Construction method for cybertwin-based double-layer internet-of-vehicles federated learning framework

Abstract

Disclosed in the present invention is a construction method for a Cybertwin-based double-layer Internet-of-vehicles federated learning framework. The method comprises: constructing an Internet-of-vehicles double-layer federated learning system, wherein a Cybertwin network framework is divided into three layers, i.e., a central cloud server, edge cloud servers j and vehicle devices i, the central cloud server being attached to a preset central-cloud-server server, and the edge cloud servers j being attached to preset road test units of the vehicle devices i; the vehicle devices i using local datasets to train local models (1), and edge clouds acquiring historical behaviors of vehicles i, calculating contribution scores of the local models of the vehicles, discarding the local models (1) that do not meet preset model requirements, and obtaining edge cloud models (2) at a moment t+1 by means of an aggregation algorithm; and the edge clouds submitting the models to a central cloud, and the central cloud server aggregating the received local models (3) to obtain a global model ωt+1 at the moment t+1. Thus, the method expands scenarios of federated learning, has stronger applicability, and can resist more backdoor attacks.

Description

A method for constructing a two-layer federated learning framework for Internet of Vehicles based on Cybertwin Technical Field

[0001] The present invention relates to the field of federated learning technology, and in particular to a method for constructing a two-layer Internet of Vehicles federated learning framework based on Cybertwin. Background Art

[0002] Industry is a key area of ​​IoT application. With the continuous advancement of science and technology, the Internet of Vehicles (IoV) has become synonymous with intelligent manufacturing and Industry 4.0. Various advanced intelligent technologies, such as artificial intelligence (AI), machine learning, augmented and virtual reality (AR / VR), digital twins / threads, and cloud / edge computing, are increasingly being integrated into every aspect of industrial production.

[0003] Intelligent vehicles are an important development trend in the future transportation sector, but achieving fully autonomous driving faces many challenges, one of which is how to obtain vehicle data for model training. Intelligent vehicles obtain perception information through a variety of sensors (such as cameras, lidar, ultrasonic sensors, radar, GPS, etc.), where cameras may generate several megabytes of data per second. Directly uploading all perception data will result in huge communication load and network resource consumption, and there are privacy issues. Therefore, existing researchers have proposed a distributed federated learning solution that uses the model crowdsourcing concept of the Internet of Vehicles to train large models for intelligent vehicles. Federated learning allows multiple vehicle devices to collaboratively train a shared global model based on local training data, and aggregates all local model parameters through a central cloud server to generate an improved global model, thereby reducing network resource consumption and protecting the privacy of vehicle devices. To achieve this goal, existing researchers have proposed a two-layer federated learning aggregation framework based on the Cybertwin network framework.

[0004] The Cybertwin network framework provides three key functions to support crowdsourcing model training: communication assistance, data loggers, and digital assets. Communication assistance ensures the accuracy of vehicle identities in connected vehicle environments. Data loggers optimize traffic flow control and road maintenance strategies by analyzing vehicle data. Digital assets, combined with digital encryption and blockchain technology, ensure the security and integrity of model data and assign value to it. These functions provide the theoretical basis and functional support for model crowdsourcing.

[0005] However, there are challenges in aggregating large models through model crowdsourcing. Factors such as distributed network attacks, differences in computing power and data structure, high vehicle dynamics, communication bandwidth limitations, and connection intermittentity can all affect the aggregation and iteration of the global model. Summary of the Invention

[0006] In view of the shortcomings of the above-mentioned existing technologies, the present invention provides a method for constructing a two-layer federated learning framework for the Internet of Vehicles based on Cybertwin. The two-layer federated learning aggregation based on the Cybertwin network framework has the ability to increase the tolerance ratio of malicious nodes and resist diverse backdoor attacks, and is effective in different vehicle scenarios.

[0007] To achieve the above effects, the technical solutions of the present invention are as follows:

[0008] In the first aspect, the present invention provides a method for constructing a two-layer federated learning framework for the Internet of Vehicles based on Cybertwin. An edge cloud server j is deployed between a vehicle device i and a central cloud server to build a two-layer federated learning system for the Internet of Vehicles. Vehicle device i, edge cloud server j, and the central cloud server collaborate to complete model training. Vehicle device i, edge cloud server j, and the central cloud server communicate with each other via a wireless link.

[0009] The specific steps include:

[0010] Step 1: Build a two-layer federated learning system for the Internet of Vehicles. The Cybertwin network framework of the Internet of Vehicles federated learning model is divided into three layers: the central cloud server, the edge cloud server j, and the vehicle device i. The central cloud server is attached to the preset central cloud server, and the edge cloud server j is attached to the preset road test unit of the vehicle device i.

[0011] The central cloud server uses its own preset private key S G Signature, corresponding to the public key P preset by edge cloud server j j The global model ω at time t t Encrypt and send to edge cloud server j;

[0012] Edge cloud server j uses its own preset private key S j The encrypted global model ω at time t t Decrypt and get the decrypted global model ω t , ensuring the communication security between edge cloud server j and central cloud server; each edge cloud server j collects the list of vehicle equipment i And the global model ω t Send to vehicle device i;

[0013] Step 2: Vehicle device i trains a local model using a local dataset And send it to the edge cloud server j, the edge cloud calculates the local model The contribution score and the historical behavior of vehicle equipment i obtained through the cybertwin node will not meet the local model requirements of the preset model. Discard and get the edge cloud model at time t+1 The local model that does not meet the preset model requirements For: Local Model Not in the default history behavior, or local model Not within the preset contribution score, or local model Not within the preset quality;

[0014] Vehicle device i uses private key S i Sign and use the public key P corresponding to the edge cloud server j G For the local model at time t+1 Encrypt and send to the drive test unit;

[0015] Edge cloud server j obtains the local model set submitted by vehicle device i within the range of the road test unit And the local model collection Aggregate to get the t+1 time model of edge cloud server j

[0016] Edge cloud server j uses its own private key S j Sign and pass the central cloud server public key P G For local models After encryption, it is sent to the central cloud server;

[0017] Step 3: The central cloud server receives the local model Aggregate to get the global model ω at time t+1 t+1 ;

[0018] Step 4: Repeat steps 1-3 above until the global iteration is completed and the final global model ω is obtained. T .

[0019] In the present invention, vehicle device i can be an IoT device such as a vehicle, and i represents the i-th vehicle. The present invention designs an efficient and secure aggregation algorithm and a two-layer federated learning framework based on the Cybertwin (network twin) framework to resist various backdoor attacks.

[0020] Furthermore, the local data set used for training in step 2 is generated by vehicle device i through information collection; in order to reduce the transmission pressure of the return link, vehicle device i only communicates with edge cloud server j, and edge cloud server j can communicate with both vehicle device i and cloud server.

[0021] Furthermore, in step 2, the vehicle device i trains the local model using the local data set include:

[0022] Assuming that a multi-classification task is currently being performed, initialize the global model ω 0 , learning rate η of vehicle equipment i, local training round k, global iteration round T;

[0023] Vehicle device i receives the initialized global model ω 0 After that, through the local dataset (x i ,y i ) Train to get the local model at time t+1 Right now

[0024] The edge cloud server j obtains the computing power of vehicle device i through the following formula:

[0025] Among them, A t+1 represents the reference value of the data volume of vehicle equipment i at time t, a t+1 represents the data volume of vehicle device i at time t+1; Indicates the model quality performance of the local model on the local test set; represents the JS divergence of the test set of vehicle device i and edge cloud server j.

[0026] Furthermore, in step 2, the vehicle device i trains the local model using the local data set Later also includes:

[0027] The local model quality of vehicle device i is calculated by the following formula:

[0028] Among them, α represents the sliding factor; represents the contribution of vehicle equipment i to the global model at time t+1; Represents the distance between vehicle i and the global model at time t+1.

[0029] Furthermore, step 2 calculates the historical behavior of vehicle device i by the following formula:

[0030] in, is a binary number, when When it is 1, it means that the local model of vehicle device i is used at time t, otherwise it is 0; represents the local model quality of vehicle equipment i at time t; the operation E(·) represents the sigmoid function operation.

[0031] Furthermore, step 2 obtains the local model by the following formula Contribution score

[0032] in, express Cosine similarity of the differences, is the local model at time t+1 The difference with the global model at time t, is the difference between the local model at time t+1 and the vehicle equipment i at time t;

[0033] If the local model Contribution score If greater than 0, add it to the aggregation list And the local model is calculated by the following formula Contribution score Normalize to get the local model Contribution score Aggregation weight

[0034] Furthermore, step 2 combines the local model Aggregate to get the t+1 time model of edge cloud server j include:

[0035] Aggregate and obtain the t+1 time model of edge cloud server j through the following formula

[0036] Furthermore, in step 3, the central cloud server receives the edge cloud model Aggregate to get the global model ω at time t+1 t+1 ,include:

[0037] The central cloud server collects the t+1 time model of edge cloud server j The t+1 time model of edge cloud server j is given by the following formula Consider and obtain the t+1 time model of edge cloud server j Quality contribution score

[0038] in, Represents the t+1 time model of edge cloud server j Performance on the central cloud server test set;

[0039] If the quality contribution score If it is greater than β', then add it to the aggregation list G t+1 ; and the quality contribution score is calculated by the following formula Normalize to get the aggregation weight;

[0040] The central cloud server will receive the local model Aggregate by aggregation weights and obtain the global model at time t+1

[0041] In a second aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute a method for constructing a two-layer vehicle network federated learning framework based on Cybertwin through the computer program.

[0042] In a third aspect, the present invention provides a computer-readable storage medium, comprising a stored computer program, wherein the computer program executes a method for constructing a two-layer vehicle network federated learning framework based on Cybertwin when running.

[0043] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0044] The present invention improves upon the shortcomings of existing robust federated learning and constructs a two-layer federated learning system for the Internet of Vehicles. The Cybertwin network framework of the Internet of Vehicles federated learning model is divided into three layers: a central cloud server, an edge cloud server j, and a vehicle device i. The central cloud server is attached to a preset central cloud server, and the edge cloud server j is attached to a preset road test unit of the vehicle device i.

[0045] Vehicle device i trains a local model using a local dataset And send it to the edge cloud server j, which obtains the local model set submitted by the vehicle device i within the range of the road test unit And the local model collection Aggregate to get the t+1 time model of edge cloud server j The central cloud server will receive the edge cloud model And aggregate to get the global model ω at time t+1 t+1 ; It expands the scenarios of federated learning, makes it more applicable, can resist more backdoor attacks, enhances the robustness of the global model, and improves the maximum fault tolerance rate of malicious nodes, making it suitable for more application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] FIG1 is a schematic diagram of a two-layer federated learning framework based on the Cybertwin network framework of the present invention. DETAILED DESCRIPTION

[0047] The following describes the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art will readily appreciate the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the various details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are intended only to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0048] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0049] Definition of noun:

[0050] RSU, short for Road Side Unit, is a roadside unit in the ETC system. It uses DSRC (Dedicated Short Range Communication) technology to communicate with the on-board unit (OBU) to identify vehicles and electronically deduct points. In highway and parking lot management, installing RSUs on the roadside creates unmanned, dedicated express lanes.

[0051] Example

[0052] This embodiment proposes a method for constructing a two-layer federated learning framework for the Internet of Vehicles (IoV) based on Cybertwin. Referring to Figure 1, an edge cloud server j is deployed between vehicle device i and a central cloud server to build an IoV two-layer federated learning system. Vehicle device i, edge cloud server j, and the central cloud server collaborate to complete model training. Vehicle device i, edge cloud server j, and the central cloud server communicate with each other via wireless links.

[0053] The specific steps include:

[0054] Step 1: Build a two-layer federated learning system for the Internet of Vehicles. The Cybertwin network framework of the Internet of Vehicles federated learning model is divided into three layers: the central cloud server, the edge cloud server j, and the vehicle device i. The central cloud server is attached to the preset central cloud server, and the edge cloud server j is attached to the preset road test unit of the vehicle device i.

[0055] The central cloud server uses its own preset private key S G Signature, corresponding to the public key P preset by edge cloud server jj The global model ω at time t t Encrypt and send to edge cloud server j;

[0056] Edge cloud server j uses its own preset private key S j The encrypted global model ω at time t t Decrypt and get the decrypted global model ω t , in order to ensure the communication security between edge cloud server j and central cloud server; each edge cloud server j collects the list of vehicle equipment i And the global model ω t Send to vehicle device i;

[0057] Step 2: Vehicle device i trains a local model using a local dataset And send it to the edge cloud server j, the edge cloud calculates the local model The contribution score and the historical behavior of vehicle equipment i obtained through the cybertwin node will not meet the local model requirements of the preset model. Discard and get the edge cloud model at time t+1 The local model that does not meet the preset model requirements For: Local Model Not in the default history behavior, or local model Not within the preset contribution score, or local model Not within the preset quality;

[0058] At time h, vehicle device i uses private key S i Sign and use the public key P corresponding to the edge cloud server j G For the local model at time t+1 Encrypt and send to the drive test unit;

[0059] Edge cloud server j obtains the local model set submitted by vehicle device i within the range of the road test unit And the local model collection Aggregate to get the t+1 time model of edge cloud server j

[0060] Edge cloud server j uses its own private key S j Sign and pass the central cloud server public key P G For local models After encryption, it is sent to the central cloud server;

[0061] Step 3: The central cloud server receives the edge cloud model Aggregate to get the global model ω at time t+1 t+1 ;

[0062] Step 4: Repeat steps 1-3 above until the global iteration is completed and the final global model ω is obtained. T .

[0063] As a preferred technical solution, in this embodiment, the local data set used for training in step 2 is generated by vehicle device i through information collection; in order to reduce the transmission pressure of the return link, vehicle device i only communicates with the edge cloud server j, and the edge cloud server j can communicate with both the vehicle device i and the cloud server.

[0064] As a preferred technical solution, in this embodiment, the vehicle device i in step 2 trains the local model with the local data set include:

[0065] Assuming that a multi-classification task is currently being performed, initialize the global model ω 0 , vehicle equipment i learning rate η, local training round k (in actual application, it can be adjusted according to the situation of vehicle equipment i to obtain better results), global iteration round T;

[0066] Vehicle device i receives the initialized global model ω 0 After that, through the local dataset (x i ,y i ) Train to get the local model at time t+1 Right now

[0067] Taking edge cloud server j as an example, we get Then start calculating its contribution and historical behavior;

[0068] The edge cloud server j obtains the computing power of vehicle device i through the following formula:

[0069] Among them, A t+1 It represents the reference value of the data volume of vehicle equipment i at time t, and the specific value depends on different tasks; a t+1 represents the data volume of vehicle device i at time t+1; Indicates the model quality performance of the local model on the local test set; represents the JS divergence of the test set of vehicle device i and edge cloud server j.

[0070] It's understood that data volume refers to the size of the dataset held by vehicle device i. For example, for multi-class image recognition, suppose vehicle device iA holds 3,000 images, while vehicle device iB holds 2,000 images. This is a common data distribution. During local model training, the dataset is divided into a training set and a test set. The local test set is the test set used for training.

[0071] As a preferred technical solution, in this embodiment, the vehicle device i in step 2 trains the local model with the local data set Later also includes:

[0072] The local model quality of vehicle device i is calculated by the following formula:

[0073] Among them, α represents the sliding factor; represents the contribution of vehicle equipment i to the global model at time t+1; Represents the distance between vehicle i and the global model at time t+1.

[0074] As a preferred technical solution, in this embodiment, step 2 calculates the historical behavior of vehicle device i by the following formula:

[0075] in, is a binary number, when When it is 1, it means that the local model of vehicle device i is used at time t, otherwise it is 0; represents the local model quality of vehicle device i at time t; the operation E(·) represents the sigmoid function operation, which makes events closer to the current time more meaningful.

[0076] As a preferred technical solution, in this embodiment, step 2 obtains the local model by the following formula: Contribution score

[0077] in, express Cosine similarity of the differences, is the local model at time t+1 The difference with the global model at time t, is the difference between the local model at time t+1 and the vehicle equipment i at time t;

[0078] If the local model Contribution score If greater than 0, add it to the aggregation list And the local model is calculated by the following formula Contribution score Normalize to get the local model Contribution score Aggregation weight

[0079] As a preferred technical solution, in this embodiment, step 2 sets the local model Aggregate to get the t+1 time model of edge cloud server j include:

[0080] Aggregate and obtain the t+1 time model of edge cloud server j through the following formula

[0081] As a preferred technical solution, in this embodiment, step 3, the central cloud server receives the edge cloud model Aggregate to get the global model ω at time t+1 t+1 ,include:

[0082] The central cloud server collects the t+1 time model of edge cloud server j The t+1 time model of edge cloud server j is given by the following formula Consider and obtain the t+1 time model of edge cloud server j Quality contribution score

[0083] in, Represents the t+1 time model of edge cloud server j The performance in the central cloud server test set is so that even if a problem occurs in an edge cloud server j, it can be identified in the central cloud server;

[0084] If the quality contribution score If it is greater than β', then add it to the aggregation list G t+1 ; and the quality contribution score is calculated by the following formula Normalize to get the aggregation weight;

[0085] The central cloud server will receive the local model Aggregate by aggregation weights and obtain the global model at time t+1

[0086] It should be noted that the vehicle device i in the present invention can be an IoT device such as a vehicle.

[0087] In another embodiment of the present invention, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the present invention implements the method described in any of the above embodiments.

[0088] In this embodiment, the computer system suitable for implementing the electronic device of the embodiment of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) or the program loaded from the storage portion into the random access memory (RAM), such as the method described in the above embodiment. Various programs and data required for system operation are also stored in the RAM. The CPU, ROM, and RAM are connected to each other via a bus. The input / output (I / O) interface is also connected to the bus.

[0089] The following components are connected to the I / O interface: an input section including a keyboard, mouse, etc.; an output section including a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section including a hard disk; and a communication section including a network interface card such as a LAN (Local Area Network) card and a modem. The communication section performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as needed. Removable media such as magnetic disks, optical disks, magneto-optical disks, semiconductor memories, etc. are installed in the drive as needed so that computer programs read from them can be installed into the storage section as needed.

[0090] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication portion, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the various functions defined in the system of the present invention are performed.

[0091] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0092] In this embodiment, the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a computer processor, the computer executes the method provided in any of the aforementioned embodiments. The computer-readable storage medium may be included in the electronic device described in the aforementioned embodiments, or may exist independently and not be incorporated into the electronic device.

[0093] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the described example embodiments can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.) or on a network, and includes a number of instructions to enable a computing device (which can be a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of the present invention.

[0094] The implementation of the embodiments of the present invention has the following beneficial effects:

[0095] 1) Technical problems to be solved by the present invention:

[0096] In the connected vehicle (IoV) network, the complexity of communication, diverse data structures, and multiple backdoor attacks complicate model aggregation. Therefore, it is necessary to design new federated learning frameworks to increase the tolerance for malicious nodes and new federated learning methods to enhance the model's stability against various backdoor attacks, while also ensuring the model's robustness in complex communications.

[0097] 2) Technical solutions adopted to solve its technical problems

[0098] The present invention comprehensively considers multiple key factors in the field of Internet of Vehicles and is used to solve the complex challenges such as multiple backdoor attacks in data acquisition and model training of intelligent vehicle devices i. In order to optimize the local model aggregation process, the local model aggregation is based on the consideration of the distance between models and the degree of heterogeneity of vehicle device i data. Aggregate to get the t+1 time model of edge cloud server j The present invention can reasonably reflect the performance and data quality of each local model, thereby ensuring that the contribution of each local model is accurately evaluated and effectively utilized. At the same time, by considering the contribution of vehicle device i to the global model, the ability to dynamically adjust the number and weight of vehicle devices i participating in training is achieved, which can maximize the use of contributing devices and improve the performance of the global model. In view of the complexity of communication of vehicle device i, the federated learning framework of the present invention is adopted to reduce the unreliability caused by communication complexity and increase the maximum ratio of maliciously tolerated nodes. The Cybertwin network framework fully considers the application of the Cybertwin network framework in the Internet of Vehicles, thereby effectively enhancing the applicability and robustness of the federated learning model of the Internet of Vehicles.

[0099] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a two-layer Internet of Vehicles federated learning framework based on Cybertwin, characterized by: Deploy edge cloud server j between vehicle device i and the central cloud server to build a two-layer federated learning system for the Internet of Vehicles. Vehicle device i, edge cloud server j, and the central cloud server collaborate to complete model training. Vehicle device i, edge cloud server j, and the central cloud server communicate with each other via wireless links. The specific steps include: Step 1: Build a two-layer federated learning system for the Internet of Vehicles. The Cybertwin network framework of the Internet of Vehicles federated learning model is divided into three layers: a central cloud server, an edge cloud server j, and a vehicle device i. The edge cloud server j is attached to the road test unit preset in the vehicle device i. The central cloud server uses its own preset private key S G Signature, corresponding to the public key P preset by edge cloud server j j The global model ω at time t t Encrypt and send to edge cloud server j; Edge cloud server j uses its own preset private key S j The encrypted global model ω at time t t Decrypt and get the decrypted global model ω t , ensuring the communication security between edge cloud server j and central cloud server; each edge cloud server j collects the list of vehicle equipment i And the global model ω t Send to vehicle device i; Step 2: Vehicle device i trains a local model using a local dataset And send it to edge cloud server j, which calculates the local model The contribution score and the historical behavior of vehicle equipment i obtained through the cybertwin node will not meet the local model requirements of the preset model. Discard and get the edge cloud model at time t+1 The local model that does not meet the preset model requirements For: Local Model Not in the default history behavior, or local model Not within the preset contribution score, or local model Not within the preset quality; Vehicle device i uses private key S i Sign and use the public key P corresponding to the edge cloud server j G For the local model at time t+1 Encrypt and send to the drive test unit; Edge cloud server j obtains the local model set submitted by vehicle device i within the range of the road test unit And the local model collection Aggregate to get the t+1 time model of edge cloud server j Edge cloud server j uses its own private key S j Sign and pass the central cloud server public key P G Model of edge cloud server j at time t+1 After encryption, it is sent to the central cloud server; Step 3: The central cloud server will receive the edge cloud model And aggregate to get the global model ω at time t+1 t+1 ; Step 4: Repeat steps 1-3 above until the global iteration is completed and the final global model ω is obtained. T .

2. The method according to claim 1, characterized in that In step 2, the local dataset used for training is generated by vehicle device i through sensors and locators. To reduce the transmission pressure of the backhaul link and the scalability of the aggregation framework, vehicle device i only communicates with edge cloud server j, while edge cloud server j can communicate with both vehicle device i and cloud servers.

3. The method according to claim 1, characterized in that Step 2 calculates the historical behavior of vehicle device i by the following formula: in, is a binary number, when When it is 1, it means that the local model of vehicle device i is used at time t, otherwise it is 0; represents the local model quality of vehicle equipment i at time t; the operation E(·) represents the sigmoid function operation.

4. The method according to claim 1, wherein In step 2, the vehicle device i trains the local model using the local data set include: Assuming that a multi-classification task is currently being performed, initialize the global model ω 0 , learning rate η of vehicle equipment i, local training round k, global iteration round T; Vehicle device i receives the initialized global model ω 0 After that, through the local dataset (x i ,y i ) Train to get the local model at time t+1 Right now The edge cloud server j obtains the computing power of vehicle device i through the following formula: Among them, A t+1 represents the reference value of the data volume of vehicle equipment i at time t, a t+1 represents the data volume of vehicle device i at time t+1; Indicates the model quality performance of the local model on the local test set; represents the JS divergence of the test set of vehicle device i and edge cloud server j.

5. The method according to claim 4, characterized in that In step 2, the vehicle device i trains the local model using the local data set Later also includes: The local model quality of vehicle device i is calculated by the following formula: Among them, α represents the sliding factor; represents the contribution of vehicle equipment i to the global model at time t+1; Represents the distance between vehicle i and the global model at time t+1.

6. The method according to claim 5, characterized in that Step 2: Get the local model by the following formula Contribution score in, express Cosine similarity of the differences, is the local model at time t+1 The difference with the global model at time t, is the difference between the local model at time t+1 and the vehicle equipment i at time t; If the local model Contribution score If greater than 0, add it to the aggregation list And the local model is calculated by the following formula Contribution score Normalize to get the local model Contribution score Aggregation weight 7. The method according to claim 6, characterized in that Step 2: Collect local models Aggregate to get the t+1 time model of edge cloud server j include: Aggregate and obtain the t+1 time model of edge cloud server j through the following formula 8. The method according to claim 5, characterized in that Step 3 The central cloud server receives the edge cloud model Aggregate to get the global model ω at time t+1 t+1 ,include: The central cloud server collects the t+1 time model of edge cloud server j The t+1 time model of edge cloud server j is given by the following formula Consider and obtain the t+1 time model of edge cloud server j Quality contribution score in, Represents the t+1 time model of edge cloud server j Performance on the central cloud server test set; If the quality contribution score If it is greater than β', then add it to the aggregation list G t+1 ; and the quality contribution score is calculated by the following formula Normalize to get the aggregation weight; The central cloud server will receive the local model Aggregate by aggregation weights and obtain the global model at time t+1 9. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the Cybertwin-based two-layer vehicle network federated learning framework construction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the Cybertwin-based two-layer vehicle network federated learning framework construction method as described in any one of claims 1 to 7.

Images