Internet of vehicles privacy calculation method and system based on adversarial federal incremental learning

Abstract

The invention provides a car networking privacy calculation method and system based on adversarial federal incremental learning, and aims to solve the privacy protection problem in user privacy data updating. Comprising the following steps: collecting Internet of Vehicles privacy data; constructing a local generation model by adopting a generative adversarial network model according to the privacy data of the Internet of Vehicles; constructing a cloud center global model by adopting a generative adversarial network model; acquiring newly added privacy data, and constructing a newly added global edge model based on federal incremental learning; obtaining a cloud update center global model based on the newly added global edge model; updating and optimizing the local generation model based on the cloud updating center global model to obtain Internet of Vehicles privacy data; according to the method, effective updating and maintenance are carried out on the premise that user privacy information is not leaked, the privacy protection level is improved by combining a multi-terminal generative adversarial training technology and a cloud global model aggregation mechanism, and the data security of an Internet of Vehicles system and the confidentiality of user privacy are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of adversarial federated incremental learning, and in particular to a vehicle network privacy computing method and system based on adversarial federated incremental learning. Background Art

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

[0003] The continuous development of social economy has driven the rapid development of the automobile industry, and many companies have continuously increased their investment in automobiles. Through real-time monitoring and data analysis of the Internet of Vehicles, driving safety, driving experience, and vehicle management and maintenance have been optimized, while promoting the development of intelligence and automation. This technology not only improves traffic efficiency and environmental protection, but also promotes the rise of new data-driven services and business models, becoming an important driving force for innovation and future development of the automobile industry.

[0004] At present, the inventors have found that there are also some problems in the widespread application of Internet of Vehicles:

[0005] First, there are issues of privacy protection and data security. Insufficient protection of data privacy in the Internet of Vehicles leads to leakage of privacy data during data storage and transmission, making the abuse of user data a potential risk.

[0006] Second, Internet of Vehicles data is usually stored and processed in a centralized manner, which cannot meet the needs of dynamically adapting to data changes in the Internet of Vehicles environment, and the system lacks flexibility and real-time performance.

[0007] Third, the existing models used require the collection of a large amount of data for training, which increases the number of data transmissions and thus increases the risk of data leakage.

[0008] Therefore, in order to ensure the privacy and security of users and better adapt to the needs of data changes in the Internet of Vehicles environment, effective data management and protection measures should be taken for the Internet of Vehicles. Summary of the invention

[0009] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a vehicle network privacy computing method and system based on adversarial federated incremental learning. By combining incremental learning, federated learning and adversarial generative network models, the privacy protection and data security problems existing in the prior art are solved. At the same time, it can generate more diverse data sets without leaking private data, improve the accuracy and robustness of the model, and use distributed collaborative computing methods to better meet the needs of dynamically adapting to data changes in the vehicle network environment, improve the flexibility and real-time performance of the system, and greatly reduce computing costs while protecting user privacy.

[0010] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0011] A first aspect of the present invention provides a privacy computing method for Internet of Vehicles based on adversarial federated incremental learning, comprising:

[0012] Collect private data of Internet of Vehicles;

[0013] According to the Internet of Vehicles privacy data, a local generation model is constructed using a generative adversarial network model;

[0014] A generative adversarial network model is used to build a global model in the cloud center;

[0015] Acquire new privacy data and build a new global edge model based on federated incremental learning;

[0016] Based on the newly added global edge model, a cloud update center global model is obtained;

[0017] Based on the global model of the cloud update center, the local generation model is updated and optimized to obtain Internet of Vehicles privacy data.

[0018] As an implementation method, the local generation model is constructed using a generative adversarial network model based on the Internet of Vehicles privacy data. The specific method is as follows:

[0019] Cleaning and standardizing the Internet of Vehicles privacy data;

[0020] Build a generator and train it with preprocessed data to generate synthetic data that looks real but does not contain individual privacy information;

[0021] Build a discriminator and train it using the preprocessed data and the synthetic data generated by the generator to distinguish between real data and synthetic data based on the loss function;

[0022] The generator and the discriminator are confronted and repeated alternating training is performed iteratively;

[0023] Generate synthetic data based on the trained generator to update and maintain the model.

[0024] As an implementation method, the specific method of acquiring the newly added privacy data and constructing a newly added global edge model based on federated incremental learning is as follows:

[0025] Obtain new privacy data from the Internet of Vehicles;

[0026] Calculate the new data proportion and the new data model impact according to the new privacy data;

[0027] Calculate the influence weight according to the influence of the newly added data model;

[0028] According to the influence of the newly added data model and the influence weight, the weight of the newly added edge data is weighted and aggregated to obtain a newly added global edge model.

[0029] As an implementation method, the cloud update center global model is obtained based on the newly added global edge model, specifically:

[0030] The cloud center server performs weighted aggregation on the parameters of the newly added global edge model received to obtain weighted aggregated parameters;

[0031] The cloud center global model is updated according to the weighted aggregated parameters to obtain a cloud update center global model.

[0032] As an implementation method, based on the global model of the cloud update center, the local generation model is updated and optimized to obtain the Internet of Vehicles privacy data. The specific calculation process is:

[0033] Send the global model of the cloud update center to each device;

[0034] Each device updates the local generation model and continues to perform adversarial generation incremental learning on the newly added data samples.

[0035] As an implementation method, the newly added global edge model calculation formula is:

[0036]

[0037] Where n' is the total amount of new data added by all devices, is the total amount of new data added by device e; θ t,e is the model parameter weight based on the amount of device data, It means that at time t, each device e i The weight of .

[0038] A second aspect of the present invention provides a vehicle networking privacy computing system based on adversarial federated incremental learning, comprising:

[0039] A data collection module configured to collect Internet of Vehicles privacy data;

[0040] A local generation model module is constructed, configured to construct a local generation model using a generative adversarial network model according to the Internet of Vehicles privacy data;

[0041] Build a cloud center global model module, and configure it to use a generative adversarial network model on the cloud to build a cloud center global model;

[0042] Build a new global edge model module, configure it to obtain new privacy data, and build a new global edge model based on federated incremental learning;

[0043] Constructing a cloud update center global model module, configured to obtain a cloud update center global model based on the newly added global edge model;

[0044] The Internet of Vehicles privacy computing module is configured to update and optimize the locally generated model based on the global model of the cloud update center.

[0045] The third aspect of the present invention aims to provide a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the method described in the first aspect of the present invention.

[0046] A fourth aspect of the present invention provides a computer device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method described in the first aspect of the present invention are implemented.

[0047] A fifth aspect of the present invention is to provide a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the methods and functions involved in any of the above-mentioned embodiments.

[0048] One or more of the above technical solutions have the following beneficial effects:

[0049] (1) In the IoV data management of this embodiment, incremental learning, federated learning and adversarial generative network models are combined to solve the privacy protection and data security problems existing in the prior art, and prevent IoV privacy data from being stolen, tampered with, or abused by unauthorized visitors during transmission and storage, thereby greatly improving the security of IoV data.

[0050] (2) In this embodiment, a distributed collaborative computing method and model update are adopted to realize decentralized storage and processing of data, which greatly reduces the risk and cost of data transmission. At the same time, it can dynamically adapt to the needs of data changes and model updates in the Internet of Vehicles environment, thereby improving the flexibility and real-time performance of the system. The cloud and edge devices are made to work together, making the Internet of Vehicles data processing and model updates more efficient and secure.

[0051] (3) In this embodiment, realistic synthetic data samples are generated by using an adversarial generative network model and combined with federated learning, which can generate more diverse data sets without leaking private data, further improving the accuracy and robustness of the model.

[0052] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0054] Figure 1 This is a principle flow chart of a vehicle networking privacy computing method based on adversarial federated incremental learning according to an embodiment of the present invention;

[0055] Figure 2 A principle flow chart of adversarial generative network modeling of a central server of a vehicle networking privacy computing method based on adversarial federated incremental learning in an embodiment of the present invention;

[0056] Figure 3 This is a flowchart of a newly added global edge model in a vehicle network privacy computing method based on adversarial federated incremental learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0058] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0059] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0060] The overall idea proposed by the present invention is as follows: in the Internet of Vehicles, the original data is first collected by each device end and transmitted to the edge end through a secure encrypted channel; at the edge end, a local generative model is constructed using generative adversarial network technology, and trained based on the original data; at the same time, a cloud center global model is constructed using generative adversarial network technology; newly added privacy data is obtained, and on the basis of the generative adversarial network model, federated incremental learning is used for the local generative model to construct a newly added global edge model; based on the newly added global edge model, a cloud update center global model is obtained; and the cloud update center global model is sent to each member end (device end and edge end), and each member end updates the local generative model and continues to perform adversarial generative incremental learning on the newly added data samples.

[0061] Embodiment 1

[0062] This embodiment discloses a privacy computing method for Internet of Vehicles based on adversarial federated incremental learning.

[0063] In order to more clearly explain this embodiment, the implementation process of Internet of Vehicles privacy computing based on adversarial federated incremental learning can be specifically described as follows:

[0064] A privacy computing method for Internet of Vehicles based on adversarial federated incremental learning, including:

[0065] S1. Collect private data of Internet of Vehicles;

[0066] S2. Building a local generation model using a generative adversarial network model based on the Internet of Vehicles privacy data;

[0067] S3, using the generative adversarial network model to build a cloud-based global model;

[0068] S4. Obtain new privacy data and build a new global edge model based on federated incremental learning;

[0069] S5. Based on the newly added global edge model, obtain a cloud update center global model;

[0070] S6. Based on the global model of the cloud update center, the local generation model is updated and optimized to obtain Internet of Vehicles privacy data.

[0071] like Figure 1 As shown, in step S1, each device collects the original data of the Internet of Vehicles and transmits it to the edge through a secure encrypted channel.

[0072] The acquired raw data of the Internet of Vehicles provides a data foundation for subsequent model construction.

[0073] like Figure 1 , Figure 2 As shown, in step S2, a local generation model is constructed using a generative adversarial network model according to the Internet of Vehicles privacy data. The specific process is as follows:

[0074] S201: Clean and standardize the Internet of Vehicles data.

[0075] In this embodiment, the least square method is used to clean the Internet of Vehicles data, and then the cleaned data attribute values ​​are converted into a unified format, that is, the attribute values ​​of the data are converted into a consistent or unified format.

[0076] In this embodiment, other existing methods can also be used to clean and standardize the Internet of Vehicles data, which will not be described in detail here.

[0077] After the above steps, problems such as inaccurate data, data duplication, and data missing are solved from the source of the data, improving data quality and providing an accurate data foundation for subsequent better models.

[0078] S202: Build a generator and use the preprocessed data for training to generate synthetic data that appears to be real but does not contain individual privacy information.

[0079] In this example, a generator is built on the device side. The generator network is usually a deconvolutional neural network with the following structure:

[0080] 1) Input layer: accepts random noise z, which is usually a multi-dimensional vector, such as a noise vector that obeys a normal distribution or a uniform distribution.

[0081] 2) Hidden layer: includes multiple deconvolution layers or fully connected layers, which gradually map the input random noise to a higher dimensional space.

[0082] 3) Output layer: Generate the final output through the last deconvolution layer This output can be an image, a piece of text, or other representation of data.

[0083] The generator network is used to generate samples that look like real data from random noise, usually using a function G that maps the input random vector z to samples in the data space. The generator formula is:

[0084]

[0085] Among them, θ g is the generator θ g Parameters that control how the generator converts random noise z into samples z is the input of the generator, representing random noise; θ g are the parameters of the generator network, learned through the training process, and used to adjust the weights and biases of the network. is the output of the generator, that is, the sample output by the model.

[0086] The generator is trained using preprocessed real data to generate synthetic data samples that appear real but do not contain individual privacy information.

[0087] After the above steps, synthetic data samples that appear to be real but do not contain individual privacy information are generated to protect privacy: the selection of a generator and subsequent continuous optimization through feedback from the discriminator can help generate more realistic forged data, which is used to replace real data in privacy computing and improve data availability: after subsequent adversarial training of the discriminator and generator, the generator can obtain a distribution closer to the real data after receiving feedback from the discriminator, thereby making the generated data more available.

[0088] S203, construct a discriminator, and use the preprocessed data and the synthetic data generated by the generator for training, and distinguish between real data and synthetic data according to the loss function.

[0089] In this embodiment, the task of the discriminator is to evaluate whether the input sample comes from the true data distribution p data The probability of (x). This is done through a binary classifier, which outputs a value between 0 and 1 to represent the possibility of being real data. Assume that the discriminator D is a neural network that accepts an input sample x and outputs a scalar value D(x) that represents the probability that the sample x comes from the real data distribution.

[0090] During the training process, the discriminator needs to maximize its ability to recognize real data and minimize the recognition error of synthetic data generated by the generator. This goal is achieved through the following loss function:

[0091]

[0092] Among them, Ex~pdata(x)[log D(x)] means that the discriminator hopes to maximize the probability that the real data x is judged as real; this expected value is calculated by the sample x in the real data set; It means that the discriminator hopes to minimize the probability that the fake data G(z) generated by the generator is discriminated as real data. This expected value can be calculated by the random noise z input to the generator. Using this loss function in training can enable the discriminator to better distinguish between real data and synthetic data.

[0093] Constructing a loss function can promote adversarial training of the generator and the discriminator: this loss function allows the generator and the discriminator to compete with each other and optimize the performance of the discriminator and the generator in the game.

[0094] After the above steps, the discriminator is trained using the preprocessed real data and the synthetic data generated by the generator, so that it can accurately identify which data is real and which is generated. That is, the discriminator feeds back to the generator, so that the generator is continuously optimized, which can help generate more realistic forged data, which is used to replace real data in privacy computing.

[0095] S204: The generator is pitted against the discriminator, and the training is repeated alternately for iteration.

[0096] In this example, the parameters θ of the generator G are g and the parameters θ of the discriminator D d Initialize; after initialization, fix the parameters of the generator to train the parameters of the discriminator, then fix the parameters of the discriminator to train the parameters of the generator, and repeat the alternating training for iteration.

[0097] In the process of training the parameters of the discriminator while fixing the parameters of the generator G, the parameters θ of D are updated by maximizing V(D,G) d When the parameters of the discriminator are fixed and the parameters of the generator are trained, the parameters θ of the generator G are updated by minimizing V(D,G) g .

[0098] After the above steps, during the training process, the generator generates data samples and adjusts them through the feedback of the discriminator to generate more realistic data; the discriminator competes with the generator to improve the recognition accuracy by distinguishing the generated synthetic data from the real data. The generator can generate samples that are closer and closer to the real data, gradually narrowing the gap with the real data. Through continuous training, the discriminator becomes more adept at distinguishing between real samples and generated samples, thereby improving its classification accuracy. Avoid the problems of mode collapse and gradient disappearance in the generator and discriminator during training.

[0099] S205. Generate synthetic data based on the trained generator to update and maintain the model.

[0100] In this example, the trained generator is used to generate synthetic data, which can be used to update and maintain the local model.

[0101] like Figure 1 , Figure 2 As shown, in step S3, a generative adversarial network model is used to build a cloud center global model. The specific process is:

[0102] S301, construct a generator and use the preprocessed data for training to generate synthetic data that looks real but does not contain individual privacy information;

[0103] S302, construct a discriminator, and use the preprocessed data and the synthetic data generated by the generator for training, and distinguish the real data from the synthetic data according to the loss function;

[0104] S303, the generator and the discriminator are confronted, and the alternating training is repeated for iteration;

[0105] S304: Generate synthetic data based on the trained generator to update and maintain the model.

[0106] The specific process is the same as the steps for building a local generation model, so it will not be repeated here.

[0107] After the above steps, a cloud center global model is obtained, which provides a model basis for the subsequent cloud update center global model.

[0108] like Figure 1 , Figure 3 As shown, in step S4, newly added privacy data is obtained, and a newly added global edge model is constructed based on federated incremental learning. The process of obtaining the newly added global edge model is as follows:

[0109] S401, obtaining new data of the Internet of Vehicles;

[0110] In this implementation, 1) a new Internet of Vehicles data sample C is obtained and preprocessed to obtain the preprocessed C e , providing reliable data for subsequent calculations.

[0111] 2) Define the initial generative adversarial network learning model and build the generator and discriminator.

[0112] 3) Distribute to each device, build a model based on local data, and generate synthetic data with similar distribution but without leaking individual information through generative adversarial networks. Distribute this initial model to each member (device).

[0113] 4) After receiving the initialized generative adversarial model, each member terminal sets the parameters θ of the discriminator G g and the parameters θ of the generator D d Initialize, perform incremental learning tasks based on local data streams or tasks without sending local data to the central server, and continuously optimize the discriminator and generator through the loss function (2). Every time the discriminator and generator cross iterate a certain number of times, the member end

[0114] After the above steps, the newly acquired Internet of Vehicles data is input into the model, and the discriminator and generator are iteratively trained to obtain a better model.

[0115] S402: Calculate the proportion of new data and the impact of the new data model based on the new privacy data.

[0116] The cloud center server receives the global edge aggregation model parameter information sent by each member, and for each newly added device data sample, obtains the ratio of the preprocessed new sample to all new data samples and the model influence of the new sample.

[0117] In this embodiment, every time a new data sample is added, the data sample C is processed according to the preprocessing. e, calculate the proportion of new samples at this time W e And the impact of the new sample model The calculation formula is as follows:

[0118]

[0119] D e is the current data volume of the device, It is the influence of the sample model at the previous moment. The model influence is the impact of the local data of the device on the model performance after the device completes the edge global model aggregation during the federated incremental learning process.

[0120] After the above steps, when the model influence is calculated, some data can be added or deleted based on the model influence, and high-quality data can be retained as much as possible, thereby optimizing the overall data quality and enhancing the performance of the model.

[0121] S403: Calculate the influence weight according to the influence of the new data model.

[0122] In this embodiment, the SoftPlus algorithm is used to calculate the parameter weights to obtain the model parameter weights based on the device data volume. The Softplus function is a smooth nonlinear function that can be used to achieve gradual growth, but the growth rate gradually slows down. The influence of the model is processed with the Softplus function so that the weight growth of the device with greater influence gradually slows down. The formula for calculating the model influence weight is:

[0123]

[0124] The output of the SoftPlus function is always non-negative, and its logarithmic function is approximately linear when the input is large, which makes it very suitable for weight generation, provides smoother gradients in training, and enhances the robustness of the model; the non-negativity and smoothness of the SoftPlus function can better handle the problem of sparsity, so this embodiment adopts the SoftPlus algorithm.

[0125] After the above steps, different features are quantified using a unified standard. Through different weights, the importance of different data can be better evaluated, thereby helping the model learn more useful feature data and reducing the risk of overfitting the model.

[0126] S404: Perform weighted aggregation on the weight of the newly added edge data according to the influence of the newly added data model and the influence weight to obtain a newly added global edge model.

[0127] In this embodiment, after calculating the influence of the cluster model, the weight of the newly added global edge data is weighted and aggregated to obtain a newly added global edge model. The calculation formula is:

[0128]

[0129] Where n' is the total amount of new data added by all devices, is the total amount of new data added by device e; θ t,e is the model parameter weight based on the amount of device data, It means that at time t, each device e i The weight of .

[0130] After the above steps, after the data is collected from the edge devices, a global model can better consider the data on all devices more comprehensively, efficiently fuse the collected data, and ensure that the model can be optimized according to the new data sample distribution after obtaining new data samples.

[0131] like Figure 1 As shown, in step S5, based on the newly added global edge model, the cloud update center global model is obtained, specifically:

[0132] S501. The cloud center server performs weighted aggregation on the received parameters of the newly added global edge model to obtain weighted aggregated parameters.

[0133] In this embodiment, after obtaining the global model of the newly added edge, the cloud center server performs weighted aggregation on the received newly added edge parameters.

[0134] Through weighted aggregation, the impact of data sources on the global model can be balanced, ensuring that model parameters are emphasized to different degrees based on the importance and quality of each device data.

[0135] S502: Update the cloud center global model according to the weighted aggregated parameters to obtain a cloud update center global model.

[0136] In this embodiment, the cloud center server adds a global weight ω of the edge ′ u,s Perform weighted aggregation and update the central global model. The formula is:

[0137]

[0138] in, is the total amount of new data of the i-th edge device, N ′ It is the amount of new data in the cloud.

[0139] Finally, the cloud center server will weight the aggregated parameter ω ′ t The original central adversarial generative network model is updated. After each update of the central learning network model, the central model is sent to each member end.

[0140] After the above steps, each time privacy sample data is added, the global model is updated according to the new sample data, and then the parameters of the global model are updated according to the new model. During the training process, the hyperbolic tangent function is used as the incremental attenuation function as the basis for modifying the model aggregation strategy. After uploading the new edge global strategy, the global model of the new sample is obtained by weighted aggregation. The privacy data calculation after weighing the weights of each device end is realized, and the latest data and features from different terminals can be integrated, thereby improving the accuracy and generalization of the model, so that the model can continuously adapt to the number of features of various terminals.

[0141] like Figure 1 As shown, in step S6, based on the global model of the cloud update center, the local generation model is updated and optimized, and the specific process is:

[0142] S601, sending the global model of the cloud update center to each member terminal;

[0143] S602. Each member terminal updates the local generation model and continues to perform adversarial generation incremental learning on the newly added data samples.

[0144] After the above steps, the local model is updated and optimized to ensure that the model on the member side is consistent with the global model. By importing the global model on the cloud, the mobility of sensitive data can be reduced, privacy can be better protected, and the versatility and robustness of the local model can be improved.

[0145] After steps S4, S5, and S6, privacy calculation of new data in the Internet of Vehicles is realized and sent to each member terminal, which enables continuous updating of the local generation model, and continues to perform adversarial generation incremental learning on new data samples to obtain a better model in a cycle.

[0146] This embodiment constructs an improved federated learning model through incremental learning, uses distributed computing resources for collaborative computing, and then uses an adversarial generative network algorithm to improve the federated incremental learning model. The Softplus algorithm is used to adjust the weight of each device according to the amount of data, and the incremental model parameters are uploaded to the central server.

[0147] At the same time, this embodiment generates a more diverse data set without leaking private data, further improving the accuracy and robustness of the model. The distributed collaborative computing method can better meet the needs of dynamically adapting to data changes in the Internet of Vehicles environment, improving the flexibility and real-time performance of the system. It greatly reduces computing costs while protecting user privacy.

[0148] Embodiment 2

[0149] The purpose of this embodiment is to provide a private computing system for Internet of Vehicles based on adversarial federated incremental learning, including:

[0150] A data collection module configured to collect Internet of Vehicles privacy data;

[0151] A local generation model module is constructed, configured to construct a local generation model using a generative adversarial network model according to the Internet of Vehicles privacy data;

[0152] Build a cloud center global model module, and configure it to use a generative adversarial network model on the cloud to build a cloud center global model;

[0153] Build a new global edge model module, configure it to obtain new privacy data, and build a new global edge model based on federated incremental learning;

[0154] Constructing a cloud update center global model module, configured to obtain a cloud update center global model based on the newly added global edge model;

[0155] The Internet of Vehicles privacy computing module is configured to update and optimize the local generation model based on the global model of the cloud update center to obtain Internet of Vehicles privacy data.

[0156] Embodiment 3

[0157] The purpose of this embodiment is to provide a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the steps in the method described in the first aspect of the present invention are implemented.

[0158] Embodiment 4

[0159] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the program.

[0160] Embodiment 5

[0161] The purpose of the present embodiment is to provide a computer program product including instructions, which, when executed on a computer, enables the computer to execute the methods and functions involved in any of the above embodiments.

[0162] The steps involved in the apparatus of the above embodiment correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0163] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0164] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A privacy computing method for Internet of Vehicles based on adversarial federated incremental learning, characterized in that: include: Collect private data of Internet of Vehicles; According to the Internet of Vehicles privacy data, a local generation model is constructed using a generative adversarial network model; A generative adversarial network model is used to build a global model in the cloud center; Acquire new privacy data and build a new global edge model based on federated incremental learning; Based on the newly added global edge model, a cloud update center global model is obtained; Based on the global model of the cloud update center, the locally generated model is updated and optimized.

2. The Internet of Vehicles privacy computing method based on adversarial federated incremental learning as claimed in claim 1, characterized in that: According to the Internet of Vehicles privacy data, a local generation model is constructed using a generative adversarial network model. The specific method is as follows: Cleaning and standardizing the Internet of Vehicles privacy data; Build a generator and train it with preprocessed data to generate synthetic data that looks real but does not contain individual privacy information; Build a discriminator and train it using the preprocessed data and the synthetic data generated by the generator to distinguish between real data and synthetic data based on the loss function; The generator and the discriminator are confronted and repeated alternating training is performed iteratively; Generate synthetic data based on the trained generator to update and maintain the model.

3. The Internet of Vehicles privacy computing method based on adversarial federated incremental learning as claimed in claim 1, characterized in that: The specific method of obtaining the newly added privacy data and constructing a new global edge model based on federated incremental learning is as follows: Obtain new privacy data from the Internet of Vehicles; Calculate the new data proportion and the new data model impact according to the new privacy data; Calculate the influence weight according to the influence of the newly added data model; According to the influence of the newly added data model and the influence weight, the weight of the newly added edge data is weighted and aggregated to obtain a newly added global edge model.

4. The Internet of Vehicles privacy computing method based on adversarial federated incremental learning as claimed in claim 1, characterized in that: Based on the newly added global edge model, the cloud update center global model is obtained, specifically: The cloud center server performs weighted aggregation on the parameters of the newly added global edge model received to obtain weighted aggregated parameters; The cloud center global model is updated according to the weighted aggregated parameters to obtain a cloud update center global model.

5. The Internet of Vehicles privacy computing method based on adversarial federated incremental learning as claimed in claim 1, characterized in that: Based on the global model of the cloud update center, the local generation model is updated and optimized to obtain the Internet of Vehicles privacy data. The specific calculation process is as follows: Send the global model of the cloud update center to each device; Each device updates the local generation model and continues to perform adversarial generation incremental learning on the newly added data samples.

6. The Internet of Vehicles privacy computing method based on adversarial federated incremental learning as claimed in claim 1, characterized in that: The new global edge model calculation formula is: Where n' is the total amount of new data for all devices, n' is the total amount of new data for device e; θ t,e is the model parameter weight based on the amount of device data, It means that at time t, each device e i The weight of .

7. A privacy computing system for Internet of Vehicles based on adversarial federated incremental learning, characterized in that: include: A data collection module configured to collect Internet of Vehicles privacy data; A local generation model module is constructed, configured to construct a local generation model using a generative adversarial network model according to the Internet of Vehicles privacy data; Build a cloud center global model module, and configure it to use a generative adversarial network model on the cloud to build a cloud center global model; Build a new global edge model module, configure it to obtain new privacy data, and build a new global edge model based on federated incremental learning; Constructing a cloud update center global model module, configured to obtain a cloud update center global model based on the newly added global edge model; The Internet of Vehicles privacy computing module is configured to update and optimize the locally generated model based on the global model of the cloud update center.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are performed.

Images