Internet of vehicles verifiable federal learning method based on block chain

Abstract

The invention discloses an Internet of Vehicles verifiable federated learning method based on a block chain. The method comprises the following steps: realizing decentration by using a block chain storage transaction; the local gradient privacy is protected by adopting an AES encryption algorithm; the transaction is signed by using an ECDSA signature algorithm; and designing a time-sensitive right proof consensus mechanism based on the Newton's cooling law to improve the training enthusiasm of the participants. Through the steps of system initialization, worker training encryption, transaction signature, verifier voting, miner processing and the like, a safer and more efficient federated learning process is realized, the global model precision is improved, powerful support is provided for intelligent application of the 6G Internet of Vehicles, the defects of traditional federated learning are overcome, and the method has important application value.

Description

Technical Field

[0001] The present invention belongs to the field of secure machine learning applications, and involves technologies such as encryption algorithms and consensus mechanisms, and in particular, a verifiable federated learning method for Internet of Vehicles based on blockchain. Background Art

[0002] In the rapidly developing 6G In the field of Internet of Vehicles, federated learning has become a highly regarded machine learning framework due to its unique advantages. With the continuous evolution of intelligent transportation systems, the amount of data generated by vehicles has exploded. These data contain huge value, but the traditional centralized data processing method faces many challenges. Federated learning came into being. It breaks the shackles of the traditional model and provides a decentralized collaborative model training method, allowing all participants to jointly train a more powerful model without leaking local data, greatly protecting data privacy.

[0003] With the help of 6G Technical excellence in performance, federated learning 6G It will add wings to the Internet of Vehicles ecosystem. 6G The ultra-low latency feature of the federated learning model makes data transmission in the model training process almost instantaneous, avoiding the problem of low training efficiency caused by latency, and enabling real-time model updates to adapt to changing road conditions and driving scenarios in a timely manner. High reliability ensures the accuracy and stability of data transmission, reduces the risk of data loss or errors, and provides a reliable data foundation for model training. Large-scale connectivity allows more vehicles, roadside units and other equipment to participate in federated learning, gathering a wider range of data, thereby improving the generalization and accuracy of the model and achieving more efficient data sharing.

[0004] However, federated learning is not always smooth sailing in practical applications and faces many severe challenges. The single point failure problem is one of the key hidden dangers. In the federated learning architecture, if a key node (such as a central server or an important participant) fails, the entire learning process may be interrupted, affecting the training and updating of the model. This not only reduces the reliability of the system, but may also cause data loss or inconsistency, bringing great risks to subsequent applications.

[0005] The potential privacy leakage of data providers should not be ignored either. Although federated learning aims to protect data privacy, there is still a risk of information leakage during data interaction and model training. For example, some malicious participants may reversely infer the sensitive data of other participants by analyzing the gradient information during the training process. Existing encryption technologies can protect data to a certain extent, but with the continuous escalation of attack methods, the security of these technologies faces new challenges. For example, some privacy protection schemes based on homomorphic encryption, although they can perform calculations on encrypted data, have low computational efficiency and are difficult to fully guarantee the privacy security of data when faced with complex attack models. These problems limit the application of federated learning in 6G The widespread application of Internet of Vehicles urgently requires innovative technologies and methods to solve it. Summary of the invention

[0006] The purpose of the present invention is to provide a verifiable federated learning method for Internet of Vehicles based on blockchain to overcome the shortcomings of the prior art.

[0007] According to the first main aspect of the present invention, a verifiable federated learning method for Internet of Vehicles based on blockchain is provided, comprising: utilizing blockchain to store transactions to achieve decentralization; and use AES Encryption algorithms protect local gradient privacy; and use ECDSA The signature algorithm signs the transaction; and A time-sensitive proof-of-stake consensus mechanism is designed based on Newton's law of cooling to enhance participants' enthusiasm for training.

[0008] Among them, the AES The encryption algorithm uses symmetric key Encrypt the local model gradient, ECDSA The algorithm uses a private key Sign the transaction with the public key Verify the signature.

[0009] Specifically, the blockchain-based Internet of Vehicles verifiable federated learning method includes at least the following steps: S1, System initialization: Define roadside units RSU Three roles of workers: , the verifier ,miner , and generate AES Symmetric key of the algorithm as well as ECDSA The public key of the algorithm and private key ; S2, for workers Local Model Train and encrypt the encrypted local model gradient and workers obtained by the reward mechanism based on Newton's law of cooling Expected Reward Encapsulated in a worker Transactions middle; S3, Worker use Algorithms for Workers Sign the transaction and sign the worker Transactions Send the specified validator ; S4, Validator To the signed worker Transactions Verify the signature and obtain the encrypted worker local model gradient , decrypt to get the worker local model gradient and vote; validators The workers Transactions , Validator For workers Voting results , Validator Verification Rewards and validators Voting rewards Encapsulated in the verifier Transactions middle; S5, Validator use Algorithm for Validators Affairs Sign and verify the signature Transactions Send the specified miner ; S6, Miner For validators Signed Transactions Verify the signature and obtain the verifier For workers Voting results , miners combine all voting results and record them as: , and retrieve the workers Encrypted Encrypted local model gradients , decrypt and aggregate effective local gradients.

[0010] As a further preferred solution, during the system initialization, the roadside unit RSU The role in each round of training cycle Dynamic allocation within the network, and the global model is stored on the blockchain For workers RSU Download and train the update.

[0011] As a further preferred solution, the verifier The voting process includes: broadcasting the signed transaction to other validators RSU , after verifying the validity of the signature, the gradient is decrypted and a majority vote is performed based on the quality of the gradient. The voting reward Positively correlated with the accuracy of voting results.

[0012] As a further preferred solution, the miner The steps of aggregating valid gradients include: screening local gradients whose voting results exceed the preset threshold, decrypting and weighted averaging to generate a global model, and writing the updated global model into the blockchain.

[0013] As a further preferred embodiment, the decay rate of Newton's cooling law is k Dynamically adjust according to network load to ensure that the reward mechanism adapts to real-time changes in the Internet of Vehicles environment.

[0014] Based on the second main aspect of the present invention, a blockchain-based Internet of Vehicles verifiable federated learning device for the aforementioned method is provided, comprising: Initialization module, used to initialize workers , the verifier ,miner ,as well as AES Symmetric key of the algorithm , ECDSA The public key of the algorithm and private key ; Training the encryption module is configured to make the worker Get the last round of global model for local model Training, adoption AES The algorithm encrypts the local model gradient and calculates the expected reward based on the reward mechanism of Newton's cooling law , and encapsulate the encrypted local model gradients and expected rewards into workers Transactions ; Signature sending module for workers use ECDSA Algorithms for Workers Transactions Sign and send the signed transaction to the designated verifier ; Verification voting module, validator This module verifies the signature of the worker's transaction. If it passes, the encrypted local model gradient is obtained and decrypted, and the local model gradient is voted on. Transactions , Voting Results , Verification Rewards and voting rewards Encapsulated in the verifier Transactions middle; Miner processing module, miner With the help of this module, transactions signed by validators are verified, voting results of all validators are merged, local model gradients encrypted by workers are retrieved and decrypted, and valid local gradients are aggregated.

[0015] Based on the third main aspect of the present invention, an electronic device is provided, comprising: at least one processor; a memory communicatively connected to the at least one processor; the memory storing a computer program which, when executed by the at least one processor, enables the at least one processor to implement the aforementioned blockchain-based verifiable federated learning method for Internet of Vehicles.

[0016] Based on the fourth main aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed, the verifiable federated learning method for the Internet of Vehicles based on blockchain as described above is implemented.

[0017] Compared with the existing technology, this paper designs a blockchain-based verifiable federated learning method for Internet of Vehicles. This method introduces blockchain technology and solves the decentralization problem by storing transactions on the chain. In order to protect the privacy of local gradients, AES The encryption algorithm encrypts the gradient, using ECDSA The signature algorithm signs the transaction and uses Newton's cooling law to design a time-sensitive proof-of-stake consensus mechanism to increase the training enthusiasm of participants. 6G Difficulties in Internet of Vehicles applications.

[0018] In terms of improving safety, the present invention has significant results. AES The encryption algorithm encrypts the local model gradient, greatly enhancing the confidentiality of data during transmission and storage. 6G In the Internet of Vehicles environment, data is frequently interacted and the risk of malicious attacks is high. AESEncryption makes it difficult for sensitive local gradient data to be stolen and cracked, effectively resisting potential data leakage threats and providing a solid guarantee for data privacy. ECDSA The signature algorithm signs the transaction, ensuring the non-repudiation and integrity of the transaction. Each transaction is strictly signed and verified, making it extremely difficult to illegally tamper with the transaction information, ensuring the credibility of data interaction in the system, and building a safe and reliable learning environment from multiple dimensions, which is difficult for traditional federated learning to achieve.

[0019] In solving the decentralization problem, the present invention innovatively introduces blockchain technology to store transactions, successfully eliminating the hidden danger of single point failure. In the traditional federated learning architecture, once the central server fails, the entire training process will be interrupted. However, in the decentralized architecture constructed by the present invention, each node cooperates with each other, and there is no single core failure point. Even if some nodes have problems, the system can still rely on other nodes to continue running, ensuring the continuity of model training and updating, significantly improving the stability and reliability of the system. This feature is very important for ensuring 6G The uninterrupted operation of connected vehicle services is crucial.

[0020] The time-sensitive proof-of-stake consensus mechanism designed by the present invention based on Newton's cooling law effectively stimulates the enthusiasm of participants. Through this mechanism, the rewards of participants are linked to the training time. The more timely the training, the higher the reward. 6G In the dynamic environment of the Internet of Vehicles, this incentive encourages vehicles and roadside units to actively participate in model training, speeds up data collection and model updates, and thus improves the accuracy of the global model. Compared with the traditional fixed reward model, this mechanism enables the model to adapt to dynamic changes such as road conditions and driving behavior more quickly, improves the accuracy and real-time performance of Internet of Vehicles applications (such as intelligent driving assistance and traffic flow prediction), and provides a reference for the development of the Internet of Vehicles. 6G It provides strong support for the efficient operation of the Internet of Vehicles.

[0021] In terms of improving model accuracy, the present invention ensures the effectiveness of model training through a multi-role collaborative verification and aggregation mechanism. Workers, validators, and miners collaborate with each other. Validators vote to verify the local model gradients of workers, and miners aggregate effective gradients. This layer-by-layer approach avoids the interference of bad data on the model, ensuring that only high-quality data participates in model updates, significantly improving the accuracy and generalization ability of the global model. 6G Various applications in the Internet of Vehicles provide more accurate model support and promote the development of intelligent transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The figure shows a comparison between the method of the present invention and other methods under different conditions in one embodiment of the present invention.

[0023] Figure 2 The figure shows a reward comparison of the reward mechanism proposed by the method of the present invention in one embodiment of the present invention. DETAILED DESCRIPTION

[0024] The preferred embodiments of the present invention will be described in detail below so that the purpose, features and advantages of the present invention can be more clearly understood. It should be understood that the following embodiments are not intended to limit the scope of the present invention, but are only intended to illustrate the essential spirit of the technical solution of the present invention.

[0025] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the relevant art will recognize that the embodiments may be practiced without one or more of these specific details. In other cases, well-known techniques associated with the present application may not be shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0026] References throughout the specification to "one embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of "in one embodiment" or "in an embodiment" in various places throughout the specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any manner in one or more embodiments.

[0027] The present invention provides a verifiable federated learning method for Internet of Vehicles based on blockchain, and the specific steps are as follows: Step 1: System initialization: The system architecture consists of a network of interconnected IoT devices, which are set up as . Roadside Unit Collect traffic flow information and perform multiple functional roles, including worker , Validator and miners , in each round Roles are assigned dynamically at runtime. As The symmetric key of the algorithm, and and Respectively represent The public and private keys of the algorithm.

[0028] Step 2: Workers Training and encrypting local models: Worker From the previous round Obtaining the global model during training , and store it on the blockchain.

[0029] Then, each worker Use your own dataset Train the local model, denoted as After training, each worker use The algorithm encrypts its local model gradient to ensure the security of data transmission. Encrypted local model gradients Send to Verification .

[0030] After this transfer, the validator Participate in a mutual broadcasting protocol through which each validator Obtained by respective workers Encrypted local model gradients for training In addition, workers Rewarding workers based on Newton's law of cooling Expected Reward .

[0031] Among them, the expected reward Includes basic rewards and time reward .

[0032] For time reward , using Newton's cooling law to design: ; in, is the maximum reward when training is completed, It is the minimum basic reward when training is completed. To control the rate at which rewards are reduced. is the time elapsed after the model has been trained.

[0033] Afterwards, the workers The encrypted local model gradient and workers award Encapsulated in a worker Transactions middle.

[0034] Step 3: Sign Worker Transactions: Worker use Algorithms for Workers Affairs Sign as follows: .

[0035] For worker affairs after signing, For private key, worker The signed worker transaction Send the specified validator .

[0036] Step 4: Validators vote: Each validator From all relevant validators Get Worker Signed Transactions , and the signed worker transaction Broadcast to all other validators Through this process, each validator Worker transaction with signature .

[0037] Then, the verifier First verify the worker Signed Transactions If the worker Transactions If the verification is successful, the verifier Due to successful verification of worker transactions The verification reward is obtained by signing .

[0038] Then the validator From Worker Affairs Extract workers from Encrypted local model gradients , decrypt the encrypted local gradient get . Validator For workers The local model gradient Voting. Validators Receive voting rewards .

[0039] Then, the verifier The workers Transactions , Validator For workers Voting results , Validator Verification Rewards and validators Voting rewards Encapsulated in the verifier Transactions middle.

[0040] like Figure 1 and Figure 2 As shown, Figure 1 The comparison between the method of the present invention and other methods under different conditions is shown. Figure 2 The reward comparison of the reward mechanism proposed by the method of the present invention is demonstrated.

[0041] Figure 1 The upper left row shows the comparison of VFL_0 / 20 (in the VFL scheme, all devices are benign), VFL_3 / 20 (in the VFL scheme, 3 out of 20 devices are malicious), VBFL_PoS 3 / 20 (in the VBFL scheme, 3 out of 20 devices are malicious), and AVBFL-1_PoS 3 / 20 (in the AVBFL scheme, 3 out of 20 devices are malicious and use an encryption algorithm).

[0042] Figure 1 The upper right side shows the comparison of VFL_0 / 20 (in the VFL scheme, all devices are benign), VFL_3 / 20 (in the VFL scheme, 3 out of 20 devices are malicious), VBFL_PoS 3 / 20 (in the VBFL scheme, 3 out of 20 devices are malicious), and AVBFL-2_TPoS 3 / 20 (in the AVBFL scheme, 3 out of 20 devices are malicious and the consensus mechanism is used).

[0043] Figure 1 The lower left row shows a comparison of VFL_0 / 20 (in the VFL scheme, all devices are benign), VFL_3 / 20 (in the VFL scheme, 3 out of 20 devices are malicious), VBFL_PoS 3 / 20 (in the VBFL scheme, 3 out of 20 devices are malicious), and AVBFL-3_TPoS 3 / 20 (in the AVBFL scheme, 3 out of 20 devices are malicious, the encryption algorithm and consensus mechanism are used, and the maximum training epoch is selected).

[0044] Figure 1The upper right side shows the comparison of AVBFL-3_TPoS 3 / 20 (in the AVBFL scheme, 3 out of 20 devices are malicious, encryption algorithms and consensus mechanisms are used, and the maximum epoch is selected), AVBFL-3mid_TPoS 3 / 20 (in the AVBFL scheme, 3 out of 20 devices are malicious, encryption algorithms and consensus mechanisms are used, and the middle epoch is selected), and AVBFL-3min_TPoS 3 / 20 (in the AVBFL scheme, 3 out of 20 devices are malicious, encryption algorithms and consensus mechanisms are used, and the minimum epoch is selected).

[0045] exist Figure 1 In the paper, VFL is a federated learning model proposed by McMahan et al. in 2017. Each device performs local model training and parameter upload. The model is updated using a central server to ensure that user data is only processed on the local device to protect user privacy (B.McMahan,E.Moore, D. Ramage, S. Hampson, and BA y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Artificial intelligence and statistics. PMLR, 2017, pp. 1273–1282.).

[0046] VBFL is a blockchain-based decentralized federated learning framework proposed by Chen et al. in 2021. VBFL introduces a new decentralized verification mechanism and a dedicated PoS consensus mechanism to achieve more robust federated learning (H. Chen, S. A. Asif, J. Park, C.-C. Shen, and M. Bennis, “Robust blockchained federated learning with model validation and proof-of stake inspired consensus. 2021,” DOI: https: / / doi. org / 10.48550 / ARXIV, vol. 2101, 2021.).

[0047] AVBFL is a solution designed by the present invention.

[0048] The matters not described in detail in the present invention are all known technologies to those skilled in the art.

[0049] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A verifiable federated learning method for Internet of Vehicles based on blockchain, characterized in that: include: Decentralization is achieved by using blockchain to store transactions; And adopt AES The encryption algorithm protects the privacy of local gradients; And use ECDSA The signature algorithm signs the transaction; and designs a time-sensitive proof-of-stake consensus mechanism based on Newton’s cooling law to increase the enthusiasm of participants to train; Said AES The encryption algorithm uses symmetric key Encrypt the local model gradient, ECDSA The algorithm uses a private key Sign the transaction with the public key Conduct signature verification; The following steps are involved: S1, System initialization: Define roadside units RSU Three roles of workers: , the verifier ,miner , and generate AES Symmetric key of the algorithm as well as ECDSA The public key of the algorithm and private key ; S2, for workers Local Model Train and encrypt the encrypted local model gradient and workers obtained by the reward mechanism based on Newton's law of cooling Expected Reward Encapsulated in a worker Transactions middle; S3, Worker use Algorithms for Workers Sign the transaction and sign the worker Transactions Send the specified validator ; S4, Validator To the signed worker Transactions Verify the signature and obtain the encrypted worker local model gradient , decrypt to get the worker local model gradient and vote; validators The workers Transactions , Validator For workers Voting results , Validator Verification Rewards and validators Voting rewards Encapsulated in the verifier Transactions middle; S5, Validator use Algorithm for Validators Affairs Sign and verify the signature Transactions Send the specified miner ; S6, Miner For validators Signed Transactions Verify the signature and obtain the verifier For workers Voting results , miners combine all voting results and record them as: , and retrieve the workers Encrypted Encrypted local model gradients , decrypt and aggregate effective local gradients.

2. The blockchain-based Internet of Vehicles verifiable federated learning method according to claim 1, characterized in that: During the system initialization, the roadside unit RSU The role in each round of training cycle Dynamic allocation within the network, and the global model is stored on the blockchain For workers RSU Download and train the update.

3. The blockchain-based Internet of Vehicles verifiable federated learning method according to claim 1, characterized in that: The verifier The voting process includes: broadcasting the signed transaction to other validators RSU , after verifying the validity of the signature, the gradient is decrypted and a majority vote is performed based on the quality of the gradient. The voting reward Positively correlated with the accuracy of voting results.

4. The blockchain-based Internet of Vehicles verifiable federated learning method according to claim 1, characterized in that: The miners The steps of aggregating valid gradients include: screening local gradients whose voting results exceed the preset threshold, decrypting and weighted averaging to generate a global model, and writing the updated global model into the blockchain.

5. The blockchain-based Internet of Vehicles verifiable federated learning method according to claim 1, characterized in that: The decay rate of Newton's law of cooling k Dynamically adjust according to network load to ensure that the reward mechanism adapts to real-time changes in the Internet of Vehicles environment.

6. A blockchain-based Internet of Vehicles verifiable federated learning device for implementing any one of the methods in claims 1-5, characterized in that: include: Initialization module, used to initialize workers , the verifier ,miner ,as well as AES Symmetric key of the algorithm , ECDSA The public key of the algorithm and private key ; Training the encryption module is configured to make the worker Get the last round of global model for local model Training, adoption AES The algorithm encrypts the local model gradient and calculates the expected reward based on the reward mechanism of Newton's cooling law , and encapsulate the encrypted local model gradients and expected rewards into workers Transactions ; Signature sending module for workers use ECDSA Algorithms for Workers Transactions Sign and send the signed transaction to the designated verifier ; Verification voting module, validator This module verifies the signature of the worker's transaction. If it passes, the encrypted local model gradient is obtained and decrypted, and the local model gradient is voted on. Transactions , Voting Results , Verification Rewards and voting rewards Encapsulated in the verifier Transactions middle; Miner processing module, miner With the help of this module, transactions signed by validators are verified, voting results of all validators are merged, local model gradients encrypted by workers are retrieved and decrypted, and valid local gradients are aggregated.

7. An electronic device, characterized in that: include: at least one processor; a memory communicatively coupled to the at least one processor; The memory stores a computer program, which, when executed by the at least one processor, enables the at least one processor to implement the blockchain-based verifiable federated learning method for Internet of Vehicles as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed, the blockchain-based verifiable federated learning method for Internet of Vehicles as described in any one of claims 1-5 is implemented.

Images