Aviation data protection method and equipment based on national cryptographic algorithm, and medium

Abstract

The invention provides an aviation data protection method and device based on a national secret algorithm and a medium, and relates to the technical field of information security, and the method comprises the steps: obtaining a target aircraft client ID list corresponding to personalized federated learning, constructing a global model corresponding to personalized federated learning, initializing Pi, 1 and Bi, 1, obtaining initial update results Pi, 2 and Bi, 2, obtaining encrypted APi, 2, and obtaining a target aircraft client ID list corresponding to personalized federated learning; obtaining an aggregation result Ci, 2 of the global model, obtaining a re-updating result Qi, 2 of the Ai, judging whether the global model converges based on the Ci, 2, if the global model converges, ending training, obtaining a target global model, and if the global model does not converge, assigning the Ci, 2 to Bi, 1, assigning the Qi, 2 to Pi, 1, and continuing circulation until the global model converges; therefore, the integrating degree of the global model and the target aircraft client is improved.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular, to an aviation data protection method, device and medium based on national cryptographic algorithms. Background Art

[0002] With the development of information technology, data privacy and security issues have become prominent. Existing machine learning methods lack security measures, and users are reluctant to upload private data. Federated learning enables clients to collaboratively train a global model without uploading private data through decentralized datasets and privacy protection mechanisms, effectively preventing data leakage. This technology has made progress in personalized federated learning and model security and will play a role in fields such as healthcare and finance in the future. Traditional federated learning is based on Google's FedAvg algorithm, where all clients share the same model and pursue global generalization performance, but ignore local personalized needs. Personalized federated learning, on the other hand, adapts to the data distribution of different clients to improve the performance of local models and better meet the personalized needs of users.

[0003] To address data heterogeneity, personalized federated learning (PFL) provides a customized model for each client instead of a unified global model. Existing PFL methods still rely on traditional frameworks, generating a single global model by aggregating the parameters of all clients, ignoring the distribution differences between clients. FL-MOOL encourages local models to improve based on historical versions, reducing the distance from the global model and enhancing personalized performance.

[0004] However, the above methods also have the following technical problems:

[0005] Federated learning enables decentralized training without sharing local data. In each round of communication, FedAvg randomly selects a subset of devices, and each selected device trains a local model using the same learning rate and local batch of local data. Each local model is updated using stochastic gradient descent. The local models transmit the updated parameters to the central server for weighted averaging. Finally, the weighted-averaged global model is sent back to each client. However, the data distributions across clients are essentially non-independent and identically distributed, and statistical heterogeneity makes it difficult for the shared global model trained and generalized to all clients to fit well. Summary of the Invention

[0006] In view of the above technical problems, the technical solution adopted by the present invention is as follows:

[0007] According to a first aspect of the present invention, there is provided an aviation data protection method based on national cryptographic algorithms, which is used for aviation data protection in personalized federated learning. The method includes the following steps:

[0008] S100, Obtain the list of target aircraft client IDs A = {A 1 , A 2 , …, A i , …, A k}, where A i is the i-th target aircraft client ID, and the value range of i is from 1 to k, where k is the number of target aircraft client IDs;

[0009] S200, Construct the global model corresponding to the personalized federated learning;

[0010] S300, Initialize P i，1 and B i，1 , where P i，1 is the personalized layer parameter corresponding to A i , and B i，1 is the global general knowledge parameter corresponding to A i ;

[0011] S400, Obtain the initial update results P i，2 and B i，2 ; among them, update P i based on the preset local dataset corresponding to A i，1 to obtain P i，2 , and update B i based on the preset local dataset corresponding to A i，1 to obtain B i，2 ;

[0012] S500, Obtain the encrypted AP i，2 , and transmit the encrypted data AP i，2 to the global model corresponding to the personalized federated learning, where the target aircraft client corresponding to A i uses the national cryptographic algorithm to encrypt P i，2 to obtain the encrypted data AP i，2 ;

[0013] S600, Obtain the aggregation result C i，2 of the global model, where the global model performs aggregation training on B i，2 to obtain C i，2 ;

[0014] S700, Obtain the re-update result Q i of A i，2 , where the target aircraft client corresponding to A i freezes the global general knowledge parameter B i，2 , and uses the preset local dataset corresponding to A i to re-update P i，2 to obtain Q i，2 ;

[0015] S800, based on C i，2 Determine whether the global model converges. If the global model converges, end the training and obtain the target global model. If the global model does not converge, assign C i，2 to B i，1 , and assign Q i，2 to P i，1 , and execute S400.

[0016] According to the second aspect of the present invention, there is provided a non-transitory computer-readable storage medium storing a computer program which is loaded and executed by a processor to implement the foregoing method.

[0017] According to the third aspect of the present invention, there is provided an electronic device including: a processor, a memory, and a computer program stored on the memory and executable on the processor, and the processor implements the foregoing method when executing the computer program.

[0018] The present invention has at least the following beneficial effects: In summary, obtain the target aircraft client ID list corresponding to personalized federated learning, construct the global model corresponding to personalized federated learning, initialize P i，1 and B i，1 , obtain the initial update results P i，2 and B i，2 , obtain the encrypted AP i，2 , obtain the aggregation result C of the global model i，2 , obtain A i 's re-update result Q i，2 , based on C i，2 Determine whether the global model converges. If the global model converges, end the training and obtain the target global model. If the global model does not converge, assign C i，2 to B i，1 , and assign Q i，2 to P i，1 , and continue to loop. The present invention sets the global general parameters corresponding to the target aircraft client, and by sharing the global general parameters, each aircraft client can achieve personalized modeling, improving the fit between the global model and the target aircraft client. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0020] Figure 1Flowchart of an aviation data protection method based on national cryptographic algorithms provided by an embodiment of the present invention. Detailed implementation manners

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the protection scope of the present invention.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above drawings are used to distinguish similar tasks, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units does not necessarily need to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0023] An embodiment of the present invention provides an aviation data protection method based on national cryptographic algorithms, as Figure 1 shown. The method is used for aviation data protection in personalized federated learning, and the method includes the following steps:

[0024] S100. Obtain a list A = {A 1 , A 2 , …, A i , …, A k} of target aircraft client IDs corresponding to personalized federated learning, where A i is the i-th target aircraft client ID, and the value range of i is from 1 to k, where k is the number of target aircraft client IDs.

[0025] Among them, the target aircraft client ID is the unique identifier of the target aircraft client.

[0026] S200. Construct a global model corresponding to personalized federated learning. Specifically, those skilled in the art know that any global model in the prior art belongs to the protection scope of the present invention. For example, the global model is a deep neural network model, or the global model is a convolutional neural network model, etc.

[0027] S300. Initialize P i，1 and B i，1, where P i，1 is the personalized layer parameter corresponding to A i , and B i，1 is the global general knowledge parameter corresponding to A i . It can be understood that in the face of the federated optimization problem of heterogeneous data, the present invention adopts a parameter decomposition strategy, corresponding the global general knowledge parameter to the target aircraft client one by one, and the personalized layer parameter to the target aircraft client one by one.

[0028] S400, obtain the initial update result P i，2 and B i，2 ; among them, based on the preset local dataset corresponding to A i update P i，1 to obtain P i，2 , and based on the preset local dataset corresponding to A i update B i，1 to obtain B i，2 .

[0029] Specifically, the target aircraft client uses the corresponding preset local dataset to train the personalized layer parameter and the global general knowledge parameter, and then uses an optimization algorithm to update the personalized layer parameter and the global general knowledge parameter. In an embodiment of the present invention, based on the preset local dataset, the gradient of the preset loss function with respect to the personalized parameter is calculated by the backpropagation algorithm, and then the personalized layer parameter and the global general knowledge parameter are updated using the stochastic gradient descent algorithm.

[0030] S500, obtain the encrypted AP i，2 , and transmit the encrypted data AP i，2 to the global model corresponding to the personalized federated learning, where the target aircraft client corresponding to A i uses the national cryptography algorithm to encrypt P i，2 to obtain the encrypted data AP i，2 .

[0031] Specifically, the present invention uses the national cryptography algorithm to encrypt P i，2 to obtain the encrypted data AP i，2 , and transmit the encrypted data AP i，2 to the global model corresponding to the personalized federated learning. By using the national cryptography algorithm for encryption, the security, integrity, and authenticity of the data transmission process are ensured.

[0032] S600, obtain the aggregation result C i，2 of the global model, where the global model performs aggregation training on B i，2 to obtain C i，2 .

[0033] In an embodiment of the present invention, in S600, the global model performs aggregation training on B i，2Perform aggregation training to obtain C i，2 , further including: C i，2 = C i，2 = (1 / k)×(Σ k i＝1 B i，2 ).

[0034] In another embodiment of the present invention, the global model performs aggregation training on B i，2 using the weighted average method to obtain C i，2 . i，2 .

[0035] S700, obtain the updated result Q i of A i，2 , where the target aircraft client corresponding to A i freezes the global general knowledge parameter B i，2 and uses the preset local dataset corresponding to A i to update P i，2 again to obtain Q i，2 .

[0036] Among them, in S700, the method for the re-update of A i is the same as the method for updating P i based on the preset local dataset corresponding to A i，1 in S400. It can be understood that P i，2 is further optimized to make it more adaptable to the data distribution of the target aircraft client and better serve the specific needs of the target aircraft client.

[0037] S800, based on C i，2 judge whether the global model converges. If the global model converges, end the training and obtain the target global model. If the global model does not converge, assign C i，2 to B i，1 , and assign Q i，2 to P i，1 , and execute S400.

[0038] Specifically, those skilled in the art know that any method for judging whether the global model converges based on C i，2 in the prior art belongs to the protection scope of the present invention and will not be elaborated here.

[0039] Furthermore, the present invention also includes that if the number of assignment times exceeds the preset threshold and the global model still does not converge, end the training and use the global model at this time as the target global model.

[0040] In summary, obtain the target aircraft client ID list corresponding to personalized federated learning, construct the global model corresponding to personalized federated learning, and initialize P i，1 and Bi，1 , obtain the initial update result P i，2 and B i，2 , obtain the encrypted AP i，2 , obtain the aggregation result C of the global model i，2 , obtain A i 's re-update result Q i，2 , based on C i，2 judge whether the global model converges. If the global model converges, end the training and obtain the target global model. If the global model does not converge, assign C i，2 to B i，1 , and assign Q i，2 to P i，1 , continue to loop. In the present invention, by setting the global general parameters corresponding to the target aircraft client, and sharing the global general parameters, each aircraft client can achieve personalized modeling, making the global model and the aircraft client more compatible.

[0041] Among them, S100 further includes:

[0042] S110, obtain the initial aircraft client ID list E = {E 1 , E 2 , …, E j , …, E n}, where E j is the j-th initial aircraft client ID corresponding to personalized federated learning, and the value range of j is from 1 to n, where n is the number of initial aircraft client IDs, and the initial aircraft clients are all aircraft clients participating in personalized federated learning. Among them, the initial aircraft client ID is the unique identifier of the initial aircraft client.

[0043] S120, use the reservoir sampling algorithm to extract k from E 1 to E n as the target aircraft client IDs, so as to obtain the target aircraft client ID list, where k is less than n.

[0044] Specifically, for the situation where parameter data packets of a large number of initial aircraft clients arrive at the server for processing simultaneously, the present invention adopts a method combining a multi-thread pool and the reservoir sampling algorithm. This method can fairly randomly extract a specified number of samples from the initial aircraft client IDs and perform authentication operations in parallel, thus significantly improving the efficiency of model authentication.

[0045] Specifically, in S500, the target aircraft client corresponding to A i uses the national secret algorithm to encrypt P i，2 to obtain the encrypted data AP i，2 further includes:

[0046] S510, for P i，2 Use the national cryptographic SM3 algorithm to perform hash calculation to obtain the message digest value D i，2 .

[0047] S520, use the national cryptographic SM2 algorithm to obtain the message signature S i，2 = S(K priv , D i，2 ), where S() is the national cryptographic SM2 algorithm, and K priv is the private key of the aircraft client.

[0048] S530, obtain the encrypted data AP i，2 = (D i，2 , S i，2 , P i，2 ). Among them, the global model also stores the public key K pub .

[0049] In summary, for P i，2 Use the national cryptographic SM3 algorithm to perform hash calculation to obtain the message digest value D i，2 , for P i，2 Use the national cryptographic SM3 algorithm to perform hash calculation to obtain the message digest value D i，2 , use the national cryptographic SM2 algorithm to obtain the message signature S i，2 = S(K priv , D i，2 ), obtain the encrypted data AP i，2 = (D i，2 , S i，2 , P i，2 ). The present invention adopts domestic cryptographic algorithms SM2 and SM3 to perform digital signature and verification on the model parameters trained by the aircraft client, realizing lightweight data authentication technology within the system.

[0050] The embodiment of the present invention also applies digital signature technology in the training of the federated learning model, performs hash operation on the personalized layer parameters with the SM3 algorithm to generate a message digest value, and uses the SM2 algorithm private key of each aircraft client to digitally sign the generated message digest value. Finally, when aggregating at the global ground layer server, the server uses the SM2 algorithm public key to verify the signature of the local client. After the verification passes, the server uses the SM3 algorithm to perform hash operation on the information again to generate a message digest, and compares it with the message digest sent by the local aircraft client. If the results are consistent, the next step is carried out.

[0051] Specifically, based on A i corresponding preset local dataset to update P i，1 to obtain P i，2 , further includes:

[0052] Pi，2 = P i，1 - η × g i , where η is the preset learning rate, and g i is the preset loss function corresponding to A i at P i，1 gradient.

[0053] Furthermore, update B i based on the preset local dataset corresponding to A i，1 to obtain B i，2 , and it also includes: B i，2 = B i，1 - η × g i .

[0054] Specifically, randomly select a sample from the preset local dataset corresponding to A i to calculate the gradient g i , and use g i to update P i，1 and B i，1 .

[0055] Even further, the client corresponding to A i freezes the global general knowledge parameter B i，2 , and uses the preset local dataset corresponding to A i to update P i，2 again to obtain Q i，2 , and it also includes the following steps:

[0056] S710, the client corresponding to A i freezes the global general knowledge parameter B i，2 .

[0057] S720, obtain q i , where q i is the preset loss function corresponding to A i at P i，2 gradient. Specifically, randomly select a sample from the preset local dataset corresponding to A i to calculate the gradient q i .

[0058] S730, obtain Q i，2 , where Q i，2 satisfies the following condition: Q i，2 = P i，2 - η × q i .

[0059] In summary, the client corresponding to A i freezes the global general knowledge parameter B i，2 , obtains q i , and based on q iObtain Q i，2 。

[0060] An embodiment of the present invention further provides a non-transitory computer-readable storage medium, which can be disposed in an electronic device to store a computer program related to a method in a method embodiment. The computer program is loaded and executed by the processor to implement the method provided in the above embodiment.

[0061] An embodiment of the present invention further provides an electronic device, including: a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method provided in the above embodiment is implemented.

[0062] An embodiment of the present invention further provides a computer program product, which includes program code. When the program product runs on an electronic device, the program code is used to cause the electronic device to execute the steps in the method according to various exemplary embodiments of the present invention described above in this specification.

[0063] Although some specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are only for illustration and not for limiting the scope of the present invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present invention.

Claims

1. A method for protecting aviation data based on a national secret algorithm, characterized in that: The method is used for aviation data protection in personalized federated learning, and the method comprises the following steps: S100, obtaining a target aircraft client ID list A corresponding to personalized federated learning = {A1, A2, ..., A i , …, A k }, A i is the i-th target aircraft client ID, i ranges from 1 to k, and k is the number of target aircraft client IDs; S200, building a global model corresponding to personalized federated learning; S300, Initialize P i，1 and B i，1 , where P i，1 Yes A i The corresponding personalization layer parameters, B i，1 Yes A i The corresponding global general parameters; S400, obtaining the initial update result P i，2 and B i，2 ; Among them, based on A i The corresponding preset local data set pair P i，1 Update to get P i，2 , based on A i The corresponding preset local data set pair B i，1 Update to get B i，2 ; S500, Get Encrypted AP i，2 , and encrypt the data AP i，2 Transferred to the global model corresponding to personalized federated learning, where A i The corresponding target aircraft client uses the national encryption algorithm to i，2 Encrypt and obtain encrypted data AP i，2 ; S600, obtaining the aggregation result C of the global model i，2 , where the global model is i，2 Perform aggregate training to obtain C i，2 ; S700, Get A i Updated result Q i，2 , where A i The corresponding target aircraft client freezes the global common parameter B i，2 、Use A i The corresponding preset local data set pair P i，2 Update again to get Q i，2 ; S800, based on C i，2 Determine whether the global model converges. If the global model converges, end the training and obtain the target global model. If the global model does not converge, C i，2 Assign to B i，1 , and Q i，2 Assign to P i，1 , execute S400.

2. The aviation data protection method based on the national secret algorithm according to claim 1 is characterized in that: The S100 also includes: S110, obtaining the initial aircraft client ID list E corresponding to the personalized federated learning = {E1, E2, ..., E j , …, E n }, E j is the jth initial aircraft client ID corresponding to the personalized federated learning, where j ranges from 1 to n, and n is the number of initial aircraft client IDs. The initial aircraft clients are all aircraft clients participating in the personalized federated learning; S120, using reservoir sampling algorithm from E1 to E n K are extracted as target aircraft client IDs, thereby obtaining a target aircraft client ID list, where k is less than n.

3. The aviation data protection method based on the national secret algorithm according to claim 1 is characterized in that: In S500, A i The corresponding target aircraft client uses the national encryption algorithm to i，2 Encrypt and obtain encrypted data AP i，2 Also includes: S510, for P i，2 Use the national secret SM3 algorithm to perform hash calculation and obtain the message digest value D i，2 ; S520, use the national secret SM2 algorithm to obtain the message signature S i，2 =S(K priv , D i，2 ), where S() is the national secret SM2 algorithm, K priv It is the aircraft client private key; S530, obtain encrypted data AP i，2 =(D i，2 , S i，2 , P i，2 ).

4. The aviation data protection method based on the national secret algorithm according to claim 1 is characterized in that: In S600, the global model is i，2 Perform aggregate training to obtain C i，2 , also includes: C i，2 =(1 / k)×(Σ k i＝1 B i，2 ).

5. The aviation data protection method based on the national secret algorithm according to claim 1 is characterized in that: Based on A i The corresponding preset local data set pair P i，1 Update to get P i，2 , also includes: P i，2 =P i，1 -η×g i , where η is the preset learning rate, g i A i The corresponding preset loss function is in P i，1 gradient.

6. The aviation data protection method based on the national secret algorithm according to claim 5 is characterized in that: Based on A i The corresponding preset local data set pair B i，1 Update to get B i，2 , also includes: B i，2 =B i，1 -η×g i .

7. The aviation data protection method based on the national secret algorithm according to claim 5 is characterized in that: A i The corresponding client freezes the global common parameter B i，2 , use A i The corresponding preset local data set pair P i，2 Update again to get Q i，2 , further comprising the following steps: S710, A i The corresponding client freezes the global common parameter B i，2 ; S720, get q i , where q i A i The corresponding preset loss function is in P i，2 The gradient of S730, Get Q i，2 , where Q i，2 The following conditions are met: Q i，2 =P i，2 -η×q i .

8. The aviation data protection method based on the national secret algorithm according to claim 1 is characterized in that: The target aircraft client ID is the unique identifier of the target aircraft client.

9. A non-transitory computer-readable storage medium, characterized in that: The storage medium stores a computer program, which is loaded and executed by a processor to implement the aviation data protection method based on the national encryption algorithm as described in any one of claims 1 to 8.

10. An electronic device, comprising: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the aviation data protection method based on the national secret algorithm as described in any one of claims 1 to 8 when executing the computer program.

Images