Internet of vehicles federal forgetting learning method and system based on group learning and homomorphic encryption

Abstract

The invention belongs to the technical field of privacy protection, and discloses an Internet of Vehicles federal forgetting learning method and system based on group learning and homomorphic encryption, and the method comprises the steps: constructing a model, and carrying out the training; the target vehicle node sends a forgetting request, the RSU node issues the global model parameters to the target vehicle node for reverse correction training, the model parameters subjected to reverse correction are uploaded to the RSU node to calculate the change rate of the model parameters, the non-target vehicle node carries out forward correction training on the model parameters subjected to reverse correction, and the change rate of the model parameters subjected to reverse correction is obtained. Sending the forward correction parameter to an RSU node for edge aggregation to obtain an SL forgetting model; and when the change rate of the model parameters is smaller than a set threshold value, the target vehicle node quits the federated system, the correction training is finished, and the RSU node uploads the SL forgetting model parameters to the server to update the global model. According to the method, toxic data samples can be effectively eliminated, the balance between the model performance and the retraining overhead is realized, and the privacy of vehicle nodes is protected while the model performance is not damaged.

Description

Technical Field

[0001] The present invention relates to the field of privacy protection technology, and in particular to a method and system for federated forgetting learning in an Internet of Vehicles based on group learning and homomorphic encryption. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the acceleration of urbanization, traditional traffic management methods can no longer effectively cope with increasingly complex traffic conditions, and the demand for intelligent transportation has increased dramatically. In this context, the concept of intelligent transportation system (ITS) came into being, providing a new solution for improving traffic efficiency and safety. In this system, the Internet of Vehicles (IoV) plays a vital role. Vehicles in the IoV continuously collect data on their own status (such as speed, location, fuel consumption, etc.) and surrounding environment (such as road conditions, pedestrians, other vehicles, etc.) through built-in sensors (such as cameras, radars, GPS, LIDAR, etc.), and exchange data and share information with other vehicles, road infrastructure, cloud systems, etc. in real time through vehicle communication systems, providing strong support for creating a safer and more efficient traffic environment. Although IoV has obvious advantages, it is still accompanied by many challenges. The high mobility and ever-changing topology of the IoV environment make users who share information face serious security and privacy risks. Therefore, it is crucial to ensure secure data sharing in IoV without leaking user-side privacy.

[0004] Machine learning (ML) is considered a promising technology to improve driving safety and achieve intelligent traffic management. However, due to factors such as latency requirements, bandwidth limitations, computing resources, and privacy protection, traditional centralized ML methods are unable to cope with the massive data generated by IoV. To this end, researchers have proposed the use of federated learning (FL) to overcome the limitations of centralized ML. As an effective means of data privacy protection, FL can not only achieve AI collaboration without exposing local data privacy, technically break data silos, but also avoid centralized storage and transmission of data, effectively reduce bandwidth and computing pressure, and thus better adapt to the complex data environment of IoV. However, the FL mechanism still has security issues such as privacy leakage. On the one hand, the model parameters or gradients are essentially the relationship mapping of the original data set. Even if the data is not directly shared, the attacker can still restore part or all of the original data based on the model parameters or gradients. On the other hand, the attacker can infer information about the original training data through model inversion attacks. In the IoV scenario, the data may contain a lot of sensitive information, including driving routes, location records, driving habits, etc. These data not only record the user's activity trajectory, but also involve personal privacy. When individual identities change or users need to protect their privacy, the leakage of this information will cause serious privacy concerns. At the same time, the EU General Data Protection Regulation (GDPR) also stipulates that citizens have the "right to be forgotten" for their own data. Therefore, effective techniques are needed to delete or forget specific training data from the trained FL model.

[0005] To address these issues, Federated Forget Learning (FU) demonstrates its unique advantages. It aims to selectively remove specific training data points and their impact on the trained model, so that the updated model is as if it has never been trained on these data points. FU provides a "subtractive" capability that can effectively remove unauthorized, malicious, or outdated data points without consuming a lot of resources for comprehensive retraining. FU is migrated from the existing forget learning technology in the context of ML. The application of this technology enables the model to more efficiently meet dynamic privacy needs and legal compliance. However, with the development of FU technology, how to optimize computing resources and communication overhead while ensuring model performance, and strengthen privacy protection in the forgetting process, remains an important direction for future research. Current research mainly focuses on protecting the privacy of original data, but rarely involves the privacy protection of forgotten parameters and non-forgotten parameters. Attackers may restore private information through parameters, causing client privacy leakage. In addition, the federated forget learning process will generate additional computing and communication overhead. Most existing solutions rely on sacrificing space in exchange for time, which makes forget learning difficult to deploy on a large scale. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a federated forgetting learning method and system for Internet of Vehicles based on group learning and homomorphic encryption. By introducing homomorphic encryption technology, the forgetting operation is effectively performed without decryption, thereby protecting the privacy of forgotten and non-forgotten parameters of vehicle nodes. Group learning technology is combined to perform intra-group forgetting, thereby achieving a balance between model performance and retraining overhead.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a vehicle network federation forgetting learning method based on group learning and homomorphic encryption, comprising the following steps: Build a federated learning model for Internet of Vehicles and train the model; When the target vehicle node issues a forget request, the RSU node to which it belongs downloads the encrypted global model parameters from the aggregation server and sends them to the target vehicle node. The target vehicle node decrypts the global model parameters and performs multiple rounds of reverse correction training on them. The reverse-corrected model parameters are encrypted and uploaded to the RSU node. The RSU node calculates the rate of change of the model parameters and sends the reverse-corrected model parameters to the non-target vehicle nodes in the same group as the target vehicle node. The non-target vehicle nodes perform forward correction training on the reverse-corrected model parameters. In each round of forward correction training, the non-target vehicle nodes in the group perform homomorphic encryption on the correction parameter updates, and exchange encrypted correction updates with adjacent non-target vehicle nodes. Other non-target vehicle nodes aggregate the received encrypted correction updates with their own correction updates for the next round of forward correction training. Finally, all non-target vehicle nodes encrypt the correction updates and send them to the RSU nodes to which the group belongs for edge aggregation to obtain the SL forgetting model. When the rate of change of the model parameters calculated by the RSU node is less than the set threshold, the target vehicle node exits the federation system, the correction training ends, and the RSU node uploads the SL forgotten model parameters to the server, which updates the global model by aggregating the SL forgotten model parameters.

[0008] As an optional implementation, the Internet of Vehicles federated learning model includes a server, an RSU node and a vehicle node, wherein a plurality of RSU nodes are arranged at the lower layer of the server, and a plurality of vehicle nodes divided into different groups are arranged in the area of ​​each RSU node.

[0009] As an optional implementation method, the Internet of Vehicles federated learning model is trained as follows: The RSU node downloads the encrypted global model parameters of the current round from the server, and then sends them to the vehicle nodes of each group in its area. The vehicle node decrypts the global model parameters and adjusts them based on local data to generate local updates. The local updates are then encrypted and uploaded to the RSU node to which the group belongs. The RSU node performs edge aggregation on the encrypted local updates uploaded by each vehicle node, generates an SL model, and submits the SL model parameters and aggregation weights to the server. The server updates the global model through the aggregation weights of each RSU node, and sends the updated global model to each RSU node for the next round of training until the global model converges to obtain a trained Internet of Vehicles federated learning model.

[0010] As an optional implementation, it also includes passively forgetting the vehicle node, and the passive forgetting is: During the training of the federated learning model of the Internet of Vehicles, before the RSU node performs edge aggregation on the encrypted local updates uploaded by each vehicle node, it will calculate the cosine similarity between the encrypted local updates submitted by each vehicle node and the SL model parameters. When the result deviates significantly, it is regarded as a malicious node. The contribution of the malicious node is removed based on the historical global model, and k rounds of retraining are performed. The new local updates are encrypted and submitted to the RSU node to bridge the remaining node weights for edge aggregation.

[0011] As an optional implementation method, a federated learning model for Internet of Vehicles is constructed, specifically: Register RSU nodes and vehicle nodes, assign initial reputation values ​​to each node, and store the successfully registered node information on the blockchain; Initialize a machine learning model and broadcast it to all registered vehicle nodes that meet the task requirements through the RSU node to respond to the task; RSU nodes and vehicle nodes that wish to participate in the federated learning task send response messages to the blockchain and select nodes with higher reputation values ​​to participate in the task; The RSU node divides the vehicle nodes that have successfully joined the training task in its area into multiple different groups based on data distribution, and all vehicle nodes in the group are connected to the blockchain.

[0012] As an optional implementation, the vehicle node decrypts the global model parameters, specifically: The nodes participating in the federated learning task jointly execute the key generation protocol to generate public keys and private key shares. The vehicle nodes interact point-to-point through the blockchain, exchange private key shares, and reconstruct the private key using Lagrange interpolation, which is used to decrypt the global model parameters.

[0013] In a second aspect, the present invention provides a federated forgetting learning system for Internet of Vehicles based on group learning and homomorphic encryption, comprising: The model building and training module is configured to: build a federated learning model for Internet of Vehicles and train the model; The model correction training module is configured as follows: when the target vehicle node sends a forget request, the RSU node to which it belongs downloads the encrypted global model parameters from the aggregation server and sends them to the target vehicle node; the target vehicle node decrypts the global model parameters and performs multiple rounds of reverse correction training on them; the reverse corrected model parameters are encrypted and uploaded to the RSU node; the RSU node calculates the rate of change of the model parameters and sends the reverse corrected model parameters to the non-target vehicle nodes in the same group as the target vehicle node; the non-target vehicle nodes perform forward correction training on the reverse corrected model parameters; in each round of forward correction training, the non-target vehicle nodes in the group perform homomorphic encryption on the correction parameter updates, and exchange encrypted correction updates with adjacent non-target vehicle nodes; other non-target vehicle nodes aggregate the received encrypted correction updates with their own correction updates for the next round of forward correction training; finally, all non-target vehicle nodes encrypt the correction updates and send them to the RSU nodes to which the group belongs for edge aggregation to obtain the SL forgetting model; The global model update module is configured as follows: when the rate of change of the model parameters calculated by the RSU node is less than the set threshold, the target vehicle node exits the federation system, the correction training ends, the RSU node uploads the SL forgotten model parameters to the server, and the server updates the global model by aggregating the SL forgotten model parameters.

[0014] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0016] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0017] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a method and system for federated forgetting learning in the Internet of Vehicles based on group learning and homomorphic encryption. Compared with the existing technology, the present invention performs edge aggregation at the RSU node, which effectively reduces the communication and computing pressure; by introducing the group learning mechanism, multiple vehicle nodes can work together to further optimize the efficiency of forgetting learning and ensure the stability of model performance; a threshold homomorphic encryption algorithm is used to protect model parameters, prevent unauthorized access, effectively protect model updates, and prevent model reversal attacks and attacks from untrusted servers.

[0018] The present disclosure proposes a federated forgetting learning method and system for the Internet of Vehicles based on group learning and homomorphic encryption. Compared with the existing technology that does not use blockchain, the present disclosure uses blockchain point-to-point communication to reconstruct the decryption private key, and enables group learning to accelerate the forgetting process; it realizes decentralized and secure federated learning, and ultimately ensures the secure on-chain storage of global models, historical updates and reputation values, providing a transparent and tamper-proof chain of evidence, laying the foundation for the sustainable development of the Internet of Vehicles.

[0019] The present disclosure proposes a method and system for federated forgetting learning in the Internet of Vehicles based on group learning and homomorphic encryption. Compared with the prior art that does not use the homomorphic encryption algorithm, the present disclosure utilizes the characteristics of the homomorphic encryption algorithm to achieve secure aggregation of encrypted local updates during the transmission of local updates, effectively preventing model inversion attacks; and utilizes the secret sharing algorithm to achieve private key splitting, ensuring data privacy and model security, and effectively guaranteeing the robustness and stability of the federated learning process.

[0020] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0022] Figure 1 A framework for a vehicle network federation forgetting learning method based on group learning and homomorphic encryption provided in Example 1 of the present invention Figure 1 ; Figure 2 A framework for a vehicle network federation forgetting learning method based on group learning and homomorphic encryption provided in Example 1 of the present invention Figure 2 ; Figure 3 A flowchart of a vehicle network federated forgetting learning method based on group learning and homomorphic encryption provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0024] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0025] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" 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.

[0026] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0027] Terminology explanation: 1. Blockchain: Blockchain is a distributed ledger technology that uses a series of encryption algorithms and consensus mechanisms to connect data in chronological order in the form of blocks to form an unalterable record chain.

[0028] 2. Federated Learning: Federated Learning (FL) is a decentralized machine learning method that allows multiple parties to collaborate to train a shared model without exposing the original data.

[0029] 3. Internet of Vehicles: Internet of Vehicles (IoV) is a comprehensive system that uses information and communication technologies to promote the interconnection between vehicles and between vehicles and the external environment. Its core concept is to achieve real-time interconnection between vehicles and the outside world through information and communication technology (ICT), thereby improving traffic efficiency, reducing accident rates, optimizing energy consumption, and providing a more intelligent driving experience.

[0030] 4. Federated Unlearning: Machine Unlearning (MU) aims to effectively remove specific training samples and their influence from the trained model while maintaining the performance of the model as much as possible. With the deployment of Federated Learning (FL) in the real world, it allows multiple parties to jointly train a shared model without sharing the original data. However, since the model parameters or gradient information are still the relational mapping of the original data set, there are still security issues such as privacy leakage. Therefore, it is necessary to extend MU to Federated Unlearning (FU) to protect the privacy of the client.

[0031] 5. Homomorphic encryption: Homomorphic encryption (HE) is a special encryption technology that allows calculations to be performed directly on ciphertext without decrypting the data, and ensures that the result obtained is consistent with the plaintext calculation result. The advantages of homomorphic encryption are small communication volume, no need for multiple rounds of communication, secure calculation process, and no leakage of the privacy data of the result party.

[0032] Example 1 like Figure 1-3 As shown, this embodiment provides a vehicle network federation forgetting learning method based on group learning and homomorphic encryption, including the following steps: S1 builds a federated learning model for Internet of Vehicles and trains the model; S2 When the target vehicle node issues a forget request, the RSU node to which it belongs downloads the encrypted global model parameters from the aggregation server and sends them to the target vehicle node. The target vehicle node decrypts the global model parameters and performs multiple rounds of reverse correction training on them. The reverse corrected model parameters are encrypted and uploaded to the RSU node. The RSU node calculates the rate of change of the model parameters and sends the reverse corrected model parameters to the non-target vehicle nodes in the same group as the target vehicle node. The non-target vehicle nodes perform forward correction training on the reverse corrected model parameters. In each round of forward correction training, the non-target vehicle nodes in the group perform homomorphic encryption on the correction parameter updates, and exchange encrypted correction updates with adjacent non-target vehicle nodes. Other non-target vehicle nodes aggregate the received encrypted correction updates with their own correction updates for the next round of forward correction training. Finally, all non-target vehicle nodes encrypt the correction updates and send them to the RSU nodes to which the group belongs for edge aggregation to obtain the SL forgetting model. S3 When the rate of change of the model parameters calculated by the RSU node is less than the set threshold, the target vehicle node exits the federation system, the correction training ends, and the RSU node uploads the SL forgotten model parameters to the server, and the server updates the global model by aggregating the SL forgotten model parameters.

[0033] First, it is necessary to build a federated learning model of the Internet of Vehicles, which includes a server, an RSU node and a vehicle node. The lower layer of the server is provided with multiple RSU nodes, and each RSU node has multiple vehicle nodes divided into different groups in its area.

[0034] The specific process of building a federated learning model for Internet of Vehicles is as follows: 1. Identity registration stage: The smart contract automatically executes the registration process of all RSU nodes and vehicle nodes, and assigns an initial reputation value to each node. The information of successfully registered nodes is stored on the chain.

[0035] 2. Task release phase: The task publisher initializes a machine learning model and uploads it to the aggregation server. After the smart contract interacts with the task publisher to complete the order confirmation, it will be broadcast through the RSU node to all registered vehicle nodes that meet the task requirements to respond to the task.

[0036] 3. Participating node selection phase: RSU nodes and vehicle nodes that wish to participate in the federated learning task send response messages to the aggregation server. The smart contract obtains the recent reputation value of the responding node from the blockchain, and the task publisher selects the node with a higher reputation value to participate in the task.

[0037] 4. Group division: Integrate group learning (SL) into federated forgetting learning (FU) to build a two-layer federated forgetting learning architecture. RSU nodes divide the vehicle nodes that successfully join the training task in their area into p different groups according to the federated learning task. The vehicle nodes in the group need to be connected to the blockchain to achieve secure peer-to-peer communication.

[0038] The constructed Internet of Vehicles federated learning model is trained as follows: Model distribution stage: The RSU node downloads the encrypted global model parameters of the current round from the server and then sends them to the vehicle nodes of each group in its area.

[0039] Privacy protection training phase: To decrypt the global model, the nodes participating in the federated learning task jointly execute the key generation protocol to generate public keys and private key shares. The vehicle nodes interact point-to-point through the blockchain, exchange private key shares, and reconstruct the private key using Lagrange interpolation, which is then used to decrypt the global model parameters.

[0040] Based on the federated learning protocol, the vehicle nodes in the group perform local training and generate local updates according to the issued global model (the global model can be public during initialization, and the subsequent global model is encrypted. The vehicle nodes need to exchange private key shares through blockchain point-to-point and reconstruct the aggregation structure through Lagrange interpolation, and finally use the private key to decrypt the global model).

[0041] Threshold homomorphic encryption: the vehicle node calls Encrypt(m, PK) to encrypt the local update and uploads the encrypted local update to the RSU node to which the group belongs, together with indicators such as the size of the local data, computing speed, and training time, as a basis for evaluating the reputation value.

[0042] Edge aggregation, RSU nodes perform edge aggregation on the encrypted local updates uploaded by each vehicle node. In the edge aggregation process, RSU nodes are used as participants in the edge aggregation process. After receiving the encrypted local updates uploaded by the SL group, the RSU node first verifies whether the identity and timestamp of the vehicle node are valid. After verification, the encrypted local updates are edge aggregated to generate the SL model. The RSU node retains the historical updates submitted by the SL group in each edge aggregation round. The RSU node submits the SL model parameters and aggregation weights aggregated by the edge to the aggregation server for global model aggregation.

[0043] Global aggregation: The aggregation server retains the historical model parameters submitted by each RSU node, and updates the global model through the aggregation weights of each RSU node. Finally, the global model is generated through intra-group aggregation and inter-group aggregation to achieve hierarchical federated learning. The updated global model is then sent to each RSU node for the next round of training. Until the global model converges, a trained IoV federated learning model is obtained. The blockchain is used to store encrypted model parameters and the reputation values ​​of participating nodes to ensure the security and traceability of model updates.

[0044] Forgetting learning stage: Forgetting learning includes active forgetting learning and passive forgetting learning. Among them, the specific process of active forgetting learning is as follows: The target vehicle node sends a forget request, assuming that i During the round of training, the target vehicle node is received When a forget request is made, the ultimate goal is to exclude the data influence of this node from the global model.

[0045] In the reverse correction training phase, the blockchain first verifies the identity of the vehicle node that initiates the forgetting request. After verification, the RSU node sends the newly downloaded encrypted global model parameters to the target vehicle node. The target vehicle node interacts point-to-point through the blockchain, exchanges private key shares, and reconstructs the private key using Lagrange interpolation. Finally, the private key is used to decrypt the global model.

[0046] The target vehicle node adjusts the local update size and update direction through the gradient ascent algorithm locally, gradually eliminating the impact of local data on the global model. The target vehicle node exchanges the modified encrypted model parameters to the RSU node once every n rounds of reverse correction training.

[0047] The RSU node calculates the rate of change of the model parameters, and the SL corrects the rate of change of the model parameters as follows:

[0048] in, is the parameter of the SL correction model in round t, is the rate of change of the parameter, when Less than the set threshold When , the model is considered to have converged, and the reverse correction training is stopped. The SL model after reverse correction training is expressed as:

[0049] in, represents the modified SL model, represents the revised update of the target vehicle node, represents the learning rate, which is a hyperparameter that controls the magnitude of each update. Represents the loss function L relative to the correction update gradient.

[0050] In the forward correction training phase, the RSU node sends the reverse correction model parameters to the non-target vehicle nodes in the same group as the target vehicle node. The non-target vehicle nodes Perform n rounds of forward correction training. In each round of forward correction training, each non-target vehicle node Will calculate the correction update , after encryption, it interacts with the adjacent non-target vehicle nodes. Other non-target vehicle nodes aggregate the received encrypted correction updates with their own correction updates for the next round of forward correction training. Finally, all non-target vehicle nodes encrypt the correction updates and send them to the RSU nodes to which the SL group belongs for edge aggregation to obtain the SL forgetting model Aggregate correction update, RSU nodes aggregate correction updates from non-target vehicle nodes To update the SL forgetting model, the aggregated results of the correction update are as follows:

[0051] in, is the weight of the non-target vehicle node.

[0052] SL forgetting model update,By aggregating the edges of the correction update, the SL forgetting model can be updated as:

[0053] Each time the target vehicle node interacts with the RSU node to correct the model parameters, forward correction training is required until the change rate of the SL correction model parameters is less than the set threshold. , the target vehicle node exits the federated system. After the correction training is completed, the RSU node uploads the SL forgetting model parameters to the server, and the server updates the global model by aggregating the SL forgetting model parameters.

[0054] The specific process of passive forgetting is as follows: Before the RSU node performs edge aggregation on the encrypted local updates uploaded by each vehicle node, the cosine similarity between the encrypted local updates submitted by each vehicle node and the SL model is calculated. When the result deviates significantly, it is regarded as a malicious node. The contribution of the malicious node is removed based on the historical global model, and k rounds of retraining are performed. The new local update is encrypted and submitted to the RSU node, and the remaining node weights are bridged for edge aggregation to offset the impact of malicious nodes on the global model.

[0055] Reconstruction phase: The server updates the global model by aggregating the SL forgetting model parameters and updates the global FL model to:

[0056] in, is the current global model Based on the revised global model.

[0057] The aggregation server, target vehicle nodes and non-target vehicle nodes repeat the above process together, and finally obtain the forgotten global model , the model has eliminated the target vehicle node The impact of data.

[0058] The present invention utilizes group learning technology, and the vehicle nodes in each group can share and jointly train the corresponding SL model to improve the adaptability and accuracy of the model. Hierarchical federated learning is realized through intra-group aggregation and inter-group aggregation to reduce communication overhead. The homomorphic encryption algorithm is used to perform model aggregation based on the ciphertext domain of the model parameters of the vehicle nodes during the training process, providing a strong privacy guarantee for FL and avoiding inference attacks by malicious RSU nodes. Before aggregation, the model parameters are screened by calculating the cosine similarity between the local model parameters of each vehicle node and the SL model parameters to prevent model poisoning attacks. During each training process, the task publisher performs accurate and effective reputation evaluation on the RSU nodes and vehicle nodes according to relevant indicators, and stores the reputation values ​​securely on the chain, ultimately realizing a safe and efficient federated forgetting process.

[0059] Example 2 This embodiment provides a federated forgetting learning system for Internet of Vehicles based on group learning and homomorphic encryption, including: The model building and training module is configured to: build a federated learning model for Internet of Vehicles and train the model; The model correction training module is configured as follows: when the target vehicle node sends a forget request, the RSU node to which it belongs downloads the encrypted global model parameters from the aggregation server and sends them to the target vehicle node; the target vehicle node decrypts the global model parameters and performs multiple rounds of reverse correction training on them; the reverse corrected model parameters are encrypted and uploaded to the RSU node; the RSU node calculates the rate of change of the model parameters and sends the reverse corrected model parameters to the non-target vehicle nodes in the same group as the target vehicle node; the non-target vehicle nodes perform forward correction training on the reverse corrected model parameters; in each round of forward correction training, the non-target vehicle nodes in the group perform homomorphic encryption on the correction parameter updates, and exchange encrypted correction updates with adjacent non-target vehicle nodes; other non-target vehicle nodes aggregate the received encrypted correction updates with their own correction updates for the next round of forward correction training; finally, all non-target vehicle nodes encrypt the correction updates and send them to the RSU nodes to which the group belongs for edge aggregation to obtain the SL forgetting model; The global model update module is configured as follows: when the rate of change of the model parameters calculated by the RSU node is less than the set threshold, the target vehicle node exits the federation system, the correction training ends, the RSU node uploads the SL forgotten model parameters to the server, and the server updates the global model by aggregating the SL forgotten model parameters.

[0060] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0061] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0062] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0063] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0064] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0065] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0066] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0067] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0068] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0069] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, and the like. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, and the like.

[0070] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0071] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A federated forgetting learning method for Internet of Vehicles based on group learning and homomorphic encryption, characterized in that: The following steps are involved: Build a federated learning model for Internet of Vehicles and train the model; When the target vehicle node issues a forget request, the RSU node to which it belongs downloads the encrypted global model parameters from the aggregation server and sends them to the target vehicle node. The target vehicle node decrypts the global model parameters and performs multiple rounds of reverse correction training on them. The reverse-corrected model parameters are encrypted and uploaded to the RSU node. The RSU node calculates the rate of change of the model parameters and sends the reverse-corrected model parameters to the non-target vehicle nodes in the same group as the target vehicle node. The non-target vehicle nodes perform forward correction training on the reverse-corrected model parameters. In each round of forward correction training, the non-target vehicle nodes in the group perform homomorphic encryption on the correction parameter updates, and exchange encrypted correction updates with adjacent non-target vehicle nodes. Other non-target vehicle nodes aggregate the received encrypted correction updates with their own correction updates for the next round of forward correction training. Finally, all non-target vehicle nodes encrypt the correction updates and send them to the RSU nodes to which the group belongs for edge aggregation to obtain the SL forgetting model. When the rate of change of the model parameters calculated by the RSU node is less than the set threshold, the target vehicle node exits the federation system, the correction training ends, and the RSU node uploads the SL forgotten model parameters to the server, which updates the global model by aggregating the SL forgotten model parameters.

2. The vehicle network federation forgetting learning method based on group learning and homomorphic encryption as claimed in claim 1, characterized in that: The Internet of Vehicles federated learning model includes a server, an RSU node and a vehicle node, wherein a plurality of RSU nodes are arranged at the lower layer of the server, and a plurality of vehicle nodes divided into different groups are arranged in the area of ​​each RSU node.

3. The vehicle network federation forgetting learning method based on group learning and homomorphic encryption as claimed in claim 1, characterized in that: Train the Internet of Vehicles federated learning model, specifically: The RSU node downloads the encrypted global model parameters of the current round from the server, and then sends them to the vehicle nodes of each group in its area. The vehicle node decrypts the global model parameters and adjusts them based on local data to generate local updates. The local updates are then encrypted and uploaded to the RSU node to which the group belongs. The RSU node performs edge aggregation on the encrypted local updates uploaded by each vehicle node, generates an SL model, and submits the SL model parameters and aggregation weights to the server. The server updates the global model through the aggregation weights of each RSU node, and sends the updated global model to each RSU node for the next round of training until the global model converges to obtain a trained Internet of Vehicles federated learning model.

4. The vehicle network federation forgetting learning method based on group learning and homomorphic encryption as claimed in claim 3 is characterized in that: It also includes passively forgetting the vehicle node, and the passive forgetting is: During the training of the federated learning model of the Internet of Vehicles, before the RSU node performs edge aggregation on the encrypted local updates uploaded by each vehicle node, it will calculate the cosine similarity between the encrypted local updates submitted by each vehicle node and the SL model parameters. When the result deviates significantly, it is regarded as a malicious node. The contribution of the malicious node is removed based on the historical global model, and k rounds of retraining are performed. The new local updates are encrypted and submitted to the RSU node to bridge the remaining node weights for edge aggregation.

5. The vehicle network federation forgetting learning method based on group learning and homomorphic encryption as claimed in claim 1, characterized in that: Construct a federated learning model for Internet of Vehicles, specifically: Register RSU nodes and vehicle nodes, assign initial reputation values ​​to each node, and store the successfully registered node information on the blockchain; Initialize a machine learning model and broadcast it to all registered vehicle nodes that meet the task requirements through the RSU node to respond to the task; RSU nodes and vehicle nodes that wish to participate in the federated learning task send response messages to the blockchain and select nodes with higher reputation values ​​to participate in the task; The RSU node divides the vehicle nodes that have successfully joined the training task in its area into multiple different groups based on data distribution, and all vehicle nodes in the group are connected to the blockchain.

6. The vehicle network federation forgetting learning method based on group learning and homomorphic encryption as claimed in claim 1, characterized in that: The vehicle node decrypts the global model parameters, specifically: The nodes participating in the federated learning task jointly execute the key generation protocol to generate public keys and private key shares. The vehicle nodes interact point-to-point through the blockchain, exchange private key shares, and reconstruct the private key using Lagrange interpolation, which is used to decrypt the global model parameters.

7. The Internet of Vehicles federated forgetting learning system based on group learning and homomorphic encryption is characterized by: include: The model building and training module is configured to: build a federated learning model for Internet of Vehicles and train the model; The model correction training module is configured as follows: when the target vehicle node sends a forget request, the RSU node to which it belongs downloads the encrypted global model parameters from the aggregation server and sends them to the target vehicle node; the target vehicle node decrypts the global model parameters and performs multiple rounds of reverse correction training on them; the reverse corrected model parameters are encrypted and uploaded to the RSU node; the RSU node calculates the rate of change of the model parameters and sends the reverse corrected model parameters to the non-target vehicle nodes in the same group as the target vehicle node; the non-target vehicle nodes perform forward correction training on the reverse corrected model parameters; in each round of forward correction training, the non-target vehicle nodes in the group perform homomorphic encryption on the correction parameter updates, and exchange encrypted correction updates with adjacent non-target vehicle nodes; other non-target vehicle nodes aggregate the received encrypted correction updates with their own correction updates for the next round of forward correction training; finally, all non-target vehicle nodes encrypt the correction updates and send them to the RSU nodes to which the group belongs for edge aggregation to obtain the SL forgetting model; The global model update module is configured as follows: when the rate of change of the model parameters calculated by the RSU node is less than the set threshold, the target vehicle node exits the federation system, the correction training ends, the RSU node uploads the SL forgotten model parameters to the server, and the server updates the global model by aggregating the SL forgotten model parameters.

8. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images