to another aspect of the present application, a fund supervision system is provided. The system includes: a central server and multiple clients, and each of the clients is used to execute any one of the above methods.

[0021] Applying the technical solution of the present application, the local business data is screened, cleaned, integrated, and transformed to obtain basic data. The basic data includes historical contract information, historical budget information, and historical fund allocation information; the basic data is subjected to feature alignment processing using an asymmetric encryption protocol to obtain aligned basic data; and the aligned basic data is used to train an initial federated learning model until the initial federated learning model converges to determine the final federated learning model. Compared with the existing solution, the artificial intelligence model can be trained in a more secure and effective way for multi-source data, thereby improving the data security and the accuracy of subsequent model prediction, and further solving the problem of how to ensure the data privacy and security of the fund supervision system while using multi-data source data for artificial intelligence model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0023] Figure 1 A schematic flowchart of a fund supervision method provided according to an embodiment of the present application is shown;

[0024] Figure 2 A schematic diagram of a horizontal federated learning framework for fund supervision based on the C / S architecture provided according to an embodiment of the present application is shown;

[0025] Figure 3 A schematic flowchart of another fund supervision method provided according to an embodiment of the present application is shown;

[0026] Figure 4 A structural block diagram of a client provided according to an embodiment of the present application is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] As introduced in the background technology, the supervision method of most existing fund supervision systems is: the fund user stores the fund usage data in the supervision system, and the supervisor manually checks and inspects the fund user's fund operation data. This method cannot perform correlation analysis through the system algorithm, and the system cannot automatically calculate and determine whether the business data involves abnormal information, abnormal fund flow and other scenarios, and cannot automatically trigger system warnings. If the abnormality of the whereabouts of funds is missed manually, the actual cost of recovery is very high, and multiple data sources cannot guarantee the accuracy of the model. In order to solve the problem of how to use multiple data sources to train artificial intelligence models while ensuring the data privacy and security of the fund supervision system, the embodiments of the present application provide a fund supervision method, client, medium and system.

[0031] Currently, the supervision functions of most existing fund supervision systems are designed such that the supervisor manually checks and examines the operation data of the fund user. This method cannot perform systematic automatic analysis on the operation data, resulting in high labor costs and low supervision efficiency. Moreover, if the abnormal fund whereabouts are missed during manual inspection, the actual cost of recovery will be very high. Currently, the information sources of most fund supervision systems are only the input data of the fund user and local historical data, and the data is relatively isolated, making it impossible to conduct correlation analysis and early warning on relevant data. If artificial intelligence algorithms are used to train the fund supervision data model and machine learning algorithm models are introduced, enabling the system to automatically give early warnings for operation data, however, the local business data volume is small and the dimensions of effective feature data are incomplete, resulting in poor performance of the model trained with local data. If the fund supervision systems of different institutions provide business data for joint modeling, relevant laws stipulate that data sharing between different institutions should be strictly restricted, thus forming the problem of data islands. How to ensure the data privacy and security of the fund supervision system while using data from multiple data sources for artificial intelligence model training has become an urgent problem to be solved.

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention.

[0033] In this embodiment, a fund supervision method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0034] This method is applied to each client in the fund supervision system. Figure 1 It is a flowchart showing the process of a fund supervision method provided according to an embodiment of the present application. As Figure 1 shown, this method includes the following steps:

[0035] Step S101: Preprocess the local business data to obtain basic data. The above preprocessing includes screening, cleaning, integration, and transformation. The above integration means organizing and merging the local business data according to the association relationship. The above basic data includes historical contract information, historical budget information, and historical fund appropriation information;

[0036] The operations in each system are roughly divided into three categories: contract information, budget information, and fund appropriation information. The above historical contract information includes the name of the first party, the account number of the first party, the name of the second party, the contract amount, and the contract type. The above historical budget information includes the name of the other party, the budget expenditure amount, and the budget item. The above historical fund appropriation information includes the name of the payer, the account number of the payer, the name of the payee, the account number of the payee, the purpose of the transfer, and the transfer amount. There will be historical warning information in the fund supervision systems of each bank, and these warning information are associated with various types of business information.

[0037] Among them, step S101: Preprocess the local business data to obtain basic data, specifically including:

[0038] Filter the above local business data to obtain filtered data, so as to filter out the feature data for abnormal warning of fund flow in federated learning;

[0039] Clean the above filtered data to remove the incomplete data in the above filtered data and obtain cleaned data;

[0040] Specifically, check whether there are missing values in each column of the data. If there are missing values, they need to be processed; for the rows or columns with missing values, you can choose to delete this data or perform filling processing. Common filling methods include filling with statistical quantities such as mean, median, and mode, or filling according to specific business logic; ensure that the format and unit of the data are consistent, such as converting all date formats to a unified date format and converting all amount units to a unified currency unit, etc. Check whether there are duplicate rows in the data. If there are duplicate data, they need to be deleted to ensure the uniqueness of the data; perform outlier detection on the data. You can use methods such as box plots and scatter plots to find outliers and process them according to specific situations, such as deleting, replacing, or retaining; convert the data into the correct data type, such as converting the string type to the numerical type, date type, etc.; perform formatting processing on the data, such as removing extra spaces and unifying the case, etc., thus obtaining the cleaned data.

[0041] Guided by the feature dimension of the data, sort and merge the data from different operations and different data table structures in the above cleaned data according to the association relationship to obtain integrated processed data;

[0042] Perform conversion processing on the above integrated processed data to obtain the above basic data, so that the above integrated processed data meets the data specifications of the above initial federated learning model.

[0043] Specifically, the data screened during the screening process are the feature data for anomaly warning of fund flow in federated learning, that is, the data types used in federated learning. By orienting towards the feature dimensions of the data, the data from different businesses and different data table structures in the above-mentioned cleaned data are sorted and merged according to the association relationship to obtain integrated processed data. That is, when using multi-source data, the multi-source data are associated to improve the prediction accuracy of the federated learning model for subsequent training.

[0044] Step S102: Use an asymmetric encryption protocol to perform feature alignment processing on the above basic data to obtain aligned basic data; and use the above aligned basic data to train the initial federated learning model until the initial federated learning model converges to determine the final federated learning model.

[0045] Build a horizontal federated learning framework for fund supervision based on the C / S architecture as follows Figure 2 As shown, this framework includes a central aggregation server and n client nodes participating in federated learning. Each client node corresponds to the fund supervision system of a different bank, and each node stores its own business and warning data locally. Under this horizontal federated learning framework for fund supervision, the central server distributes the model parameters to each bank client node. Each bank client conducts local model training based on the local data. After the local training is completed, the data are processed based on the homomorphic encryption technology and then uploaded to the central server. The central server aggregates and evaluates the parameters uploaded by multiple bank clients, and returns the aggregated processed parameters to each bank client for local model update. The central server and each node continuously repeat the above steps until the training model converges to obtain a fund supervision joint model. The converged fund supervision joint model is synchronized to each local fund supervision system node. When new business data are added locally to each system, corresponding local anomaly data warning information can be obtained through local model calculation.

[0046] The C / S architecture is the abbreviation of the Client / Server architecture, which is a computer network architecture model and is usually used to describe the interaction relationship between the client and the server in a software system. In the C / S architecture, the client is responsible for processing the user interface and user input, while the server is responsible for processing data storage, processing, and transmission. The client and the server communicate through a network connection.

[0047] Homomorphic encryption technology is a type of encryption technology that allows computations to be performed on encrypted data without decrypting the data. This means that computations can be carried out without exposing the data, thereby protecting the privacy of the data. There are two main types of homomorphic encryption technology: partial homomorphic encryption and full homomorphic encryption. Partial homomorphic encryption can only support certain computational operations, such as addition or multiplication, while full homomorphic encryption can support any computational operation. Homomorphic encryption technology has extensive applications in fields such as cloud computing, privacy protection, and secure computing. By using homomorphic encryption technology, users can upload data to the cloud for computation, and the cloud cannot access the users' original data, thereby enhancing the privacy and security of the data.

[0048] Among them, the feature alignment process of the above basic data using the asymmetric encryption protocol in step S102 to obtain the aligned basic data specifically includes:

[0049] Based on the above asymmetric encryption protocol, combining the feature values of the above local business data and the generation time of the above local business data, perform feature alignment processing on the above basic data to obtain the above aligned basic data.

[0050] The specific implementation method of obtaining the above aligned basic data:

[0051] Extract the first feature value and time feature to be aligned from the above local business data, and use the above time feature as the second feature value;

[0052] Use the asymmetric encryption algorithm to encrypt the above first feature value and the above second feature value to obtain the first encrypted data and the second encrypted data;

[0053] Align and match the first encrypted data and the second encrypted data with the corresponding features in other data sources or datasets respectively, establish a corresponding relationship to obtain the matched data;

[0054] Integrate the above matched data to obtain the above aligned basic data.

[0055] Specifically, the asymmetric encryption protocol needs to use public keys and private keys to encrypt and decrypt data. By performing feature alignment processing on the basic data, the encryption and decryption processes can be made simpler and more efficient, enhancing the security of the data; by performing feature alignment processing on the basic data, data redundancy and repetition can also be reduced, improving the reliability and accuracy of the data, ensuring the integrity and consistency of the data; performing feature alignment processing on the basic data can make the data more standardized and easier to operate, improving the operability and maintainability of the data, reducing the complexity and difficulty of data processing; by performing feature alignment processing on the basic data, the data processing efficiency and response speed can also be improved, reducing the time and cost of data processing, and enhancing the efficiency and effect of data processing.

[0056] During feature alignment, the system does not expose data belonging to the same participating party, and the entire alignment work is automatically completed by the system.

[0057] In an embodiment of the present application, the above-mentioned fund supervision system further includes a central server, and the initial federated learning model is trained using the above-mentioned aligned basic data until the final federated learning model is determined when the above-mentioned initial federated learning model converges, including a first receiving step, a first processing step, a second processing step, and a second receiving step, where,

[0058] The first receiving step receives the initial encryption parameters sent by the above-mentioned central server, and the above-mentioned initial encryption parameters represent the encrypted data of the initial parameters of the above-mentioned initial federated learning model;

[0059] The first processing step uses the above-mentioned aligned basic data and the above-mentioned initial encryption parameters to perform local model training on the above-mentioned initial federated learning model, calculates the training gradient and loss value of the above-mentioned initial federated learning model to update the above-mentioned initial federated learning model;

[0060] The second processing step uploads the parameters of the trained above-mentioned initial federated learning model to the above-mentioned central server using encryption or differential privacy technology, so that the above-mentioned central server performs weighted averaging processing on the parameters of the trained above-mentioned initial federated learning model of each above-mentioned client to obtain corrected parameters, and distributes the above-mentioned corrected parameters to each above-mentioned client;

[0061] The second receiving step receives the above-mentioned corrected parameters sent by the above-mentioned central server, updates the parameters of the trained above-mentioned initial federated learning model using the above-mentioned corrected parameters, and repeats the above-mentioned first receiving step, the above-mentioned first processing step, and the above-mentioned second processing step until the above-mentioned final federated learning model is determined when the above-mentioned initial federated learning model converges.

[0062] Among them, during the model training process, the interaction between the client and the central server is realized through encryption or differential privacy technology. The central server will perform weighted averaging processing on the parameters of the models of multiple clients received each time, and then encrypt and feedback the processed parameters to each client, and then the client trains the model again. In this way, the joint data model training between different systems is realized, enabling the data to be analyzed associatively, and effectively improving the early warning recognition accuracy of abnormal fund supervision behaviors.

[0063] In an embodiment of the present application, after determining the final federated learning model, the above-mentioned method further includes:

[0064] At each preset period, based on the newly added associated service feature data and warning situations within the above preset period, execute the above first receiving step, the above first processing step, the above second processing step, and the above second receiving step to update the above final federated learning model.

[0065] Specifically, the preset period can be one week. According to the time period for model update set in the system, the client nodes for bank fund supervision obtain the newly added warning and associated service feature data within the periodic time interval, and repeat the above first receiving step, the above first processing step, the above second processing step, and the above second receiving step to obtain and mutually transmit the latest gradient and loss data, and obtain the updated federated learning model, thereby ensuring that the model of the present application is updated periodically and improving the accuracy of model prediction.

[0066] In step S103, use the above final federated learning model to predict the real-time data, determine whether the current fund flow is abnormal, and in the case of determining that the current fund flow is abnormal, generate a warning message to prompt that the current fund flow is abnormal.

[0067] In the above steps, the local service data is screened, cleaned, integrated, and transformed to obtain the basic data. The basic data includes historical contract information, historical budget information, and historical fund appropriation information; the asymmetric encryption protocol is used to perform feature alignment processing on the above basic data to obtain the aligned basic data; and the above aligned basic data is used to train the initial federated learning model until the initial federated learning model converges to determine the final federated learning model. Compared with the existing solutions, it can train the artificial intelligence model in a more secure and effective way for multi-source data, thereby improving the data security and the accuracy of subsequent model prediction, and further solving the problem of how to ensure the data privacy and security of the fund supervision system while using multi-data source data for artificial intelligence model training.

[0068] The asymmetric encryption protocol is an encryption communication protocol that uses a pair of keys: a public key and a private key. The public key is used to encrypt data, and the private key is used to decrypt data. This encryption method ensures the security of data because only the party holding the private key can decrypt the data.

[0069] Common asymmetric encryption protocols include RSA (Rivest-Shamir-Adleman), Diffie-Hellman, ECC (Elliptic Curve Cryptography), etc. These protocols play an important role in protecting data security, identity authentication, and digital signatures. Asymmetric encryption protocols are usually used in scenarios such as secure communication, encrypted storage, and digital signatures.

[0070] In an embodiment of the present application, the above-mentioned final federated learning model is used to predict real-time data to determine whether the current fund flow is abnormal, including:

[0071] Use the above-mentioned final federated learning model to predict real-time data to obtain an anomaly confidence level;

[0072] In the case where the above-mentioned anomaly confidence level is greater than or equal to the confidence level threshold, it is determined that the above-mentioned current fund flow is abnormal;

[0073] In the case where the above-mentioned anomaly confidence level is less than the above-mentioned confidence level threshold, it is determined that the above-mentioned current fund flow is normal.

[0074] Specifically, the value range of the confidence level threshold is [70%, 90%]. A horizontal federated learning system framework for fund supervision based on the C / S architecture is constructed, and a fund supervision federated learning model is introduced for each fund supervision system node, enabling the system to automatically identify operation risk factors, reducing manual judgment and intervention of abnormal fund behaviors, and improving supervision efficiency. A fund supervision system based on horizontal federated learning is designed. Through the cyclic process of "parameter conduction - gradient aggregation - model update" between the central server and each fund supervision system node, a joint fund supervision model is obtained. Thus, the data island problem is effectively solved. While ensuring the data privacy of different systems, joint data model training between different systems is achieved, enabling the system data to be analyzed associatively and improving the accuracy of early warning and identification of abnormal fund supervision behaviors.

[0075] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the fund supervision method of the present application will be described in detail below with specific embodiments.

[0076] This embodiment relates to a specific fund supervision method, as Figure 3 shown, including:

[0077] Preprocess the local business data to obtain basic data. The preprocessing includes screening, cleaning, integration, and transformation. The integration process represents organizing and merging the local business data according to the association relationship. The basic data includes historical contract information, historical budget information, and historical fund allocation information;

[0078] Specifically, the local business data is screened to obtain screened data, so as to screen out the feature data for abnormal early warning of fund flow in federated learning; the screened data is cleaned to remove the incomplete data in the screened data, obtaining cleaned data; guided by the feature dimension of the data, the data from different services and different data table structures in the cleaned data is sorted and merged according to the association relationship, obtaining integrated processed data; the integrated processed data is transformed to obtain basic data, so that the integrated processed data meets the data specifications of the initial federated learning model.

[0079] Based on the asymmetric encryption protocol, combined with the feature value of the local business data and the generation time of the local business data, the basic data is subjected to feature alignment processing to obtain aligned basic data;

[0080] Specifically, the first feature value and time feature to be aligned are extracted from the local business data, and the time feature is used as the second feature value; the first feature value and the second feature value are encrypted using the asymmetric encryption algorithm to obtain the first encrypted data and the second encrypted data; the first encrypted data and the second encrypted data are respectively aligned and matched with the corresponding features in other data sources or data sets to establish a corresponding relationship to obtain the matched data; the matched data is integrated to obtain aligned basic data.

[0081] The first receiving step: The client receives the initial encryption parameters sent by the central server, and the initial encryption parameters represent the encrypted data of the initial parameters of the initial federated learning model;

[0082] The first processing step: The client uses the aligned basic data and the initial encryption parameters to perform local model training on the initial federated learning model, calculates the training gradient and loss value of the initial federated learning model to update the initial federated learning model;

[0083] The second processing step: The client uploads the parameters of the trained initial federated learning model to the central server using encryption or differential privacy technology, so that the central server performs weighted averaging processing on the parameters of the trained initial federated learning model of each client to obtain corrected parameters, and distributes the corrected parameters to each client;

[0084] The second receiving step: The client updates the parameters of the trained initial federated learning model using the corrected parameters, and repeats the first receiving step, the first processing step and the second processing step until the initial federated learning model converges to determine the final federated learning model;

[0085] Every preset period, based on the newly added associated business feature data and early warning situation within the preset period, the first receiving step, the first processing step, the second processing step and the second receiving step are executed to update the final federated learning model;

[0086] Use the final federated learning model to predict real-time data, determine whether the current fund flow is abnormal, and generate a warning message when it is determined that the current fund flow is abnormal to prompt the abnormality of the current fund flow.

[0087] Specifically, use the final federated learning model to predict real-time data to obtain an anomaly confidence level; when the anomaly confidence level is greater than or equal to the confidence level threshold, determine that the current fund flow is abnormal; when the anomaly confidence level is less than the confidence level threshold, determine that the current fund flow is normal.

[0088] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0089] The embodiment of the present application also provides a client. It should be noted that the client of the embodiment of the present application can be used to execute the fund supervision method provided by the embodiment of the present application. The client is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can implement a predetermined function. Although the client described in the following embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0090] The following introduces the client provided by the embodiment of the present application.

[0091] Figure 4 is a structural block diagram of a client provided according to an embodiment of the present application. As Figure 4 shown, the client includes:

[0092] A first processing unit 41, configured to preprocess local service data to obtain basic data. The above preprocessing includes screening processing, cleaning processing, integration processing, and transformation processing. The above integration processing represents sorting and merging the local service data according to the association relationship. The above basic data includes historical contract information, historical budget information, and historical fund appropriation information;

[0093] A second processing unit 42, configured to perform feature alignment processing on the above basic data using an asymmetric encryption protocol to obtain aligned basic data; and use the above aligned basic data to train an initial federated learning model until the final federated learning model is determined when the initial federated learning model converges;

[0094] The third processing unit 43 is configured to use the above final federated learning model to predict real-time data, determine whether the current fund flow is abnormal, and generate a warning message when it is determined that the current fund flow is abnormal, so as to indicate that the current fund flow is abnormal.

[0095] The above client performs screening, cleaning, integration, and transformation processing on local business data to obtain basic data, where the basic data includes historical contract information, historical budget information, and historical fund appropriation information; performs feature alignment processing on the above basic data using an asymmetric encryption protocol to obtain aligned basic data; and uses the above aligned basic data to train an initial federated learning model until the above initial federated learning model converges to determine a final federated learning model. Compared with the existing solution, it can train an artificial intelligence model in a more secure and effective way for multi-source data, thereby improving the data security and the accuracy of subsequent model prediction, and further solving the problem of how to ensure the data privacy and security of the fund supervision system while using multi-data source data for artificial intelligence model training.

[0096] In an embodiment of the present application, the fund supervision system further includes a central server, and the second processing unit includes a first processing module, a second processing module, a third processing module, and a fourth processing module; the first processing module is used for the first receiving step: receiving the initial encryption parameters sent by the above central server, where the initial encryption parameters represent the encrypted data of the initial parameters of the above initial federated learning model; the second processing module is used for the first processing step: using the above aligned basic data and the above initial encryption parameters to perform local model training on the above initial federated learning model, calculating the training gradient and loss value of the above initial federated learning model to update the above initial federated learning model; the third processing module is used for the second processing step: using encryption or differential privacy technology to upload the parameters of the trained above initial federated learning model to the above central server, so that the above central server performs weighted averaging processing on the parameters of the trained above initial federated learning model of each above client to obtain corrected parameters, and sending the above corrected parameters to each above client; the fourth processing module is used for the second receiving step: receiving the above corrected parameters sent by the above central server, using the above corrected parameters to update the parameters of the trained above initial federated learning model, and repeating the above first receiving step, the above first processing step, and the above second processing step until the above initial federated learning model converges to determine the above final federated learning model.

[0097] In an embodiment of the present application, the client further includes a fourth processing unit, which is configured to, after determining the final federated learning model, perform the above-mentioned first receiving step, the above-mentioned first processing step, the above-mentioned second processing step, and the above-mentioned second receiving step every preset period based on the newly added associated business feature data and the warning situation within the above-mentioned preset period to update the above-mentioned final federated learning model.

[0098] In an embodiment of the present application, the first processing unit includes a fifth processing module, a sixth processing module, a seventh processing module, and an eighth processing module; screening the above-mentioned local business data to obtain screened data so as to screen out the feature data for the federated learning of abnormal fund flow warning; cleaning the above-mentioned screened data to remove the incomplete data in the above-mentioned screened data to obtain cleaned data; guiding by the feature dimension of the data, sorting and merging the data from different businesses and different data table structures in the above-mentioned cleaned data according to the association relationship to obtain integrated processed data; performing conversion processing on the above-mentioned integrated processed data to obtain the above-mentioned basic data so that the above-mentioned integrated processed data meets the data specification of the above-mentioned initial federated learning model.

[0099] In an embodiment of the present application, the third processing unit includes a ninth processing module, a first determination module, and a second determination module. The ninth processing module is configured to use the above-mentioned final federated learning model to predict real-time data to obtain an abnormal confidence level; the first determination module is configured to determine that the current fund flow is abnormal when the above-mentioned abnormal confidence level is greater than or equal to the confidence level threshold; the second determination module is configured to determine that the current fund flow is normal when the above-mentioned abnormal confidence level is less than the above-mentioned confidence level threshold.

[0100] In an embodiment of the present application, the second processing unit includes a tenth processing module, which is configured to perform feature alignment processing on the above-mentioned basic data based on the above-mentioned asymmetric encryption protocol in combination with the feature value of the above-mentioned local business data and the generation time of the above-mentioned local business data to obtain the above-mentioned aligned basic data.

[0101] In an embodiment of the present application, the tenth processing module includes a first processing sub-module, a second processing sub-module, a third processing sub-module, and a fourth processing sub-module. The first processing sub-module is configured to extract a first eigenvalue and a time feature to be aligned from the above-mentioned local service data, and use the above-mentioned time feature as a second eigenvalue; the second processing sub-module is configured to perform encryption processing on the above-mentioned first eigenvalue and the above-mentioned second eigenvalue by using an asymmetric encryption algorithm to obtain first encrypted data and second encrypted data; the third processing sub-module is configured to align and match the first encrypted data and the second encrypted data with corresponding features in other data sources or data sets respectively to establish a corresponding relationship to obtain matched data; the fourth processing sub-module is configured to integrate the above-mentioned matched data to obtain the above-mentioned alignment basic data.

[0102] The above-mentioned client includes a processor and a memory. The above-mentioned first processing unit, second processing unit, third processing unit, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to implement corresponding functions. The above-mentioned modules are all located in the same processor; or, the above-mentioned each module is located in different processors in any combination form.

[0103] The processor contains a kernel, and the kernel retrieves the corresponding program unit from the memory. One or more kernels can be set, and by adjusting the kernel parameters, the problem of how to ensure the data privacy and security of the fund supervision system while using data from multiple data sources for artificial intelligence model training can be solved.

[0104] The memory may include non-permanent memory in a computer-readable medium, forms such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0105] An embodiment of the present invention provides a computer-readable storage medium. The above-mentioned computer-readable storage medium includes a stored program, wherein when the above-mentioned program runs, it controls the device where the above-mentioned computer-readable storage medium is located to execute the above-mentioned fund supervision method.

[0106] An embodiment of the present invention provides a processor. The above-mentioned processor is used to run a program, wherein when the above-mentioned program runs, it executes the above-mentioned fund supervision method.

[0107] An embodiment of the present invention provides a device, which includes a processor, a memory, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements at least the following steps: preprocess local service data to obtain basic data. The above preprocessing includes screening processing, cleaning processing, integration processing, and transformation processing. The above integration processing represents organizing and merging the above local service data according to the association relationship. The above basic data includes historical contract information, historical budget information, and historical fund appropriation information; perform feature alignment processing on the above basic data using an asymmetric encryption protocol to obtain aligned basic data; and use the above aligned basic data to train an initial federated learning model until the above initial federated learning model converges to determine a final federated learning model; use the above final federated learning model to predict real-time data to determine whether the current fund flow is abnormal, and generate a warning message when it is determined that the above current fund flow is abnormal to indicate the abnormality of the above current fund flow. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0108] The present application also provides a computer program product, which, when executed on a data processing device, is adapted to execute a program initialized with at least the following method steps: preprocess local service data to obtain basic data. The above preprocessing includes screening processing, cleaning processing, integration processing, and transformation processing. The above integration processing represents organizing and merging the above local service data according to the association relationship. The above basic data includes historical contract information, historical budget information, and historical fund appropriation information; perform feature alignment processing on the above basic data using an asymmetric encryption protocol to obtain aligned basic data; and use the above aligned basic data to train an initial federated learning model until the above initial federated learning model converges to determine a final federated learning model; use the above final federated learning model to predict real-time data to determine whether the current fund flow is abnormal, and generate a warning message when it is determined that the above current fund flow is abnormal to indicate the abnormality of the above current fund flow.

[0109] The present application also provides a fund supervision system, which includes: a central server and multiple clients, and each of the above clients is used to execute any one of the above methods. Screen, clean, integrate, and transform the local business data to obtain basic data, where the above basic data includes historical contract information, historical budget information, and historical fund appropriation information; perform feature alignment processing on the above basic data using an asymmetric encryption protocol to obtain aligned basic data; and use the above aligned basic data to train an initial federated learning model until the final federated learning model is determined when the above initial federated learning model converges. Compared with the existing solutions, it can train the artificial intelligence model in a more secure and effective way for multi-source data, thereby improving the data security and the accuracy of subsequent model prediction, and further solving the problem of how to ensure the data privacy and security of the fund supervision system while using multi-source data to train the artificial intelligence model.

[0110] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. They can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order from here, or they can be separately made into individual integrated circuit modules, or multiple modules or steps among them can be made into a single integrated circuit module to implement. In this way, the present invention is not limited to any specific combination of hardware and software.

[0111] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0112] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1One or more processes and / or blocks Figure 1 means for the functions specified in one or more blocks.

[0113] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device that implements the processes Figure 1 One or more processes and / or blocks Figure 1 the functions specified in one or more blocks.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide for implementing the processes Figure 1 One or more processes and / or blocks Figure 1 the steps of the functions specified in one or more blocks.

[0115] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0116] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0117] Computer-readable media includes permanent and non-permanent, removable and non-removable media and can be implemented by any method or technology for information storage. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0118] It should also be noted that the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0119] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0120] 1), The fund supervision method of the present application screens, cleans, integrates and transforms local business data to obtain basic data, and the above basic data includes historical contract information, historical budget information and historical fund appropriation information; uses an asymmetric encryption protocol to perform feature alignment processing on the above basic data to obtain aligned basic data; and uses the above aligned basic data to train an initial federated learning model until the above initial federated learning model converges to determine the final federated learning model. Compared with the existing solutions, it can train an artificial intelligence model in a more secure and effective way for multi-source data, thereby improving the data security and the accuracy of subsequent model prediction, and further solving the problem of how to ensure the data privacy and security of the fund supervision system while using multi-source data to train an artificial intelligence model.

[0121] 2), The fund supervision system of the present application screens, cleans, integrates and transforms local business data to obtain basic data, and the above basic data includes historical contract information, historical budget information and historical fund appropriation information; uses an asymmetric encryption protocol to perform feature alignment processing on the above basic data to obtain aligned basic data; and uses the above aligned basic data to train an initial federated learning model until the above initial federated learning model converges to determine the final federated learning model. Compared with the existing solutions, it can train an artificial intelligence model in a more secure and effective way for multi-source data, thereby improving the data security and the accuracy of subsequent model prediction, and further solving the problem of how to ensure the data privacy and security of the fund supervision system while using multi-source data to train an artificial intelligence model.

[0122] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fund supervision method, characterized in that, This method is applied to each client in the fund supervision system, including: Preprocessing local business data to obtain basic data. The preprocessing includes screening, cleaning, integration, and transformation. The integration represents organizing and merging the local business data according to the association relationship. The basic data includes historical contract information, historical budget information, and historical fund appropriation information; Performing feature alignment processing on the basic data using an asymmetric encryption protocol to obtain aligned basic data; and training an initial federated learning model using the aligned basic data until the initial federated learning model converges to determine the final federated learning model; Using the final federated learning model to predict real-time data to determine whether the current fund flow is abnormal, and generating a warning message when it is determined that the current fund flow is abnormal to indicate the abnormality of the current fund flow.

2. The method according to claim 1, wherein The fund supervision system further includes a central server. Training the initial federated learning model using the aligned basic data until the initial federated learning model converges to determine the final federated learning model includes: The first receiving step: receiving the initial encryption parameters sent by the central server, where the initial encryption parameters represent the encrypted data of the initial parameters of the initial federated learning model; The first processing step: performing local model training on the initial federated learning model using the aligned basic data and the initial encryption parameters, calculating the training gradient and loss value of the initial federated learning model to update the initial federated learning model; The second processing step: uploading the parameters of the trained initial federated learning model to the central server using encryption or differential privacy technology, so that the central server performs weighted averaging on the parameters of the trained initial federated learning model of each client to obtain corrected parameters, and sending the corrected parameters to each client; The second receiving step: receiving the corrected parameters sent by the central server, updating the parameters of the trained initial federated learning model using the corrected parameters, and repeating the first receiving step, the first processing step, and the second processing step until the initial federated learning model converges to determine the final federated learning model.

3. The method according to claim 2, wherein After determining the final federated learning model, the method further includes: Every preset period, based on the newly added associated business feature data and warning situations within the preset period, performing the first receiving step, the first processing step, the second processing step, and the second receiving step to update the final federated learning model.

4. The method according to claim 1, wherein Preprocessing local business data to obtain basic data, including: Performing screening processing on the local business data to obtain screened data to screen out the feature data for federated learning of abnormal fund flow warning; Performing cleaning processing on the screened data to remove the incomplete data in the screened data to obtain cleaned data; Orienting to the feature dimensions of the data, sorting and merging the data from different services and different data table structures in the cleaned data according to the association relationship to obtain integrated processed data; Performing transformation processing on the integrated processed data to obtain the basic data, so that the integrated processed data meets the data specifications of the initial federated learning model.

5. The method according to claim 1, wherein Using the final federated learning model to predict the real-time data to determine whether the current fund flow is abnormal, including: Using the final federated learning model to predict the real-time data to obtain an abnormal confidence level; When the abnormal confidence level is greater than or equal to the confidence level threshold, determining that the current fund flow is abnormal; When the abnormal confidence level is less than the confidence level threshold, determining that the current fund flow is normal.

6. The method according to claim 1, characterized in that, Performing feature alignment processing on the basic data by using an asymmetric encryption protocol to obtain aligned basic data, including: Based on the asymmetric encryption protocol, combining the feature value of the local service data and the generation time of the local service data to perform feature alignment processing on the basic data to obtain the aligned basic data.

7. The method according to claim 6, wherein Based on the asymmetric encryption protocol, combining the feature value of the local service data and the generation time of the local service data to perform feature alignment processing on the basic data to obtain the aligned basic data, including: Extracting a first feature value and a time feature to be aligned from the local service data, and using the time feature as a second feature value; Using an asymmetric encryption algorithm to encrypt the first feature value and the second feature value to obtain first encrypted data and second encrypted data; Aligning and matching the first encrypted data and the second encrypted data with the corresponding features in other data sources or data sets respectively to establish a corresponding relationship to obtain matched data; Integrating the matched data to obtain the aligned basic data.

8. A client, characterized in that, Including: A first processing unit, configured to preprocess the local service data to obtain basic data, where the preprocessing includes screening processing, cleaning processing, integration processing, and transformation processing, the integration processing represents sorting and merging the local service data according to the association relationship, and the basic data includes historical contract information, historical budget information, and historical fund appropriation information; A second processing unit, configured to perform feature alignment processing on the basic data by using an asymmetric encryption protocol to obtain aligned basic data; and training the initial federated learning model by using the aligned basic data until the initial federated learning model converges to determine the final federated learning model; A third processing unit, configured to use the final federated learning model to predict the real-time data to determine whether the current fund flow is abnormal, and generate a warning message to prompt that the current fund flow is abnormal when it is determined that the current fund flow is abnormal.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, where when the program runs, it controls the device where the computer-readable storage medium is located to execute the method according to any one of claims 1 to 7.

10. A fund supervision system, characterized in that, Including: A central server and multiple clients, each of the clients being configured to execute the method according to any one of claims 1 to 7.

Images