A malware propagation control method based on GAN data enhancement for internet of vehicles

Abstract

The application discloses a vehicle networking malicious software propagation control method based on GAN data enhancement, and the method is characterized in that: the propagation law of malicious software in vehicle networking is modeled, the Pontryagin maximum principle is applied to obtain an optimal control solution of the malicious software propagation in vehicle networking, and then the scale of a data set is expanded by using the optimal state control of GAN, so as to obtain a training data set of a neural network for controlling the malicious software propagation in vehicle networking, the propagation of the malicious software in vehicle networking can be greatly prevented, and the adverse influence of the malicious software on vehicles and users is reduced.

Description

Technical Field

[0001] This invention relates to the technical field of information security in the Internet of Vehicles (IoV), and specifically to a method for controlling the propagation of malicious software in the IoV based on GAN data enhancement. Background Technology

[0002] The Internet of Vehicles (IOV) is the application of Cyber-physical Systems (CPS) in Intelligent Transportation Systems (ITS). It highly integrates advanced technologies such as mobile communication, information processing, sensor technology, and automatic control, organically connecting traffic participants such as people, vehicles, and roads. This enables rapid information sharing and exchange between vehicles, between vehicles and people, and between vehicles and road infrastructure, transforming moving vehicles into a large-scale wireless mobile network. Furthermore, as a highly efficient information exchange network, the IIV can play a significant role in reducing traffic accident rates, alleviating traffic congestion, and providing personalized services, thus possessing very broad application prospects and economic benefits.

[0003] However, in the Internet of Vehicles (IoV), vehicle nodes heavily rely on wireless communication technology for information transmission and presentation to users through upper-layer applications. While this brings convenience, the IoV's openness, real-time nature, and dynamic topology significantly increase the risk of malware intrusion and propagation. Like ordinary networks, the IoV faces cybersecurity threats from malicious attacks such as viruses, worms, and Trojans. The spread of this malware can leak vehicle and user privacy data and affect vehicles' reception and response to traffic information, potentially leading to traffic accidents and seriously threatening users' lives and property. Summary of the Invention

[0004] This invention addresses the information security issues of the Internet of Vehicles (IoV) and proposes a method for controlling the propagation of malicious software in IoV based on GAN data augmentation. This method is implemented through the following technical solution:

[0005] A method for controlling the propagation of vehicle-to-everything (V2X) malware based on GAN data augmentation, comprising:

[0006] S1: Establish a traffic flow model to obtain the expression for vehicle distance under IDM (Intelligent Driver Mode) balanced flow conditions;

[0007] S2: Establish a vehicle-to-vehicle communication channel model to obtain the probability of successfully receiving wireless signals;

[0008] S3: Establish a vehicle-to-everything (V2X) malware propagation model and obtain the fractional differential equation for V2X malware propagation;

[0009] S4: The optimal control solution for the fractional differential equation of malware propagation in the Internet of Vehicles is calculated based on the Pondrikin principle.

[0010] S5: Based on the optimal state control of GAN, the original dataset is expanded to obtain a new dataset.

[0011] Preferably, in S1, the expression for the vehicle distance under the IDM balanced flow state is as follows:

[0012]

[0013] Among them, s ij =x i -x j The distance between vehicle j and its adjacent preceding vehicle i represents the net distance, T represents the reaction time, s0 represents the minimum bumper clearance when traffic is completely stopped, v represents the vehicle speed in a balanced flow state, v0 represents the desired speed in a free flow, δ represents the acceleration exponent, and s e Indicates the distance between vehicles.

[0014] More preferably, in S2, the probability density function of the received wireless signal power obtained from the distance between the vehicles in S1 is as follows:

[0015]

[0016] and

[0017]

[0018] Where, p t This indicates the wireless signal transmission power of the equipped vehicle; λ represents the wavelength, G t G represents the gain of the transmitting antenna; r f(p) represents the gain of the receiving antenna. r (s ij )) represents the received wireless signal power p r (s ij The probability density function of ).

[0019] Better still, by giving a critical received power p su That is, the signal power received by the vehicle is greater than p su If so, the wireless signal can be successfully transmitted; the probability that the vehicle successfully receives the wireless signal is as follows:

[0020]

[0021]

[0022] Even better, in S3, the infection rate probability is expanded by combining traffic flow and vehicle-to-vehicle communication models as follows:

[0023]

[0024] Where R represents the communication radius of the vehicle; combined, the fractional differential equation for the propagation of vehicle-to-everything (V2X) malware is obtained.

[0025] Preferably, in step S4, a new fractional differential equation for the propagation of vehicle-to-everything (V2X) malware is obtained by combining control strategies of immunity and treatment. The Pondrikin principle is then introduced to solve the optimal control problem of the fractional differential equation for the propagation of V2X malware, and finally the control quantity that minimizes the objective function J(u1(t),u2(t),u3(t)) of the fractional differential equation for the propagation of V2X malware is obtained.

[0026] Preferably, in step S5, a GAN model is obtained through an objective function, and the training process is viewed as a minimax game process. New optimal state control pairs are synthesized from the learned distribution of the simulated data. The objective function expression is as follows:

[0027]

[0028] in, Let x represent the expected value of the distribution of the real data, and z represent random noise. The expected value of the Gaussian noise distribution, D(x) represents the probability that it is real data, G(z) represents the data generated by the generator from the noise z, and D(G(z)) represents the probability that the generated data is real data.

[0029] Preferably, in step S5, the dataset of the optimal control solution is expanded by introducing GAN, increasing the size of the dataset and making the trained neural network more robust, thereby enhancing the robustness of controlling the spread of vehicle network malware through the neural network.

[0030] Compared with the prior art, the present invention also has the following advantages:

[0031] (1) This invention is based on reality and combines vehicle traffic behavior and communication factors to establish a model, which has strong practical significance;

[0032] (2) This patent aims to solve the problem of small size of the optimal control dataset. It expands the dataset based on GAN (Generative Adversarial Network) to enhance the robustness of controlling the spread of malware through neural networks. Attached Figure Description

[0033] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0034] Figure 1 This is a flowchart illustrating the implementation of a GAN-based data augmentation-based method for controlling the propagation of malware in the Internet of Vehicles (IoV).

[0035] Figure 2 This invention is a state transition diagram based on Pondrikin's extreme value principle, incorporating control quantities.

[0036] Figure 3 This is a structural diagram of the GAN data augmentation method of this invention. Detailed Implementation

[0037] The following detailed description of a method for controlling the propagation of malware in the Internet of Vehicles based on GAN data augmentation, with reference to specific embodiments, is provided. These embodiments are for comparison and explanation purposes only, and the present invention is not limited to these embodiments.

[0038] like Figure 1 As shown, a method for controlling the propagation of vehicle-to-everything (V2X) malware based on GAN data augmentation includes:

[0039] A method for controlling the propagation of vehicle-to-everything (V2X) malware based on GAN data augmentation, comprising:

[0040] S1: Establish a traffic flow model to obtain the expression for vehicle distance under IDM (Intelligent Driver Mode) balanced flow conditions;

[0041] S2: Establish a vehicle-to-vehicle communication channel model to obtain the probability of successfully receiving wireless signals;

[0042] S3: Establish a vehicle-to-everything (V2X) malware propagation model and obtain the fractional differential equation for V2X malware propagation;

[0043] S4: The optimal control solution for the fractional differential equation of malware propagation in the Internet of Vehicles is calculated based on the Pondrikin principle.

[0044] S5: Based on the optimal state control of GAN, the original dataset is expanded to obtain a new dataset.

[0045] Preferably, in S1, the expression for the vehicle distance under the IDM balanced flow state is as follows:

[0046]

[0047] Among them, s ij =x i -x jThe distance between vehicle j and its adjacent preceding vehicle i represents the net distance, T represents the reaction time, s0 represents the minimum bumper clearance when traffic is completely stopped, v represents the vehicle speed in a balanced flow state, v0 represents the desired speed in a free flow, δ represents the acceleration exponent, and s e Indicates the distance between vehicles.

[0048] More preferably, in S2, the probability density function of the received wireless signal power obtained from the distance between the vehicles in S1 is as follows:

[0049]

[0050] and

[0051]

[0052] Where, p t This indicates the wireless signal transmission power of the equipped vehicle; λ represents the wavelength, G t G represents the gain of the transmitting antenna; r f(p) represents the gain of the receiving antenna. r (s ij )) represents the received wireless signal power p r (s ij The probability density function of ).

[0053] Better still, by giving a critical received power p su That is, the signal power received by the vehicle is greater than p su If so, the wireless signal can be successfully transmitted; the probability that the vehicle successfully receives the wireless signal is as follows:

[0054]

[0055] Even better, in S3, the infection rate probability is expanded by combining traffic flow and vehicle-to-vehicle communication models as follows:

[0056]

[0057] Where R represents the communication radius of the vehicle; combined, a fractional differential equation for the propagation of malware in the Internet of Vehicles is obtained.

[0058] Preferably, in step S4, by combining the control strategies of immunity and treatment, a fractional differential equation for the propagation of new vehicle-to-everything (V2X) malware is obtained, and the Pondrikin extreme value principle is introduced to solve the optimal control problem of the fractional differential equation for the propagation of new V2X malware. Finally, the control quantity that minimizes the objective function J(u1(t),u2(t),u3(t)) of the fractional differential equation for the propagation of new V2X malware is obtained.

[0059] Preferably, in step S5, a GAN model is obtained through an objective function, and the training process is viewed as a minimax game process. New optimal state control pairs are synthesized from the learned distribution of the simulated data. The objective function expression is as follows:

[0060]

[0061] in, Let x represent the expected value of the distribution of the real data, and z represent random noise. The expected value of the Gaussian noise distribution, D(x) represents the probability that it is real data, G(z) represents the data generated by the generator from the noise z, and D(G(z)) represents the probability that the generated data is real data.

[0062] Preferably, in step S5, the dataset of the optimal control solution is expanded by introducing GAN, increasing the size of the dataset and making the trained neural network more robust, thereby enhancing the robustness of controlling the spread of vehicle network malware through the neural network.

[0063] In one embodiment, S1: Establish a traffic flow model to obtain an expression for vehicle distance under IDM (Intelligent Driver Mode) balanced flow conditions;

[0064] The purpose of establishing an Intelligent Driver Model (IDM) is to accurately describe the dynamics of urban traffic and contribute to building a more objective model of vehicle movement. The Intelligent Driver Model (IDM), as a simple, efficient, and widely accepted traffic flow theory model, is used in this patent to describe vehicle mobility. Compared to other traffic flow models, the IDM is also very simple and exhibits clear and intuitive physical meaning. The dynamic equations of the IDM are as follows:

[0065]

[0066] In the formula: a represents the acceleration of vehicle i at time t. i b represents the maximum acceleration of vehicle i. i v represents the maximum deceleration of vehicle i, v0 represents the desired velocity in free traffic flow, and Δv ij =v j -v i δ represents the relative velocity between vehicle j and its adjacent vehicle i in front, δ represents the acceleration exponent and typically ranges from 1 to 5, and s * s represents the expected safe distance between the two vehicles, s0 represents the minimum bumper spacing when traffic is completely stopped, s ij =x i -x jT represents the net distance between vehicle j and its adjacent vehicle i in front, and T represents the reaction time.

[0067] However, this invention considers all vehicles to be in a balanced flow state, that is, all vehicles maintain the same speed v and vehicle spacing s. e , that is Therefore, the expression for vehicle distance under IDM balanced flow conditions is obtained:

[0068]

[0069] In the formula: v is the speed of the vehicle under balanced flow conditions.

[0070] S2: Establish a vehicle-to-vehicle communication channel model to obtain the probability of successfully receiving wireless signals;

[0071] Essentially, the channel model describes the connectivity and efficiency of malicious information transmission between vehicles under traffic conditions, particularly the distance and speed between vehicles. Malicious software packets share the same communication channel as normal packets, typically following the IEEE 802.11p protocol (also known as DSRC). IEEE 802.11p is an approved revision of the IEEE 802.11 standard, adding wireless access to the vehicle communication system. Wireless communication transmission is affected by factors such as other vehicles and buildings; therefore, incorporating this crucial factor of the communication channel into the model is of great significance to the method of this invention.

[0072] Rayleigh fading is a classic method for describing signal attenuation and is the theory used in this invention. It considers the numerous reflections and scattering waves during signal transmission, making it suitable for analyzing wireless signal transmission in dense urban traffic environments. Combined with the distance s between vehicles in S1... ij We can obtain the probability density function of the received wireless signal power:

[0073]

[0074] and

[0075]

[0076] Where, p t This indicates the wireless signal transmission power of the equipped vehicle; λ represents the wavelength, G t G represents the gain of the transmitting antenna; r f(p) represents the gain of the receiving antenna. r (s ij )) represents the received wireless signal power p r (s ij The probability density function of ).

[0077] By giving a critical received power p su That is, the signal power received by the vehicle is greater than p su If so, the wireless signal can be successfully transmitted; the probability that the vehicle successfully receives the wireless signal is as follows:

[0078]

[0079]

[0080] S3: Establish a vehicle-to-everything (V2X) malware propagation model and obtain the fractional differential equation for V2X malware propagation;

[0081] Vehicles in the Internet of Vehicles (IoV) can communicate with each other through communication channels to send and receive data. Due to the openness, real-time nature, and dynamic changes in topology of the IoV, it is vulnerable to malware attacks. Furthermore, considering the memory-based nature of information propagation, and the fact that fractional-order theory more accurately describes this process, this invention uses epidemiological theory and fractional-order theory to describe the propagation behavior of malware. In the IoV, vehicles are the primary actors, roadside devices are traffic assistance devices, and user mobile devices are the users. Therefore, this invention considers the propagation of malware among vehicle groups, roadside devices, and mobile devices in the IoV, and also considers the infection resulting from mutual communication.

[0082] The physical distance between two vehicles must be less than the communication distance for communication between them to occur. Therefore, considering communication factors and the characteristics of traffic dynamics, the infection rate probability is extended by combining the S1 traffic flow model and the S2 vehicle-to-vehicle communication model as follows:

[0083]

[0084] Where R represents the vehicle's communication radius, the fractional differential equation for the propagation of malware in the vehicle network is then derived based on infectious disease dynamics and fractional theory:

[0085]

[0086] Specifically, the SEIR (Susceptible-Exposed-Infectious-Recovered) model is used for vehicle groups, the SIR (Susceptible-Infectious-Recovered) model is used for roadside equipment groups, and the SIS (Susceptible-Infectious-Susceptible) model is used for mobile devices. This indicates that fractional order theory is applied, with order α, start time 0, and end time t; S1(t), S2(t), and S3(t) represent the number of different groups in a susceptible state at time t; I2(t) and I3(t) represent the number of different groups in an infected state at time t; E1(t) represents the number of vehicles in a latent state at time t; subscript 1 represents the various states of the vehicle group, subscript 2 represents the various states of the roadside equipment group, and subscript 3 represents the various states of the user's mobile device.

[0087] Vulnerable states (S1, S2, S3): Individuals in these states lack defensive measures and are vulnerable to attacks by malware. This state is represented by the letter S.

[0088] Latent state (E1): Individuals in this state have been attacked by malware, but do not yet have the ability to spread malware in the vehicle network. This state is represented by E.

[0089] Infection status (I1, I2, I3): Individuals in this status have been successfully attacked by malware and have the ability to spread malware to other individuals in the vehicle network. This status is represented by I.

[0090] Recovery Status (R1, R2): Individuals in this status have updated the security software provided by the operator and are functioning normally. They will not spread malware in the vehicle network. This status is represented by R. Other parameters are explained in Table 1 below:

[0091] Table 1. Annotation Table for Fractional Differential Equations of Malware Propagation in Vehicle Networks

[0092]

[0093]

[0094] S4: The optimal control solution for the fractional differential equation of malware propagation in the Internet of Vehicles is calculated based on the Pondrikin principle.

[0095] Implementing control strategies against vehicles capable of spreading malware and whose geographical locations are constantly changing is extremely difficult. Therefore, this invention considers controlling the spread of malware on roadside and mobile devices to reduce the impact of malware on vehicles. Immunization and treatment are two commonly used methods in infectious disease prevention and control theory, and are also feasible methods to suppress malware spread. Combining the control strategies of immunization and treatment, the new fractional differential equation for malware propagation in the Internet of Vehicles is obtained as follows:

[0096]

[0097] The state transition diagram after adding the control variable is as follows: Figure 2As shown, u1(t) represents the treatment rate of roadside devices successfully attacked by malware at time t; u2(t) represents the immunity rate of roadside devices susceptible to malware attacks at time t; and u3(t) represents the treatment rate of mobile devices successfully attacked by malware at time t.

[0098] The goal of implementing control is to achieve maximum benefit at minimum cost, i.e., to maximize the number of devices that function correctly and minimize the number of devices vulnerable to malware attacks. Therefore, the Pondrigen optimality principle is introduced to solve this optimal control problem.

[0099] First, establish the following objective function:

[0100]

[0101] Where c1 represents the cost coefficient for treating roadside equipment; c2 represents the cost coefficient for immunizing roadside equipment; and c3 represents the cost coefficient for treating mobile devices. The Lagrange function is:

[0102]

[0103] Based on the cost function, the Hamiltonian function is constructed as follows:

[0104]

[0105] Where λ i i = 1, 2, 3, ..., 9 are costate variables.

[0106] ① According to Pondrikin's maximum principle, the costate variables must satisfy the following equation:

[0107]

[0108] ② Transverse condition: λ i (t f )=0,i=1,2,…,9;

[0109] ③Optimality condition:

[0110] Right now:

[0111]

[0112] Solving for:

[0113]

[0114] Finally, we obtain the control variables that minimize the objective function J(u1(t),u2(t),u3(t)). yes:

[0115]

[0116] In other words, the optimal solution to this optimal control problem can guide us on what strategies to adopt to control the spread of malware at what time, such as immunization measures or treatment measures, and the amount to be implemented.

[0117] S5: Based on the optimal state control of GAN, the original dataset is expanded to obtain a new dataset.

[0118] Before employing deep neural networks to control the spread of malware in connected vehicles, the neural network needs to be trained. However, the optimal state control obtained from the Pondrigen extreme principle has a dataset size insufficient to reflect the complexity of all changes and trends throughout the entire period. To address this issue, data augmentation methods are needed to generate simulated data for robust deep learning model training. Therefore, this invention addresses the problem of scarce training data through a GAN-based data augmentation method.

[0119] like Figure 3 As shown, a GAN has two neural networks: a generator and a discriminator. The generator aims to produce data that is difficult to distinguish from real data, while the discriminator aims to identify whether the data is real or fake. The ultimate goal is for the generator to continuously optimize its generated data so that the discriminator cannot distinguish it, and for the discriminator to also optimize itself to make its judgments more accurate.

[0120] In this embodiment of the invention, the generator network employs an LSTM (Long Short-Term Memory) network with a ReLU activation function. The generator network learns the joint probability of the input and output data, reconstructs the input data from the output data, and generates new data nodes from the training dataset. The discriminator network employs a multilayer perceptron network with a ReLU activation function. The discriminator distinguishes whether the data comes from the training dataset (real data) or from the generator, outputting a binary decision (0 or 1). Therefore, the objective function of the GAN is obtained:

[0121]

[0122] In the formula, Let x represent the expected value of the distribution of the real data, and z represent random noise. The expected value of the Gaussian noise distribution, D(x) represents the probability that it is real data, G(z) represents the data generated by the generator from the noise z, and D(G(z)) represents the probability that the generated data is real data.

[0123] Using the above objective function, we can obtain a GAN model, and the training process can be regarded as a minimax game process. By synthesizing new optimal state control pairs from the learned distribution, we can solve the problem of scarce training data for neural network control.

[0124] The optimal control state obtained based on the Pondrikin extreme principle has a very limited dataset size, which is not conducive to the training of neural networks. Therefore, GAN (Generative Adversarial Network) is introduced to expand the dataset of the optimal control obtained based on the Pondrikin extreme principle, increase the size of the dataset, and make the trained neural network more robust, so as to enhance the robustness of controlling the spread of vehicle-to-everything (V2X) malware through neural networks.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for controlling the propagation of malicious software in the Internet of Vehicles based on GAN data augmentation, characterized in that, include: S1: Establish a traffic flow model to obtain the expression for vehicle distance under IDM balanced flow conditions; S2: Establish a vehicle-to-vehicle communication channel model to obtain the probability of successfully receiving wireless signals; S3: Establish a vehicle-to-everything (V2X) malware propagation model and obtain the fractional differential equation for V2X malware propagation; S4: The optimal control solution for the fractional differential equation of malware propagation in the Internet of Vehicles is calculated based on the Pondrikin principle. S5: Based on the optimal state control of GAN, the original dataset is expanded to obtain a new dataset; In S3, the infection rate probability is expanded by combining the traffic flow model and the vehicle-to-vehicle communication channel model as follows: Where R represents the vehicle's communication radius; s ij The net distance between vehicle j and its adjacent preceding vehicle i is represented by the fractional differential equation for the propagation of malware in the Internet of Vehicles, which is then derived based on infectious disease dynamics and fractional theory. In S4, by combining the control strategies of immunity and treatment, a new fractional differential equation for the propagation of vehicle-to-everything (V2X) malware is obtained. The Pondrikin extreme value principle is introduced to solve the optimal control problem of the new V2X malware propagation fractional differential equation. Finally, the control quantity that minimizes the objective function J(u1(t),u2(t),u3(t)) of the V2X malware propagation fractional differential equation is obtained. u1(t) represents the treatment rate of roadside devices successfully attacked by malware at time t; u2(t) represents the immunity rate of roadside devices vulnerable to malware attacks at time t; u3(t) represents the treatment rate of mobile devices successfully attacked by malware at time t. In S2, the probability density function of the received wireless signal power is obtained from the distance between the vehicles in S1 as follows: and Where, p t This indicates the wireless signal transmission power of the equipped vehicle; λ represents the wavelength, G t G represents the gain of the transmitting antenna; r f(p) represents the gain of the receiving antenna. r (s ij )) represents the received wireless signal power p r (s ij The probability density function of ). By giving a critical received power p su That is, the signal power received by the vehicle is greater than p su If so, the wireless signal can be successfully transmitted; the probability that the vehicle successfully receives the wireless signal is as follows:

2. The method for controlling the propagation of vehicle-to-everything (V2X) malware based on GAN data augmentation according to claim 1, characterized in that, In S1, the expression for the vehicle distance under the IDM balanced flow state is as follows: Where T represents reaction time, s0 represents the minimum bumper spacing when traffic comes to a complete stop, v represents the vehicle speed in a balanced flow state, v0 represents the desired speed in a free flow, δ represents the acceleration exponent, and s e Indicates the distance between vehicles.

3. The method for controlling the propagation of malicious software in the Internet of Vehicles based on GAN data augmentation according to claim 1, characterized in that, In step S5, a GAN model is obtained through an objective function, and the training process is viewed as a minimax game. New optimal state control pairs are synthesized from the learned distribution of the simulated data. The objective function expression is as follows: in, Let x represent the expected value of the distribution of the real data, and z represent random noise. The expected value of the Gaussian noise distribution, D(x) represents the probability that it is real data, G(z) represents the data generated by the generator from the noise z, and D(G(z)) represents the probability that the generated data is real data.

4. The method for controlling the propagation of vehicle-to-everything (V2X) malware based on GAN data augmentation according to claim 1, characterized in that, In S5, the dataset of the optimal control solution is expanded by introducing GAN, increasing the size of the dataset and making the trained neural network more robust, thereby enhancing the robustness of controlling the spread of vehicle-to-everything (V2X) malware through the neural network.

Images