A privacy enhancement method and device for power federated learning against data heterogeneity

Abstract

The present invention relates to the field of information security technology, and specifically provides a method and device for enhancing the privacy of power federated learning that is resistant to data heterogeneity. The present invention allows participants to select different privacy protection technologies based on their own resource conditions, which is suitable for scenarios where resources are unbalanced among the parties; and effectively integrates different types of privacy-processed data to achieve model training in a federated learning scenario, thereby improving the efficiency of model training while ensuring data privacy and security, and obtaining a usable global training model.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular to a method and device for enhancing privacy of power federated learning that is resistant to data heterogeneity. Background Art

[0002] With the rapid development of digital technology, privacy and security issues in the process of global data circulation and sharing are becoming increasingly prominent. In the power industry, the use of data is becoming increasingly extensive and in-depth, covering many fields such as energy dispatching, demand-side management, and smart grid construction. However, given that power data involves a large amount of sensitive information, such as user electricity consumption records and equipment operating status, its privacy protection requirements are also more stringent. In order to ensure privacy security while ensuring data utilization, privacy computing technology has emerged and has become an important solution. It adopts a distributed computing architecture and integrates multiple technical means such as multi-party secure computing, federated learning, homomorphic encryption, and differential privacy to achieve data analysis and calculation without exposing the original data, thereby ensuring privacy security. The introduction of the privacy computing framework into the power industry is of great significance. It not only ensures the privacy security of power data, but also promotes the circulation, sharing, and efficient use of data. With the help of this framework, all parties can conduct data analysis and cooperation under the premise of strictly protecting sensitive information, and promote the intelligent upgrade of energy management and dispatching.

[0003] At present, the technologies widely used in the academic community for the study of privacy issues of power data are mainly encryption, anonymity and random perturbation. Among them, encryption technology can maintain accurate statistical characteristics, but the computational overhead is large; anonymity technology solves the problem of computational overhead, but under the condition that the attacker has certain background knowledge, anonymity technology will have the risk of privacy information leakage; differential privacy, the most widely used random perturbation method at present, strictly defines the level of privacy protection, and even if the attacker has a lot of background knowledge, he cannot obtain information about the real data.

[0004] However, in practical applications, the differences in resources among the parties involved in power data cooperation have brought new challenges to privacy computing. These differences are mainly reflected in data quality, network environment, computing power and other aspects. Therefore, a single privacy computing method is often difficult to meet diverse practical needs. To meet this challenge, it is necessary to continue to conduct in-depth research and exploration of more advanced and flexible privacy computing technologies to adapt to changes in different scenarios and needs. Continuous innovation and technological upgrades are expected to provide stronger privacy and security protection for the digital transformation of the power industry. Summary of the invention

[0005] In order to overcome the above-mentioned defects, the present invention proposes a privacy enhancement method and device for power federated learning that is resistant to data heterogeneity.

[0006] In a first aspect, a method for enhancing privacy of power federated learning against data heterogeneity is provided, wherein the method for enhancing privacy of power federated learning against data heterogeneity comprises:

[0007] The deadline T for each round of data submission is published to each client, so that the client selects the initial privacy protection technology to be used for executing the privacy protection task based on the deadline T, and informs the server of the initial privacy protection technology;

[0008] Determine the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client, so that after the client performs the training task on the local training model, it uses its corresponding final privacy protection technology to perform privacy protection processing on the model parameters of the local training model, and uploads the model parameters after the privacy protection processing to the server;

[0009] After the deadline T is reached, the heterogeneous data uploaded by each client is grouped and aggregated according to the final privacy protection technology used by each client to perform the privacy protection task, and the aggregation result is used to update the global model.

[0010] Preferably, the privacy protection technology includes: differential privacy technology and secure multi-party computing technology.

[0011] Preferably, the determining, based on the initial privacy protection technology corresponding to each client, the final privacy protection technology adopted by each client to perform the privacy protection task includes:

[0012] Step a. Determine whether the number of clients whose initial privacy protection technologies are secure multi-party computing technologies is 1. If so, select a client from other clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is closest to the deadline T as a matching client, and set the current privacy protection technology corresponding to the matching client to the secure multi-party computing technology, set the current privacy protection technology corresponding to the clients other than the matching client to the differential privacy technology, and then execute step b. Alternatively, set the current privacy protection technology corresponding to all clients to the differential privacy technology and then execute step b. Otherwise, execute step b.

[0013] Step b. Determine whether the total number of clients is less than a preset threshold. If so, set the current privacy protection technology corresponding to all clients to the secure multi-party computing technology and then execute step d. Otherwise, execute step c.

[0014] Step c. Determine whether the number of clients whose current privacy protection technology is differential privacy technology is less than a preset threshold. If so, select n clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is the largest among the clients whose current privacy protection technology is secure multi-party computing technology as matching clients, and set the current privacy protection technology corresponding to the matching clients to differential privacy technology before executing step d. Otherwise, execute step d.

[0015] Step d. Determine whether the number of clients whose current privacy protection technologies are secure multi-party computing technologies is 0 or greater than or equal to 2. If so, use the current privacy protection technology corresponding to the client as the final privacy protection technology corresponding to the client. Otherwise, use the final privacy protection technology corresponding to all clients as secure multi-party computing technology.

[0016] Here, n is the difference between the preset threshold and the number of differential privacy technologies corresponding to the client's current privacy protection technology.

[0017] Furthermore, the preset threshold is 10.

[0018] Preferably, the grouping and aggregating the heterogeneous data uploaded by each client according to the final privacy protection technology adopted by each client to perform the privacy protection task includes:

[0019] Aggregate the data uploaded by the client whose final privacy protection technology used to perform the privacy protection task is the secure multi-party computing technology into a first group of data, and aggregate the data uploaded by the client whose final privacy protection technology used to perform the privacy protection task is the differential privacy technology into a second group of data;

[0020] Performing a decryption operation on the first grouped data using a secure multi-party computing technology to obtain a mean value corresponding to the first grouped data, and performing a mean estimation on the second grouped data using a perturbed data correction scheme in a differential privacy technology to obtain a mean value corresponding to the second grouped data;

[0021] An aggregation result is determined based on the mean value corresponding to the first grouped data and the mean value corresponding to the second grouped data.

[0022] Furthermore, the aggregation result is: Aw1+Bw2, A is the mean value corresponding to the first grouped data, w1 is the first weight, B is the mean value corresponding to the second grouped data, and w2 is the second weight.

[0023] In a second aspect, a privacy enhancement method for power federated learning that resists data heterogeneity is provided, the method comprising:

[0024] Selecting an initial privacy protection technology for executing the privacy protection task based on the deadline T sent by the server, and informing the server of the initial privacy protection technology, so that the server determines a final privacy protection technology for each client to execute the privacy protection task based on the initial privacy protection technology corresponding to each client;

[0025] After executing the training task on the local training model, the model parameters of the local training model are privacy-protected using the corresponding final privacy protection technology, and the model parameters after privacy protection are uploaded to the server, so that after the deadline T is reached, the server groups and aggregates the heterogeneous data uploaded by each client according to the final privacy protection technology used by each client to execute the privacy protection task, and uses the aggregation result to update the global model.

[0026] Preferably, the selecting an initial privacy protection technology for executing the privacy protection task based on the deadline T sent by the server includes:

[0027] Obtain the sum t1 of the time required for the client to perform the training task and the time required to perform the privacy protection task using differential privacy technology, and the sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using secure multi-party computing technology;

[0028] Determine whether the client's t2 is less than T. If so, let the client use the multi-party secure computing technology as the initial privacy protection technology to perform the privacy protection task. Otherwise, let the client use the differential privacy technology as the initial privacy protection technology to perform the privacy protection task.

[0029] Preferably, when the client fails to complete the local training model execution training task and the privacy protection task within the deadline T, no data is uploaded to the server.

[0030] In a third aspect, a power federated learning privacy enhancement device resistant to data heterogeneity is provided, wherein the power federated learning privacy enhancement device resistant to data heterogeneity comprises:

[0031] The server is used to publish the deadline T for each round of data submission to each client;

[0032] A client, configured for the client to select an initial privacy protection technology to be used for executing the privacy protection task based on the deadline T, and to inform the server of the initial privacy protection technology;

[0033] The server is used to determine the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client;

[0034] The client is used to perform privacy protection processing on the model parameters of the local training model using its corresponding final privacy protection technology after executing the training task on the local training model, and upload the model parameters after privacy protection processing to the server;

[0035] The server side is used to group and aggregate the heterogeneous data uploaded by each client according to the final privacy protection technology adopted by each client to perform the privacy protection task after the deadline T is reached, and update the global model using the aggregation result.

[0036] In a fourth aspect, a computer device is provided, comprising: one or more processors;

[0037] The processor is configured to execute one or more programs;

[0038] When the one or more programs are executed by the one or more processors, the power federated learning privacy enhancement method that is resistant to data heterogeneity is implemented.

[0039] In a fifth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed, the privacy enhancement method for power federated learning that is resistant to data heterogeneity is implemented.

[0040] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0041] The present invention relates to the field of information security technology, and specifically provides a method and device for enhancing the privacy of power federated learning against data heterogeneity, including: a server side, for publishing the deadline T for each round of data submission to each client; a client side, for the client to select the initial privacy protection technology used to perform the privacy protection task based on the deadline T, and inform the server side of the initial privacy protection technology; the server side, for determining the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client; the client side, for performing privacy protection processing on the model parameters of the local training model using the corresponding final privacy protection technology after performing the training task on the local training model, and uploading the model parameters after the privacy protection processing to the server side; the server side, for grouping and aggregating the heterogeneous data uploaded by each client according to the final privacy protection technology used by each client to perform the privacy protection task after reaching the deadline T, and updating the global model using the aggregation result. The present invention realizes data aggregation in the federated learning scenario under the condition of multi-party resource imbalance, and efficiently updates the global model without leaking the local model parameters of each client. Based on the client's local resources and data volume, select appropriate privacy protection technology to complete model training, data processing and transmission in a limited time as much as possible, to ensure the efficiency and effectiveness of global model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of the main steps on the server side of the privacy enhancement method for power federated learning against data heterogeneity according to an embodiment of the present invention;

[0043] Figure 2 It is a flow chart of the main steps of the client of the privacy enhancement method for power federated learning that is resistant to data heterogeneity in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The specific implementation modes of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] As disclosed in the background technology, with the rapid development of digital technology, the privacy and security issues in the process of global data circulation and sharing are becoming increasingly prominent. In the power industry, the use of data is becoming increasingly extensive and in-depth, covering many fields such as energy dispatching, demand-side management, and smart grid construction. However, given that power data involves a large amount of sensitive information, such as user electricity consumption records, equipment operating status, etc., its requirements for privacy protection are also more stringent. In order to ensure privacy security while ensuring data utilization, privacy computing technology has emerged and has become an important solution. It adopts a distributed computing architecture and integrates multiple technical means such as multi-party secure computing, federated learning, homomorphic encryption, and differential privacy to achieve data analysis and calculation without exposing the original data, thereby ensuring privacy security. It is of great significance to introduce the privacy computing framework into the power industry, which not only ensures the privacy security of power data, but also promotes the circulation, sharing, and efficient use of data. With the help of this framework, all parties can carry out data analysis and cooperation under the premise of strictly protecting sensitive information, and promote the intelligent upgrade of energy management and dispatching.

[0047] At present, the technologies widely used in the academic community for the study of privacy issues of power data are mainly encryption, anonymity and random perturbation. Among them, encryption technology can maintain accurate statistical characteristics, but the computational overhead is large; anonymity technology solves the problem of computational overhead, but under the condition that the attacker has certain background knowledge, anonymity technology will have the risk of privacy information leakage; differential privacy, the most widely used random perturbation method at present, strictly defines the level of privacy protection, even if the attacker has a lot of background knowledge, he cannot obtain information about the real data.

[0048] However, in practical applications, the differences in resources among the parties involved in power data cooperation have brought new challenges to privacy computing. These differences are mainly reflected in data quality, network environment, computing power and other aspects. Therefore, a single privacy computing method is often difficult to meet diverse practical needs. To meet this challenge, it is necessary to continue to conduct in-depth research and exploration of more advanced and flexible privacy computing technologies to adapt to changes in different scenarios and needs. Continuous innovation and technological upgrades are expected to provide stronger privacy and security protection for the digital transformation of the power industry.

[0049] In order to improve the above problems, the present invention relates to the field of information security technology, and specifically provides a method and device for enhancing the privacy of power federated learning against data heterogeneity, including: a server side, for publishing the deadline T for each round of data submission to each client; a client side, for the client to select the initial privacy protection technology used to perform the privacy protection task based on the deadline T, and inform the server side of the initial privacy protection technology; a server side, for determining the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client; a client side, for performing privacy protection processing on the model parameters of the local training model using the corresponding final privacy protection technology after performing the training task on the local training model, and uploading the model parameters after privacy protection processing to the server side; a server side, for grouping and aggregating the heterogeneous data uploaded by each client according to the final privacy protection technology used by each client to perform the privacy protection task after reaching the deadline T, and updating the global model using the aggregation result. The present invention realizes data aggregation in the federated learning scenario under the condition of multi-party resource imbalance, and efficiently updates the global model without leaking the local model parameters of each client. Based on the client's local resources and data volume, select appropriate privacy protection technology to complete model training, data processing and transmission in a limited time as much as possible, to ensure the efficiency and effectiveness of global model training.

[0050] The above scheme is described in detail below.

[0051] Example 1

[0052] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart of the main steps of a privacy enhancement method for power federated learning that resists data heterogeneity according to an embodiment of the present invention. Figure 1 As shown, the privacy enhancement method for power federated learning against data heterogeneity in the embodiment of the present invention mainly includes the following steps:

[0053] Step S101: publishing the deadline T for each round of data submission to each client, so that the client selects an initial privacy protection technology to be used for executing the privacy protection task based on the deadline T, and notifies the server of the initial privacy protection technology;

[0054] Step S102: determining the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client, so that after the client performs the training task on the local training model, it uses its corresponding final privacy protection technology to perform privacy protection processing on the model parameters of the local training model, and uploads the model parameters after the privacy protection processing to the server;

[0055] Step S103: after the deadline T is reached, the heterogeneous data uploaded by each client is grouped and aggregated according to the final privacy protection technology adopted by each client to perform the privacy protection task, and the global model is updated using the aggregation result.

[0056] The privacy protection technology includes differential privacy technology and secure multi-party computing technology. In this embodiment, the differential privacy technology can adopt a segmentation mechanism under differential privacy, and the secure multi-party computing technology can adopt a Paillier encryption algorithm under multi-party secure computing.

[0057] In this embodiment, the final privacy protection technology used by each client to perform the privacy protection task is determined based on the initial privacy protection technology corresponding to each client, including:

[0058] Step a. Determine whether the number of clients whose initial privacy protection technologies are secure multi-party computing technologies is 1. If so, select a client from other clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is closest to the deadline T as a matching client, and set the current privacy protection technology corresponding to the matching client to the secure multi-party computing technology, set the current privacy protection technology corresponding to the clients other than the matching client to the differential privacy technology, and then execute step b. Alternatively, set the current privacy protection technology corresponding to all clients to the differential privacy technology and then execute step b. Otherwise, execute step b.

[0059] Step b. Determine whether the total number of clients is less than a preset threshold. If so, set the current privacy protection technology corresponding to all clients to the secure multi-party computing technology and then execute step d. Otherwise, execute step c.

[0060] Step c. Determine whether the number of clients whose current privacy protection technology is differential privacy technology is less than a preset threshold. If so, select n clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is the largest among the clients whose current privacy protection technology is secure multi-party computing technology as matching clients, and set the current privacy protection technology corresponding to the matching clients to differential privacy technology before executing step d. Otherwise, execute step d.

[0061] Step d. Determine whether the number of clients whose current privacy protection technologies are secure multi-party computing technologies is 0 or greater than or equal to 2. If so, use the current privacy protection technology corresponding to the client as the final privacy protection technology corresponding to the client. Otherwise, use the final privacy protection technology corresponding to all clients as secure multi-party computing technology.

[0062] Here, n is the difference between the preset threshold and the number of differential privacy technologies corresponding to the client's current privacy protection technology.

[0063] Among them, the preset threshold is 10.

[0064] In this embodiment, the grouping and aggregating of heterogeneous data uploaded by each client according to the final privacy protection technology used by each client to perform the privacy protection task includes:

[0065] Aggregate the data uploaded by the client whose final privacy protection technology used to perform the privacy protection task is the secure multi-party computing technology into a first group of data, and aggregate the data uploaded by the client whose final privacy protection technology used to perform the privacy protection task is the differential privacy technology into a second group of data;

[0066] Performing a decryption operation on the first grouped data using a secure multi-party computing technology to obtain a mean value corresponding to the first grouped data, and performing a mean estimation on the second grouped data using a perturbed data correction scheme in a differential privacy technology to obtain a mean value corresponding to the second grouped data;

[0067] An aggregation result is determined based on the mean value corresponding to the first grouped data and the mean value corresponding to the second grouped data.

[0068] In one embodiment, the aggregation result is: Aw1+Bw2, where A is the mean value corresponding to the first grouped data, w1 is the first weight, B is the mean value corresponding to the second grouped data, and w2 is the second weight.

[0069] Example 2

[0070] See attached Figure 2 , Figure 2 FIG. 1 is a flow chart of the main steps of a privacy enhancement method for power federated learning that resists data heterogeneity according to an embodiment of the present invention. Figure 2 As shown, the privacy enhancement method for power federated learning against data heterogeneity in the embodiment of the present invention mainly includes the following steps:

[0071] Step S201: selecting an initial privacy protection technology for executing the privacy protection task based on the deadline T sent by the server, and informing the server of the initial privacy protection technology, so that the server determines the final privacy protection technology for each client to execute the privacy protection task based on the initial privacy protection technology corresponding to each client;

[0072] Step S202: After executing the training task on the local training model, the model parameters of the local training model are subjected to privacy protection processing using the corresponding final privacy protection technology, and the model parameters after the privacy protection processing are uploaded to the server side, so that after the deadline T is reached, the server side groups and aggregates the heterogeneous data uploaded by each client according to the final privacy protection technology used by each client to execute the privacy protection task, and uses the aggregation result to update the global model.

[0073] In this embodiment, the initial privacy protection technology used to perform the privacy protection task is selected based on the deadline T sent by the server, including:

[0074] Obtain the sum t1 of the time required for the client to perform the training task and the time required to perform the privacy protection task using differential privacy technology, and the sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using secure multi-party computing technology;

[0075] Determine whether the client's t2 is less than T. If so, let the client use the multi-party secure computing technology as the initial privacy protection technology to perform the privacy protection task. Otherwise, let the client use the differential privacy technology as the initial privacy protection technology to perform the privacy protection task.

[0076] In this embodiment, when the client fails to complete the local training model execution training task and the privacy protection task within the deadline T, no data is uploaded to the server.

[0077] Example 3

[0078] Based on the same inventive concept, the present invention also provides a power federated learning privacy enhancement device that is resistant to data heterogeneity, and the power federated learning privacy enhancement device that is resistant to data heterogeneity includes:

[0079] The server is used to publish the deadline T for each round of data submission to each client;

[0080] A client, configured for the client to select an initial privacy protection technology to be used for executing the privacy protection task based on the deadline T, and to inform the server of the initial privacy protection technology;

[0081] The server is used to determine the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client;

[0082] The client is used to perform privacy protection processing on the model parameters of the local training model using its corresponding final privacy protection technology after executing the training task on the local training model, and upload the model parameters after privacy protection processing to the server;

[0083] The server side is used to group and aggregate the heterogeneous data uploaded by each client according to the final privacy protection technology adopted by each client to perform the privacy protection task after the deadline T is reached, and update the global model using the aggregation result.

[0084] In a specific implementation, the simulation experiment uses a public data set of smart grids, which contains 60,000 data, and the server hopes to build an SVM classification model. At the same time, it is assumed that the experiment includes 20 clients, of which 15 clients have 4-core processors and 4GB of memory, and the other 5 clients have 8-core processors and 8GB of memory. During the simulation, the data set is randomly divided into 50,000 training data and 10,000 test data. The 50,000 training data are randomly distributed to 20 clients as their local data according to exponential distribution (less data is allocated to clients with small computing resources). Table 1 lists the situations in which clients select different privacy technologies when the server sets different deadlines. N1 is the number of clients that select differential privacy technology, N2 is the number of clients that select secure multi-party computing technology, and the weight setting when the final aggregation results are obtained.

[0085] Table 1

[0086] T 1min 2min 3min N1,N2 18，2 16，4 10，10 <![CDATA[w1,w2]]> 0.5，0.5 0.3，0.7 0.2，0.8

[0087] In order to evaluate the effectiveness of the design of the present invention, the classification accuracy of the global model trained on the server side will be verified under different deadline settings, and the effectiveness of the server-side estimation of the mean will also be measured by the mean square error indicator. The experimental results are listed in Table 2. The original accuracy of SVM on the smart grid dataset is about 97%.

[0088] Table 2

[0089] T 1min 2min 3min Accuracy 91% 93% 96% MSE 0.1 0.03 0.007

[0090] As shown in Table 1, when the deadline T is set longer, more clients will choose secure multi-party computing technology. Compared with differential privacy technology, the encryption and decryption algorithms in secure multi-party computing consume more computing overhead, but will not affect the accuracy of the final statistical results, so it will bring higher data utility, as shown in Table 2, that is, the accuracy (or error) will increase with the increase of the deadline.

[0091] Example 4

[0092] Based on the same inventive concept, the present invention also provides a computer device, which includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions in a computer storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of a power federated learning privacy enhancement method against data heterogeneity in the above embodiment.

[0093] Example 5

[0094] Based on the same inventive concept, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It can be understood that the computer-readable storage medium here can include both built-in storage media in a computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of a privacy enhancement method for power federated learning that resists data heterogeneity in the above embodiment.

[0095] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take 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 code.

[0096] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0097] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A privacy enhancement method for power federated learning that is resistant to data heterogeneity, characterized in that: The method comprises: The deadline T for each round of data submission is published to each client, so that the client selects the initial privacy protection technology to be used for executing the privacy protection task based on the deadline T, and informs the server of the initial privacy protection technology; Determine the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client, so that after the client performs the training task on the local training model, it uses its corresponding final privacy protection technology to perform privacy protection processing on the model parameters of the local training model, and uploads the model parameters after the privacy protection processing to the server; After the deadline T is reached, the heterogeneous data uploaded by each client is grouped and aggregated according to the final privacy protection technology used by each client to perform the privacy protection task, and the aggregation result is used to update the global model; The determining of the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client includes: Step a. Determine whether the number of clients whose initial privacy protection technologies are secure multi-party computing technologies is 1. If so, select a client from other clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is closest to the deadline T as a matching client, and set the current privacy protection technology corresponding to the matching client to the secure multi-party computing technology, set the current privacy protection technology corresponding to the clients other than the matching client to the differential privacy technology, and then execute step b. Alternatively, set the current privacy protection technology corresponding to all clients to the differential privacy technology and then execute step b. Otherwise, execute step b. Step b. Determine whether the total number of clients is less than a preset threshold. If so, set the current privacy protection technology corresponding to all clients to the secure multi-party computing technology and then execute step d. Otherwise, execute step c. Step c. Determine whether the number of clients whose current privacy protection technology is differential privacy technology is less than a preset threshold. If so, select n clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is the largest among the clients whose current privacy protection technology is secure multi-party computing technology as matching clients, and set the current privacy protection technology corresponding to the matching clients to differential privacy technology before executing step d. Otherwise, execute step d. Step d. Determine whether the number of clients whose current privacy protection technologies are secure multi-party computing technologies is 0 or greater than or equal to 2. If so, use the current privacy protection technology corresponding to the client as the final privacy protection technology corresponding to the client. Otherwise, use the final privacy protection technology corresponding to all clients as secure multi-party computing technology. Wherein, n is the difference between the preset threshold and the number of differential privacy technologies corresponding to the client; The client selects an initial privacy protection technology for executing the privacy protection task based on the deadline T, including: Obtain the sum t1 of the time required for the client to perform the training task and the time required to perform the privacy protection task using differential privacy technology, and the sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using secure multi-party computing technology; Determine whether the client's t2 is less than T. If so, let the client use the multi-party secure computing technology as the initial privacy protection technology to perform the privacy protection task. Otherwise, let the client use the differential privacy technology as the initial privacy protection technology to perform the privacy protection task.

2. The method according to claim 1, characterized in that The privacy protection technology includes: differential privacy technology and secure multi-party computing technology.

3. The method according to claim 1, characterized in that The preset threshold is 10.

4. The method according to claim 1, characterized in that The method of grouping and aggregating the heterogeneous data uploaded by each client according to the final privacy protection technology used by each client to perform the privacy protection task includes: Aggregate the data uploaded by the client whose final privacy protection technology used to perform the privacy protection task is the secure multi-party computing technology into a first group of data, and aggregate the data uploaded by the client whose final privacy protection technology used to perform the privacy protection task is the differential privacy technology into a second group of data; Performing a decryption operation on the first grouped data using a secure multi-party computing technology to obtain a mean value corresponding to the first grouped data, and performing a mean estimation on the second grouped data using a perturbed data correction scheme in a differential privacy technology to obtain a mean value corresponding to the second grouped data; An aggregation result is determined based on the mean value corresponding to the first grouped data and the mean value corresponding to the second grouped data.

5. The method according to claim 4, characterized in that The aggregation result is: Aw1+Bw2, A is the mean value corresponding to the first grouped data, w1 is the first weight, B is the mean value corresponding to the second grouped data, and w2 is the second weight.

6. A privacy enhancement method for power federated learning that is resistant to data heterogeneity, characterized in that: The method comprises: Selecting an initial privacy protection technology for executing the privacy protection task based on the deadline T sent by the server, and informing the server of the initial privacy protection technology, so that the server determines a final privacy protection technology for each client to execute the privacy protection task based on the initial privacy protection technology corresponding to each client; After executing the training task on the local training model, the model parameters of the local training model are processed for privacy protection using the corresponding final privacy protection technology, and the model parameters after the privacy protection processing are uploaded to the server, so that the server groups and aggregates the heterogeneous data uploaded by each client according to the final privacy protection technology used by each client to execute the privacy protection task after the deadline T, and updates the global model using the aggregation result; The determining of the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client includes: Step a. Determine whether the number of clients whose initial privacy protection technologies are secure multi-party computing technologies is 1. If so, select a client from other clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is closest to the deadline T as a matching client, and set the current privacy protection technology corresponding to the matching client to the secure multi-party computing technology, set the current privacy protection technology corresponding to the clients other than the matching client to the differential privacy technology, and then execute step b. Alternatively, set the current privacy protection technology corresponding to all clients to the differential privacy technology and then execute step b. Otherwise, execute step b. Step b. Determine whether the total number of clients is less than a preset threshold. If so, set the current privacy protection technology corresponding to all clients to the secure multi-party computing technology and then execute step d. Otherwise, execute step c. Step c. Determine whether the number of clients whose current privacy protection technology is differential privacy technology is less than a preset threshold. If so, select n clients whose sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using the secure multi-party computing technology is the largest among the clients whose current privacy protection technology is secure multi-party computing technology as matching clients, and set the current privacy protection technology corresponding to the matching clients to differential privacy technology before executing step d. Otherwise, execute step d. Step d. Determine whether the number of clients whose current privacy protection technologies are secure multi-party computing technologies is 0 or greater than or equal to 2. If so, use the current privacy protection technology corresponding to the client as the final privacy protection technology corresponding to the client. Otherwise, use the final privacy protection technology corresponding to all clients as secure multi-party computing technology. Wherein, n is the difference between the preset threshold and the number of differential privacy technologies corresponding to the client; The initial privacy protection technology used to perform the privacy protection task is selected based on the deadline T sent by the server, including: Obtain the sum t1 of the time required for the client to perform the training task and the time required to perform the privacy protection task using differential privacy technology, and the sum t2 of the time required to perform the training task and the time required to perform the privacy protection task using secure multi-party computing technology; Determine whether the client's t2 is less than T. If so, let the client use the multi-party secure computing technology as the initial privacy protection technology to perform the privacy protection task. Otherwise, let the client use the differential privacy technology as the initial privacy protection technology to perform the privacy protection task.

7. The method according to claim 6, characterized in that When the client fails to complete the local training model execution training task and the privacy protection task within the deadline T, no data will be uploaded to the server.

8. A device for enhancing the privacy of power federated learning against data heterogeneity based on any one of claims 1 to 7, characterized in that: The device comprises: The server is used to publish the deadline T for each round of data submission to each client; A client, configured for the client to select an initial privacy protection technology to be used for executing the privacy protection task based on the deadline T, and to inform the server of the initial privacy protection technology; The server is used to determine the final privacy protection technology used by each client to perform the privacy protection task based on the initial privacy protection technology corresponding to each client; The client is used to perform privacy protection processing on the model parameters of the local training model using its corresponding final privacy protection technology after executing the training task on the local training model, and upload the model parameters after privacy protection processing to the server; The server side is used to group and aggregate the heterogeneous data uploaded by each client according to the final privacy protection technology adopted by each client to perform the privacy protection task after the deadline T is reached, and update the global model using the aggregation result.

9. A computer device, characterized in that: include: one or more processors; The processor is used to store one or more programs; When the one or more programs are executed by the one or more processors, the privacy enhancement method for power federated learning that is resistant to data heterogeneity as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, the privacy enhancement method for power federated learning resistant to data heterogeneity as described in any one of claims 1 to 7 is implemented.

Images