Internet of vehicles privacy protection federal learning method and system
By employing a certificateless conditional authentication scheme based on SM2 and a federated learning method for privacy protection in vehicle networks, the problems of low efficiency and insufficient security in existing technologies are solved, achieving efficient privacy protection and secure vehicle authentication.
Patent Information
- Application Number
- CN202411065494.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing federated learning schemes for connected vehicles suffer from low efficiency, high computational overhead, and insufficient security in terms of privacy protection, making it difficult to effectively protect user privacy in connected vehicles.
A certificateless conditional privacy-preserving identity authentication scheme based on SM2 is adopted. Combining elliptic curve cryptography, the Chinese Remainder Theorem, and the SM2 digital signature algorithm, a privacy-preserving federated learning method for vehicle networking is designed. Through dynamic pseudonyms and group key management, vehicle identity authentication and secure transmission of model parameters are achieved.
It improves the efficiency and security of federated learning in vehicle networking, reduces complex encryption operations, supports efficient privacy protection and dynamic member management, and ensures the authenticity and integrity of model parameters.
Smart Images

Figure CN118945650A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of Internet of Vehicles data privacy protection, and more specifically to an Internet of Vehicles privacy protection federated learning method and system. Background Art
[0002] With the rapid development of intelligent transportation systems and Internet of Things technologies, the Internet of Vehicles has become an important part of smart cities. Through communication between vehicles and between vehicles and infrastructure, the Internet of Vehicles can achieve real-time sharing of traffic information and intelligent vehicle coordination. In addition, with the continuous advancement of machine learning technology, many automobile companies are applying machine learning to the Internet of Vehicles to provide more intelligent and efficient services. By collecting a large amount of vehicle data to train models, these companies are able to provide applications such as autonomous driving and traffic flow prediction. However, traditional centralized model training requires the collection of vehicle data to a central server for training. Since these vehicle data usually contain a lot of personal information, such as driving habits, driving routes, home and work locations, many users are worried about privacy leakage and are reluctant to send data to a central server. In addition, the latest data security regulations prohibit automobile companies from collecting user data without authorization. To address these privacy issues, federated learning came into being. Federated learning is a decentralized machine learning method in which multiple clients (such as smartphones, vehicles, or other devices) jointly train a shared model under the coordination of a central server, while the data remains locally. Instead of sending raw data to a central server, the client processes the data locally and only shares model updates (such as gradients or parameters) with the central server. The server then aggregates these updates to form a global model. Currently, federated learning has been widely used in various Internet of Vehicles scenarios, such as trajectory prediction, advanced driver assistance systems, and traffic flow prediction and management.
[0003] Although federated learning solves the data silo problem, researchers have found that attackers can still infer private information about user data if the transmitted model parameters are not properly protected. In addition, when the central server aggregates the parameters, the server may attempt to infer the original data information from the uploaded model parameters. Furthermore, due to the openness of the Internet of Vehicles, attackers can easily eavesdrop on and manipulate messages transmitted between vehicles, obtain the true identity of the vehicle and further track its behavior, threatening user privacy.
[0004] To solve the problem of privacy leakage in federated learning, current solutions are mainly divided into differential privacy and encryption technology. Differential privacy protects the privacy of original data by adding random noise to model parameters. Although differential privacy-based solutions can be extended to all machine learning algorithms in deep learning, the added random noise will reduce the accuracy of the model and prolong the convergence time of the model. Encryption-based solutions can be divided into homomorphic encryption and secure multi-party computation. Homomorphic encryption allows the central server to directly summarize the model parameters in ciphertext form, but the complex homomorphic operations lead to high computational overhead and cannot be applied to all machine learning algorithms. As an encryption scheme, secure multi-party computation enables distributed participants to collaboratively calculate the target function without revealing data. A typical example is secret sharing. Compared with homomorphic encryption, this method reduces the computational overhead, but increases the number of communication rounds and communication overhead, thereby hindering the training efficiency of federated learning. In addition, encryption-based solutions make it impossible for cloud servers to directly access plaintext local model parameters during the aggregation process. This hinders the integration with the Byzantine robust federated learning defense mechanism, because the existing Byzantine robust defense mechanism focuses on directly performing similarity calculations on plaintext model parameters.
[0005] Therefore, it is necessary to study a privacy-preserving federated learning solution for Internet of Vehicles that can strike a balance between efficiency and security. Summary of the invention
[0006] In view of this, the present invention provides a method and system for Internet of Vehicles privacy protection federated learning.
[0007] In order to achieve the above object, the present invention adopts the following technical solution:
[0008] The present invention first discloses a privacy-preserving federated learning method for Internet of Vehicles, comprising the following steps:
[0009] System initialization: The trusted center initializes the system and generates system public parameters;
[0010] Registration and group key generation: The trusted center performs vehicle and roadside unit registration tasks and generates secret keys for distribution to each vehicle. It also generates the relevant keys for communication between the trusted center, vehicles, and roadside units based on the secret keys and system public parameters.
[0011] Local training and model aggregation: The cloud server sends global model parameters to each vehicle in the coverage area. The vehicle trains the local data to obtain local model parameters, signs the local model parameters according to the relevant key to generate verification information and uploads it anonymously to the roadside unit; the roadside unit verifies the local model parameters according to the verification message, locally aggregates the verified local model parameters, and obtains a local aggregate model. The roadside unit then signs the local aggregate model with its own private key and uploads it to the cloud server. The cloud server verifies the local model uploaded by the roadside unit and globally aggregates the verified local models to obtain the global model parameters. At this point, one round of training iteration is completed, and multiple rounds of iterations are performed until the model converges.
[0012] The method further includes a group member management step: the trusted center updates the group key based on the global model parameters, system public parameters, and related keys to achieve dynamic management of the vehicle.
[0013] Furthermore, the trusted center performs system initialization and generates system public parameters, which specifically includes the following steps:
[0014] The trusted center uses the security parameter λ to generate two prime numbers p and q, where p>q.
[0015] Let E(F p ) represents a finite field F p Elliptic curve on, G represents the elliptic curve E(F p ) with order q;
[0016] Random selection of trusted centers As the system master private key, and calculate P pub =s·G as the system public key, and select five one-way and collision-resistant hash functions: And publish the system public parameters
[0017] Furthermore, the trusted center performs the registration task of the vehicle and the roadside unit and generates a secret key to distribute to each vehicle, and generates the relevant keys for communication between the trusted center, the vehicle and the roadside unit based on the secret key and the system public parameters, specifically including the following steps:
[0018] Vehicle Registration:
[0019] Each vehicle V i Generate a random secret value As the first part of the private key of the vehicle, and calculate the first part of the public key X of the vehicle based on the first part of the private key i =x i G;
[0020] The first part of the public key X i The real ID corresponding to the vehicle i Sent to a trusted center via a secure channel;
[0021] The trusted center receives (RID i ,X i ), calculate h i =H(X i ||P pub ), y i =s·h i and Y i =y i ·G,(Y i ,y i ) i and i As vehicle V i The second part of the public key and private key;
[0022] The trusted center then provides V i Randomly select a prime number as a secret key for subsequent group communications;
[0023] Send y through the safe channel i ,Y i and sk i Send to vehicle V i ;
[0024] Vehicle V i Received y i ,Y i and sk i Then, set its complete public key to (X i ,Y i ), the complete private key is (x i ,y i );
[0025] Roadside Unit Registration:
[0026] Roadside Unit RSU j Register their real identity ID with a trusted center RSUj Register and the trusted center will issue a pair of public and private keys to it
[0027] Group key generation:
[0028] The trusted center builds a communication group C based on the secret keys of n vehicles and the Chinese remainder theorem. n ,calculate a i =θ / sk i and b i =(ai ) -1 modsk i ;
[0029] Let c i =a i b i , Then randomly select a group key Calculate the group public key β = K u and D pub =K·G, T K is the validity period of the group key;
[0030] Then use the trusted center's own private key sk TA For the message {β,D pub ,T K}Sign and broadcast the message To communication group C n All vehicles and roadside units in the.
[0031] Furthermore, local training and model aggregation specifically include the following steps:
[0032] During the training of round t, the vehicles V in the communication group i Receive the global model sent by the cloud server Then, use the vehicle V i Local data for global model Train to get the local model W i t ,Right now
[0033] Using Vehicle V i The private key and group key pair of the local model W i t The signature is then sent to adjacent roadside units for aggregation.
[0034] Furthermore, using the vehicle V i The private key and group key pair of the local model W i t Signing and sending to nearby roadside units for aggregation specifically includes:
[0035] Signature: Vehicle V i Randomly generate a Generate pseudonymous PID i =(PID i,1 ,PID i,2 ), where PID i,1 =c i G,
[0036] calculate in Indicates PID i,2 The bit length is converted into two bytes, a and b are F p The elements in T i Indicates the current timestamp;
[0037] Calculate the group key K≡βmodsk i and the signing key
[0038] Random Selection Calculate K i =k i G = (x1, y1), e i =H5(Z i ||W i t |T i ), r i =e i +x1modq,s i =(1+sgk i ) -1 ·(k i -r i ·sgk i )modq, judge r i =0 or r i +k i =q? and s i =0? If it satisfies, choose a new k i Calculate and finally generate the signature σ i =(r i ,s i );
[0039] Vehicle delivery to a communicating roadside unit or other vehicle;
[0040] Local aggregation:
[0041] Roadside Unit RSU j Received After that, perform signature verification as follows:
[0042] First, the roadside unit RSU j Check timestamp T i Freshness: The arrival time is T a , if ΔT ≥ T a -T i , then continue, otherwise discard the message;
[0043] Then calculate e i =H5(Zi ||W i t ||T i ), h i =H(Y i ||X i ||P pub ), t i =r i +s i modq,
[0044] Judge R = e i +x1'=r i ? If yes, the signature is valid and the message comes from a legitimate vehicle, otherwise it is discarded;
[0045] Roadside Unit RSU j Use the FedAvg algorithm to aggregate the data from n vehicles {V1,V2,...,V n}Received local model parameters Get local aggregation results Where i∈[1,n]. Then RSU j Use your own private key pair Sign and send Perform global aggregation for cloud servers;
[0046] Global aggregation: The cloud server receives data from nearby roadside units. After that, first verify its effectiveness;
[0047] Then the local aggregation results that pass the verification Perform global aggregation to obtain the global model for the next round of training Where j∈[1,m], m represents the number of RSUs;
[0048] Finally, the cloud server signs the global model using its own private key and broadcasts the message to said roadside units and vehicles within the communication group;
[0049] At this point, one round of training is completed, and multiple rounds of iterations are performed until convergence.
[0050] Furthermore, group member management specifically includes the following steps:
[0051] Tracing
[0052] When the roadside unit RSU j When abnormal behavior of the vehicle is detected, it is reported to the trusted center, which uses the system master private key to restore the true identity of the vehicle based on the pseudonym.
[0053] Revocation
[0054] The trusted center obtains the real identity RID of the malicious vehicle i Then remove the malicious vehicle from the communication group C as follows n Cancel:
[0055] Calculate u' = uc i Delete legal information related to the vehicle;
[0056] Select a new group key Calculate the new group public key β' = K'·u', D' pub = K'·G, then broadcast the message to roadside units and other vehicles;
[0057] After receiving the message, the vehicles in the group use their own key sk j The new group key K'≡β'modsk can be obtained through a modulo operation j ;
[0058] Add to
[0059] When the vehicle V i Initiate to join communication group C n When a request is made, the trusted center adds it to the group as follows:
[0060] The trusted center selects a new group key Calculate θ' = θ·f i , a′ i =θ' / f i and b′ i =(a′ i ) - 1 modsk i , c′ i =a' i b′ i ,
[0061] Calculate the new group public key β' = K'·u', D' pub = K'·G, then broadcast the message to roadside units and other vehicles;
[0062] After receiving the message, the vehicles in the group use their own key sk i The new group key K'≡β'modf can be obtained through a modulo operation i .
[0063] Based on the above method, the present invention also discloses a privacy-preserving federated learning system for Internet of Vehicles, including a trusted center, a roadside unit, and a cloud server;
[0064] Wherein, the trusted center includes an initialization module and a registration and group key generation module;
[0065] The initialization module is used to initialize the system and generate system common parameters;
[0066] The registration and group key generation module is used to perform vehicle and roadside unit registration tasks and generate secret keys to distribute to each vehicle, and generate relevant keys for communication between the trusted center, vehicles and roadside units based on the secret keys and system public parameters;
[0067] The cloud server transmits the global model parameters to each vehicle in the coverage area, each vehicle trains the local data to obtain the local model parameters, signs the local model parameters according to the relevant keys to generate verification information and uploads it anonymously to the roadside unit; the roadside unit verifies the local model parameters according to the verification message, locally aggregates the verified local model parameters to obtain the local aggregate model; the roadside unit signs the local aggregate model with its own private key and uploads it to the cloud server; the cloud server verifies the local model uploaded by the roadside unit and globally aggregates the verified local models to obtain the global model parameters. At this point, one round of training iteration is completed, and multiple rounds of iterations are performed until the model converges.
[0068] Preferably, the above system further comprises a group member management module, which is located in the trusted center and is used to update the group key based on the global model parameters, system public parameters, and related keys and realize dynamic management of vehicles.
[0069] Preferably, training the local data specifically includes the following steps:
[0070] During the training of round t, the vehicles V in the communication group i Receive the global model sent by the cloud server Then, use the vehicle V i Local data for global model Train to get the local model W i t ,Right now
[0071] Using Vehicle V i The private key (x i ,y i ) and the group key K for the local model W i t Signed and sent to adjacent roadside units for aggregation, where xi For each vehicle V i The first part of the private key, specifically the vehicle V i Generate a random secret value y i For each vehicle V i The second part of the private key is determined by the trusted center based on the vehicle V i The corresponding real identity is calculated.
[0072] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and system for Internet of Vehicles privacy protection federated learning, which has the following beneficial effects:
[0073] The present invention designs a certificateless conditional privacy-preserving identity authentication scheme based on SM2 and combines it with federated learning to provide privacy protection for vehicles participating in federated learning in the Internet of Vehicles. Unlike most existing privacy-preserving federated learning schemes, it does not require complex encryption operations or introduce random noise to affect model performance. Instead, it designs a conditional privacy-preserving identity authentication scheme, which allows vehicles to participate in training by using dynamic pseudonyms, thereby concealing the connection between model parameters and the real identity of the vehicle and maintaining the efficiency of federated learning.
[0074] In addition, the present invention is a certificateless authentication scheme based on elliptic curve cryptography (ECC), Chinese remainder theorem (CRT) and SM2 digital signature algorithm. Compared with most existing schemes, it does not require complex bilinear pairing operations, and the design of dynamic pseudonyms enables it to achieve efficient privacy protection, identity authentication and dynamic member management, and supports batch verification. It can ensure the authenticity and integrity of local model parameters, thereby achieving secure vehicle authentication. In addition, its integration with the national secret SM2 digital signature algorithm makes it have the potential for wide application.
[0075] In the present invention, the private key of the vehicle during the signature verification process consists of two parts, which improves the security of the signature verification.
[0076] In addition, the combination of the present invention and the standard algorithm makes it easier to apply and promote. BRIEF DESCRIPTION OF THE DRAWINGS
[0077] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0078] Figure 1A schematic diagram of the overall steps of the Internet of Vehicles privacy protection federated learning method provided in an embodiment of the present invention.
[0079] Figure 2 A schematic diagram of an application scenario of the method provided in an embodiment of the present invention.
[0080] Figure 3 A schematic diagram of the complete steps for generating vehicle public and private keys during vehicle registration provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0081] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0082] On the one hand, an embodiment of the present invention discloses a privacy-preserving federated learning method for an Internet of Vehicles. Figure 1 As shown, the following steps are included:
[0083] System initialization: The trusted center initializes the system and generates system public parameters;
[0084] Registration and group key generation: The trusted center performs vehicle and roadside unit registration tasks and generates secret keys for distribution to each vehicle. It also generates the relevant keys for communication between the trusted center, vehicles, and roadside units based on the secret keys and system public parameters.
[0085] Local training and model aggregation: The cloud server sends global model parameters to each vehicle in the coverage area. The vehicle trains the local data to obtain local model parameters, signs the local model parameters according to the relevant key to generate verification information and uploads it anonymously to the roadside unit; the roadside unit verifies the local model parameters according to the verification message, locally aggregates the verified local model parameters, and obtains a local aggregate model. The roadside unit then signs the local aggregate model with its own private key and uploads it to the cloud server. The cloud server verifies the local model uploaded by the roadside unit and globally aggregates the verified local models to obtain the global model parameters. At this point, one round of training iteration is completed, and multiple rounds of iterations are performed until the model converges.
[0086] Combine the following Figure 2 The specific application scenarios shown further illustrate the specific steps of the present invention in detail.
[0087] 1. System initialization
[0088] The Trusted Authority (TA) uses the security parameter λ to generate two large prime numbers p and q, where p>q. Let E(F p ) represents a finite field F p Elliptic curve on, G represents the elliptic curve E(F p ) with order q. TA randomly selects As the system master private key, and calculate P pub =s·G as the system public key, and select five one-way and collision-resistant hash functions: And publish the system public parameters
[0089]
[0090] 2. Registration
[0091] 2.1 Vehicle Registration
[0092] like Figure 3 As shown, the vehicle registration steps are as follows: Vehicle V i Generate a random secret value As the first part of its private key, and then calculate the first part of the public key X i =x i G, X i and its real identity RID i It is sent to the trusted center TA through a secure channel. The trusted center TA receives (RID i ,X i ), calculate h i =H(X i ||P pub ), y i =s·h i and Y i =y i ·G,(Y i ,y i ) as the second part of the vehicle's public and private key. The trusted center TA then i Randomly select a prime number As the secret key for subsequent group communication. Finally, y i ,Y i and sk i Send to V i . V i Received y i ,Y i and sk i Then, set its complete public key to (X i ,Y i ), the complete private key is (x i ,yi ).
[0093] In this step, the complete public key and private key are composed of two parts and are used in the subsequent signing step.
[0094] 2.2 Roadside Unit RSU Registration
[0095] RSU j Register your real ID with TA RSUj Register and TA issues a pair of public and private keys to it It is assumed here that all participants in the system know the public keys and system parameters of RSU and TA, and there is no risk of privacy leakage and attack in the registration step.
[0096] 2.3 Group Key Generation
[0097] The trusted center TA uses the secret key sk of n vehicles i Construct a communication group C with the Chinese remainder theorem n ,calculate a i =θ / sk i and b i =(a i ) -1 modsk i Let c i =a i b i , Then randomly select a group key Calculate the group public key β = K u and D pub =K·G, T K is the validity period of the group key. Then use your own private key sk TA For the message {β,D pub ,T K}Sign and broadcast the message
[0098] To Group C n All vehicles and RSUs in Group C n After receiving the message, the vehicle in the group can obtain the group key K≡βmodsk by performing a modulo operation with its own secret key according to the Chinese remainder theorem. i .
[0099] 3. Local training
[0100] 3.1 Local Model Training
[0101] During the tth round of training, the vehicles V in the group i Receive the global model sent by the Cloud Server (CS) After that, you can use your local data to analyze the global model Train to get the local model W i t ,Right now Then use your own private key and group key to encrypt the local model W i t It is signed and then sent to the adjacent RSU for local aggregation.
[0102] 3.2 Signature
[0103] First, the vehicle randomly generates a Generate pseudonymous PID i =(PID i,1 ,PID i,2 ), where PID i,1 =c i G,
[0104] calculate in Indicates PID i,2 The bit length is converted into two bytes, a and b are F p The elements in F p An elliptic curve on T i Indicates the current timestamp;
[0105] The group key during registration is randomly selected by the trusted center TA. The trusted center TA sends the group public key β to the vehicles in the area. After receiving β, the vehicle can calculate the group key K. Specifically, the group key K≡βmodsk is calculated i and the signing key Random Selection Calculate K i =k i G = (x1, y1), e i =H5(Z i ||W i t |T i ), r i =e i +x1modq,
[0106] s i =(1+sgk i ) -1 ·(k i -r i ·sgk i )modq, judge r i =0 or r i +k i =q? and si =0? If it satisfies, choose a new k i Calculate and finally generate the signature σ i =(r i ,s i ). Vehicle dispatch To communicate with the RSU or other vehicles.
[0107] 4 Model Aggregation
[0108] 4.1 Local Aggregation
[0109] RSU j Received After that, perform signature verification as follows:
[0110] First, RSUs j Check timestamp T i The freshness of the arrival time is T a , if ΔT ≥ T a -T i , then continue, otherwise discard the message;
[0111] Then calculate e i =H5(Z i ||W i t ||T i ), h i =H(Y i ||X i ||P pub ), t i =r i +s i modq,
[0112] Judge R = e i +x1'=r i ? If yes, the signature is valid and the message comes from a legitimate vehicle, otherwise it is discarded.
[0113] When RSU j Receive a batch of n vehicles {V1,V2,...,V n}Send local model parameters It can also perform batch verification as follows:
[0114] First, RSUs j Check timestamp T i The validity of, where i = 1, 2, ..., n. If T iIf it is valid, continue; otherwise, discard it. In order to prevent conspiracy attacks while ensuring non-repudiation, this scheme also uses a set of small exponents {v1,v2,...,v n} to perform batch verification, where v i ∈[1,2 t ], and t is a small integer.
[0115] Then RSU j calculate
[0116] examine Is it true? If so, all messages are valid; otherwise, some of the messages are invalid.
[0117] RSU j Use the FedAvg algorithm to aggregate the data from n vehicles {V1,V2,...,V n}Received local model parameters Get local aggregation results Where i∈[1,n]. Then RSU j Use your own private key pair Sign and send Perform global aggregation for cloud servers.
[0118] 4.2 Global Aggregation
[0119] CS receives the signal sent by the nearby RSU After that, first verify its validity. Then the local aggregation results that have passed the verification Perform global aggregation to obtain the global model for the next round of training Where j∈[1,m], m represents the number of RSUs. Finally, CS signs the global model with its own private key and broadcasts the message To the RSUs and vehicles in the group. At this point, a round of training is completed, and multiple rounds of iterations are performed until the model converges.
[0120] 5. Group member management
[0121] Tracing
[0122] When RSU j When abnormal behavior of the vehicle is detected, such as identity forgery or sending malicious local model parameters, it can be reported to TA, who can use the system master private key to restore the real identity of the vehicle based on the pseudonym
[0123] Revocation
[0124] TA obtains the RID of the malicious vehicle iThen you can remove it from the communication group C as follows n Cancel:
[0125] Calculate u' = uc i Delete legal information related to the vehicle;
[0126] Select a new group key Calculate the new group public key β' = K'·u', D' pub = K'·G, then broadcast the message to RSUs and vehicles;
[0127] After receiving the message, the vehicles in the group use their own key sk j The new group key K'≡β'modsk can be obtained through a modulo operation j Since u' no longer contains the malicious vehicle V i The legitimate information of the malicious vehicle V i The updated group key cannot be obtained, achieving efficient revocation of malicious vehicles.
[0128] When a vehicle leaves the communication group, it can also be revoked according to the above steps.
[0129] Add to
[0130] When the vehicle V i Initiate to join communication group C n When a request is made, TA adds it to the group as follows:
[0131] TA chooses a new group key Calculate θ' = θ·f i , a′ i =θ' / f i and b′ i =(a′ i ) - 1 modsk i , c′ i =a' i b′ i ,
[0132] Calculate the new group public key β' = K'·u', D' pub = K'·G, then broadcast the message to RSUs and vehicles;
[0133] After receiving the message, the vehicles in the group use their own key sk i The new group key K'≡β'modf can be obtained through a modulo operation i .
[0134] In addition to the above method, the present invention also discloses a privacy-preserving federated learning system for Internet of Vehicles, including a trusted center, a roadside unit, and a cloud server;
[0135] Wherein, the trusted center includes an initialization module and a registration and group key generation module;
[0136] The initialization module is used to initialize the system and generate system common parameters;
[0137] The registration and group key generation module is used to perform vehicle and roadside unit registration tasks and generate secret keys to distribute to each vehicle, and generate relevant keys for communication between the trusted center, vehicles and roadside units based on the secret keys and system public parameters;
[0138] The cloud server transmits the global model parameters to each vehicle in the coverage area, each vehicle trains the local data to obtain the local model parameters, signs the local model parameters according to the relevant keys to generate verification information and uploads it anonymously to the roadside unit; the roadside unit verifies the local model parameters according to the verification message, locally aggregates the verified local model parameters to obtain the local aggregate model; the roadside unit signs the local aggregate model with its own private key and uploads it to the cloud server; the cloud server verifies the local model uploaded by the roadside unit and globally aggregates the verified local models to obtain the global model parameters. At this point, one round of training iteration is completed, and multiple rounds of iterations are performed until the model converges.
[0139] The above system also includes a group member management module, which is located in the trusted center and is used to update the group key based on the global model parameters, system public parameters, and related keys and realize dynamic management of vehicles.
[0140] In the above system, training local data specifically includes the following steps:
[0141] During the training of round t, the vehicles V in the communication group i Receive the global model sent by the cloud server Then, use the vehicle V i Local data for global model Train to get the local model W i t ,Right now
[0142] Using Vehicle V i The private key (x i ,y i ) and the group key K for the local model W i tSigned and sent to adjacent roadside units for aggregation, where x i For each vehicle V i The first part of the private key, specifically the vehicle V i Generate a random secret value y i For each vehicle V i The second part of the private key is determined by the trusted center based on the vehicle V i The corresponding real identity is calculated.
[0143] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0144] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A privacy-preserving federated learning method for Internet of Vehicles, characterized in that: The following steps are involved: System initialization: The trusted center initializes the system and generates system public parameters; Registration and group key generation: The trusted center performs vehicle and roadside unit registration tasks and generates secret keys for distribution to each vehicle. It also generates the relevant keys for communication between the trusted center, vehicles, and roadside units based on the secret keys and system public parameters. Local training and model aggregation: The cloud server sends global model parameters to each vehicle in the coverage area. Each vehicle trains local data to obtain local model parameters, signs the local model parameters according to the relevant keys to generate verification information and uploads it anonymously to the roadside unit; the roadside unit verifies the local model parameters according to the verification message, locally aggregates the verified local model parameters to obtain a local aggregate model; the roadside unit uses its own private key to sign the local aggregate model and uploads it to the cloud server; the cloud server verifies the local model uploaded by the roadside unit and globally aggregates the verified local models to obtain the global model parameters. At this point, one round of training iteration is completed, and multiple rounds of iterations are repeated until convergence.
2. The method for privacy-preserving federated learning in Internet of Vehicles according to claim 1, characterized in that: It also includes group member management steps: the trusted center updates the group key based on the global model parameters, system public parameters, and related keys to achieve dynamic management of vehicles.
3. The method for privacy-preserving federated learning in Internet of Vehicles according to claim 1, characterized in that: The trusted center initializes the system and generates system public parameters, which specifically includes the following steps: The trusted center uses the security parameter λ to generate two prime numbers p and q, where p>q. Let E(F p ) represents a finite field F p Elliptic curve on, G represents the elliptic curve E(F p ) with order q; Random selection of trusted centers As the system master private key, and calculate P pub =s·G as the system public key, and select five one-way and collision-resistant hash functions: And publish the system public parameters 4. The method for privacy-preserving federated learning in Internet of Vehicles according to claim 1, characterized in that: The trusted center performs the registration task of the vehicle and the roadside unit and generates a secret key to distribute to each vehicle, and generates the relevant keys for communication between the trusted center, the vehicle and the roadside unit based on the secret key and the system public parameters, which specifically includes the following steps: Vehicle Registration: Each vehicle V i Generate a random secret value As the first part of the private key of the vehicle, and calculate the first part of the public key X of the vehicle based on the first part of the private key i =x i ·G; The first part of the public key X i The real ID corresponding to the vehicle i Sent to a trusted center via a secure channel; The trusted center receives (RID i ,X i ), calculate h i =H(X i ||P pub ), y i =s·h i and Y i =y i ·G,(Y i ,y i ) i and i As vehicle V i The second part of the public key and private key; The trusted center then provides each vehicle V i Randomly select a prime number as a secret key for subsequent group communications; Send y through the safe channel i ,Y i and sk i Send to vehicle V i ; Vehicle V i Received y i ,Y i and sk i Then, set its complete public key to (X i ,Y i ), the complete private key is (x i ,y i ); Roadside Unit Registration: Roadside Unit RSU j Register their real identity ID with a trusted center RSUj Register and the trusted center will issue a pair of public and private keys to it Group key generation: The trusted center builds a communication group C based on the secret keys of n vehicles and the Chinese remainder theorem. n ,calculate a i =θ / sk i and b i =(a i ) -1 modsk i ; Let c i =a i b i , Then randomly select a group key Calculate the group public key β = K u and D pub =K·G, T K is the validity period of the group key; Then use the trusted center's own private key sk TA For the message {β,D pub ,T K }Sign and broadcast the message To communication group C n All vehicles and roadside units in the.
5. The method for privacy-preserving federated learning in Internet of Vehicles according to claim 1, characterized in that: Local training and model aggregation specifically include the following steps: During the training of round t, the vehicles V in the communication group i Receive the global model sent by the cloud server Then, use the vehicle V i Local data for global model Train to get the local model W i t ,Right now Using Vehicle V i The private key (x i ,y i ) and group key K for the local model W i t Signed and then sent to the adjacent roadside units for aggregation; where x i For each vehicle V i The first part of the private key, specifically the vehicle V i Generate a random secret value y i For each vehicle V i The second part of the private key is determined by the trusted center based on the vehicle V i The corresponding real identity is calculated.
6. The method for privacy-preserving federated learning in Internet of Vehicles according to claim 5, characterized in that: Using Vehicle V i The private key (x i ,y i ) and the group key K for the local model W i t Signed and sent to nearby roadside units for aggregation, including: Signature: Vehicle V i Generate a random element based on a finite field Based on the generated element c i Generate pseudonymous PID i =(PID i,1 ,PID i,2 ), where PID i,1 =c i G, H i represents a hash function; calculate in Indicates PID i,2 The bit length is converted into two bytes, a and b are F p The elements in T i Indicates the current timestamp; Calculate the group key K≡βmodsk i and the signing key Random Selection Calculate K i =k i G = (x1, y1), e i =H5(Z i ||W i t |T i ), r i =e i +x1modq,s i =(1+sgk i ) -1 ·(k i -r i ·sgk i )modq, judge r i =0 or r i +k i =q? and s i =0? If it satisfies, choose a new k i Calculate and finally generate the signature σ i =(r i ,s i ); Vehicle delivery to adjacent roadside units; Local aggregation: Roadside Unit RSU j Received After that, perform signature verification as follows: First, the roadside unit RSU j Check timestamp T i Freshness: The arrival time is T a , if ΔT ≥ T a -T i , then continue, otherwise discard the message; Then calculate e i =H5(Z i ||W i t ||T i ), h i =H(Y i ||X i ||P pub ), t i =r i +s i modq, Judge R = e i +x1'=r i ? If yes, the signature is valid and the message comes from a legitimate vehicle, otherwise it is discarded; Roadside Unit RSU j Use the FedAvg algorithm to aggregate the data from n vehicles {V1,V2,...,V n }Received local model parameters Get local aggregation results Where i∈[1,n]. Then RSU j Use your own private key pair Sign and send Perform global aggregation for cloud servers; Global aggregation: The cloud server receives data from nearby roadside units. After that, first verify its effectiveness; Then the local aggregation results that pass the verification Perform global aggregation to obtain the global model for the next round of training Where j∈[1,m], m represents the number of RSUs; Finally, the cloud server signs the global model using its own private key and broadcasts the message to said roadside units and vehicles within the communication group; At this point, one round of training is completed, and multiple rounds of iterations can be performed until the model converges.
7. The method for privacy-preserving federated learning in Internet of Vehicles according to claim 2, characterized in that: Tracking revocation management specifically includes the following steps: Tracing When the roadside unit RSU j When abnormal behavior of the vehicle is detected, it is reported to the trusted center, which uses the system master private key to restore the true identity of the vehicle based on the pseudonym. Revocation The trusted center obtains the real identity RID of the malicious vehicle i Then remove the malicious vehicle from the communication group C as follows n Cancel: Calculate u' = uc i Delete legal information related to the vehicle; Select a new group key Calculate the new group public key β' = K'·u', D' pub = K'·G, then broadcast the message to roadside units and other vehicles; After receiving the message, the vehicles in the group use their own key sk j The new group key K'≡β'modsk can be obtained through a modulo operation j ; Add to When the vehicle V i Initiate to join communication group C n When a request is made, the trusted center adds it to the group as follows: The trusted center selects a new group key Calculate θ' = θ·f i , a' i =θ' / f i and b' i =(a' i ) -1 modsk i , c' i =a' i ·b' i , Calculate the new group public key β' = K'·u', D' pub = K'·G, then broadcast the message to roadside units and other vehicles; After receiving the message, the vehicles in the group use their own key sk i The new group key K'≡β'mod f can be obtained by a single modulo operation i .
8. A privacy-preserving federated learning system for Internet of Vehicles, characterized in that: Including, trusted center, roadside unit and cloud server; Wherein, the trusted center includes an initialization module and a registration and group key generation module; The initialization module is used to initialize the system and generate system common parameters; The registration and group key generation module is used to perform vehicle and roadside unit registration tasks and generate secret keys to distribute to each vehicle, and generate relevant keys for communication between the trusted center, vehicles and roadside units based on the secret keys and system public parameters; The cloud server transmits the global model parameters to each vehicle in the coverage area, each vehicle trains the local data to obtain the local model parameters, signs the local model parameters according to the relevant keys to generate verification information and uploads it anonymously to the roadside unit; the roadside unit verifies the local model parameters according to the verification message, locally aggregates the verified local model parameters to obtain the local aggregate model; the roadside unit signs the local aggregate model with its own private key and uploads it to the cloud server; the cloud server verifies the local model uploaded by the roadside unit and globally aggregates the verified local models to obtain the global model parameters. At this point, one round of training iteration is completed, and multiple rounds of iterations are performed until the model converges.
9. The Internet of Vehicles privacy protection federated learning system according to claim 8, characterized in that: It also includes a group member management module, which is located in the trusted center and is used to update the group key based on the global model parameters, system public parameters, and related keys and realize dynamic management of vehicles.
10. The Internet of Vehicles privacy protection federated learning system according to claim 8, characterized in that: Training local data specifically includes the following steps: During the training of round t, the vehicles V in the communication group i Receive the global model sent by the cloud server Then, use the vehicle V i Local data for global model Train to get the local model W i t ,Right now Using Vehicle V i The private key (x i ,y i ) and group key K for the local model W i t Signed and sent to adjacent roadside units for aggregation, where x i For each vehicle V i The first part of the private key, specifically the vehicle V i Generate a random secret value y i For each vehicle V i The second part of the private key is determined by the trusted center based on the vehicle V i The corresponding real identity is calculated.
Citation Information
Patent Citations
Asynchronous federated learning method, system, device and terminal for mobile vehicle
CN114116198A
Multilayer center asynchronous federal learning method applied to Internet of Vehicles
CN114827198A
Batch aggregation method for federated learning in Internet of Vehicles environment
CN115549901A
Privacy enhancement multi-layer federal learning method applied to Internet of Vehicles
CN116881902A
Internet of vehicles identity authentication and privacy protection method based on federated learning and unmanned aerial vehicle
CN118042464A
Cited By
Batch authentication and dynamic group key negotiation method for Internet of Vehicles
CN122294110A
