Internet of vehicles computing unloading method and system based on information age and privacy protection

Abstract

The invention discloses an Internet of Vehicles computing unloading method and system based on information age and privacy protection, and the method comprises the steps: building an optimization problem with information age and energy consumption as targets through building a vehicle task model and an energy consumption time delay model, and introducing a geographical undistinguishable constraint to protect location privacy, and the real position of the vehicle is disturbed by using a planar Laplace mechanism. And modeling calculation unloading as a Markov decision process, and selecting an unloading strategy by an intelligent agent based on the disturbed position information, environment state and task characteristics. And solving by adopting a GRPO reinforcement learning algorithm, updating the strategy network based on advantage estimation and KL divergence constraint, and obtaining an optimal unloading decision. According to the method, collaborative optimization of privacy security, information freshness and energy efficiency in the Internet of Vehicles is realized through information age modeling, geographical indistinguishability location disturbance and GRPO reinforcement learning.

Description

Technical Field

[0001] This invention relates to the field of vehicle network computing offloading technology, specifically to a vehicle network computing offloading method and system based on information age and privacy protection. Background Technology

[0002] With the rapid development of intelligent transportation and the Internet of Vehicles (IoV), the massive amounts of data and computing tasks generated by vehicles place extremely high demands on computing resources and communication latency. Traditional cloud computing, due to significant transmission delays caused by physical distance, struggles to meet the needs of some tasks in IoV scenarios, especially applications such as autonomous driving and collision warning, which require millisecond-level response and high reliability. Mobile edge computing, by deploying computing, storage, and network resources closer to the network edge of the vehicle, has become a key technology for solving this problem, effectively improving the computing load of in-vehicle devices and enhancing overall efficiency and service reliability.

[0003] However, in actual computational offloading, while traditional performance metrics can measure basic computational and communication efficiency, they are insufficient to fully characterize and optimize the timeliness value of information in high-speed, rapidly changing environments. Simultaneously, computational offloading requires vehicles to disclose their location to edge servers for resource coordination and task allocation, inadvertently opening a significant privacy risk. Malicious attackers or untrusted edge service providers could exploit frequent location uploads to continuously track and map target vehicles, posing a serious threat to users' physical location security and behavioral privacy. Existing research, when optimizing offloading decisions, often over-relies on task latency as an optimization objective, lacking effective modeling and optimization of information age, making it difficult to accurately describe information freshness and make optimal offloading decisions accordingly. Furthermore, while protective measures such as location obfuscation and differential privacy exist to address location privacy leaks, they often face significant challenges in balancing protection strength, implementation costs, and applicability to highly dynamic vehicle environments, making it difficult to maintain high performance and high availability of edge computing services while meeting privacy protection requirements. In addition, selecting appropriate reinforcement learning algorithms for optimizing computational offloading decisions in complex and dynamic environments remains challenging. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a vehicle-to-everything (V2X) computing offloading method and system based on information age and privacy protection. By constructing an optimization model that integrates information age, energy consumption, and privacy leakage parameters, and combining position perturbation based on the planar Laplace mechanism and the GRPO reinforcement learning algorithm, the invention achieves joint optimization of information freshness and system energy efficiency under computing task offloading while ensuring vehicle location privacy.

[0005] On the one hand, vehicle-to-everything (V2X) computing offloading methods based on information age and privacy protection include:

[0006] S1. The current vehicle environment is abstracted into a vehicle network model, and a vehicle task model and a vehicle position disturbance model are constructed. Based on the vehicle task model, a task latency and energy consumption calculation model is constructed. Based on the task latency calculation model, an information age model of the vehicle task is constructed. Based on the information age model and the task energy consumption calculation model, an optimization problem is constructed.

[0007] S2 treats vehicles in the vehicle-to-everything (V2X) model as intelligent agents, defines the computational offloading problem faced by vehicles when generating tasks as a Markov decision process, and uses a vehicle position perturbation model to perturb the vehicle's real position information to generate perturbed vehicle position information during the task offloading process.

[0008] S3. Based on the Markov decision process and vehicle disturbance location information, the GRPO algorithm is used to solve the optimization problem to obtain the optimal task unloading strategy, and the task is unloaded based on the optimal task unloading strategy.

[0009] Furthermore, in S1, the task latency and energy consumption calculation model is used to calculate the total latency and total energy consumption generated in completing the task.

[0010] Total delay in completing the task The calculation formula is as follows:

[0011] ;

[0012] Among them, parameters Used to describe the unloading status of a task; This refers to the processing latency incurred when tasks are processed locally. This indicates the number of CPU cycles required for the current task to be computed locally. This indicates the CPU computing frequency of the local vehicle. The processing latency incurred by the task being computed on the edge server. This indicates the number of CPU cycles required for tasks offloaded to the edge server. This indicates the CPU computing frequency of the edge server; The transmission latency caused by sending tasks to the edge server for computation. Indicates the amount of task data. Indicates the transmission rate;

[0013] Total energy consumption generated in completing the task The calculation formula is as follows:

[0014] ;

[0015] in, For vehicle equipment operating power, Indicates the operating power of the edge server. This refers to the transmission power of the local vehicle.

[0016] Furthermore, in S1, the vehicle position disturbance model satisfies geographical indistinguishability, which specifically means that within a specified range, the vehicle's actual position and the disturbance position are indistinguishable in geographical coordinates.

[0017] Generating disturbance location The probability of satisfies the following formula:

[0018] ;

[0019] in, Privacy budgets that influence the strength of privacy protection The smaller the value, the stronger the privacy protection. The actual location of the vehicle and The Euclidean distance between them; This indicates the actual location of the vehicle. Input the vehicle position disturbance model M to obtain the disturbance position ; This indicates the actual location of the vehicle. Input the vehicle position disturbance model M to obtain the disturbance position ; Represents probability;

[0020] The vehicle position perturbation model uses a planar Laplace mechanism to perturb the vehicle's true position while satisfying geographical indistinguishability. The probability density function of the planar Laplace mechanism is as follows:

[0021] ;

[0022] in, The Euclidean distance between the true location and the perturbation location; The angle between the line connecting the true location and the perturbation location and the polar coordinate axis; These are normalization parameters; This represents the probability density function of the plane Laplace mechanism.

[0023] Furthermore, in S1, an information age model for vehicle tasks is constructed based on the task delay calculation model. Specifically, this includes: assuming the current vehicle-to-everything (V2X) environment is a discrete time-slot system, considering the evolution of information age under zero-wait conditions, and deriving the formula based on the task delay model to obtain the information age of the [missing information]. Information age for each task ,in This represents the total delay in completing the nth task.

[0024] Furthermore, in S1, the optimization problem is defined as follows:

[0025] ;

[0026] in, and This indicates the weighting of information age and energy consumption in a balanced manner; This indicates that a constraint is imposed on the latency of the task. This indicates the maximum time constraint for processing this task. It is a constant representing the penalty term when the task's latency exceeds the task's maximum constraint time; A privacy leakage parameter is defined to characterize the extent to which disturbed locations compromise vehicle location privacy.

[0027] Furthermore, in S2, the computational offloading problem faced by the vehicle when generating a task is defined as a Markov decision process. The Markov decision process specifically includes:

[0028] The agent interacts with the environment, observes the current state of the environment, selects and executes an action from the action space, and then receives a corresponding reward. The environment then transitions to the state of the next moment with a certain probability.

[0029] The formula for calculating the reward function is as follows:

[0030] ;

[0031] in, Represents the action space; This represents the reward function.

[0032] Furthermore, in S2, the specific steps for generating perturbed vehicle position information by perturbing the vehicle's true position information using the vehicle position perturbation model are as follows:

[0033] S21, from uniform distribution The sampling angle is used as the angle of the perturbation position in the polar coordinate system. ;

[0034] S22, from uniform distribution Generate random numbers ;

[0035] S23, make the random number generated in step S22 Equal to the radius of the disturbance position in polar coordinates cumulative distribution function ;

[0036] S24, Equation obtained based on cumulative distribution function. And the radius of the disturbance location in the polar coordinate system is obtained. ;

[0037] S25, based on the angle of the disturbance location in polar coordinates and radius Generate disturbance location information ;in, Indicates the location of the disturbance; Indicates the initial position.

[0038] Furthermore, in S3, the GRPO algorithm is used to solve the optimization problem based on the Markov decision process and vehicle disturbance location information to obtain the optimal task offloading strategy, specifically including:

[0039] S301, Initialize the policy network for the GRPO algorithm. Obtain vehicle network environment status information ;

[0040] S302, Execute actions based on the current time sequence of tasks and environmental states. And receive a reward ;

[0041] S303, store the current state, action, reward and next state into the experience replay pool;

[0042] S304, Update current state ;

[0043] S305, if the experience replay pool is full, the number of samples per group is... The action is recorded as Otherwise, execute S302;

[0044] S306, calculate each sampled action based on the reward function. Rewards ;

[0045] S307, normalize the reward for the set of actions performed by the agent at time sequence t, and the agent performs the actions. The advantage reward obtained is expressed as ,in, This represents the average reward within the group. The standard deviation of the group's rewards;

[0046] S308, Calculate the current policy network The policy network obtained from the previous training round Unbiased estimation of KL divergence between ;

[0047] S309, Update the objective function of the GRPO algorithm, the calculation formula is as follows:

[0048] ;

[0049] in, Describe the objective function. Indicates the policy network parameters; For hyperparameters; This indicates the current state based on the strategy obtained in the previous round. The agent performs actions The probability of; Indicates the current state based on the current round's strategy. The agent performs actions The probability of; Represents the mathematical expectation; Indicates the number of sampling actions. Indicates the size of the set. This indicates that the parameters are being clipped; These are parameters used to control the cutting range;

[0050] S310, Calculate the gradient of the objective function. ;

[0051] S311, Update the policy network parameters. , The learning rate;

[0052] S312, determine whether the total number of training rounds has been reached. If the total number of training rounds has been reached, output the optimal unloading strategy; otherwise, reset the environment and execute S302.

[0053] On the other hand, vehicle-to-everything (V2X) computing offloading systems based on information age and privacy protection include:

[0054] The optimization problem construction module abstracts the current vehicle environment into a vehicle network model and constructs a vehicle task model and a vehicle position disturbance model. Based on the vehicle task model, a task latency and energy consumption calculation model is constructed. Based on the task latency calculation model, an information age model of the vehicle task is constructed. Based on the information age model and the task energy consumption calculation model, the optimization problem is constructed.

[0055] The perturbation module treats vehicles in the vehicle-to-everything (V2X) model as intelligent agents, defines the computational offloading problem faced by vehicles when generating tasks as a Markov decision process, and uses a vehicle position perturbation model to perturb the vehicle's real position information to generate perturbed vehicle position information during the task offloading process.

[0056] The unloading module uses the GRPO algorithm based on Markov decision process and vehicle disturbance location information to solve the optimization problem, obtain the optimal task unloading strategy, and unload the task based on the optimal task unloading strategy.

[0057] The present invention adopts the above technical solution and has the following beneficial effects:

[0058] (1) By constructing an information age model and a task delay and energy consumption calculation model for vehicle tasks, this invention can effectively evaluate the task processing efficiency and energy consumption under different unloading strategies. By combining the optimization problems of information age and energy consumption, it can achieve effective utilization of resources and minimization of energy consumption while ensuring service quality.

[0059] (2) The present invention adopts a vehicle position perturbation model that satisfies geographical indistinguishability, which ensures the privacy of the vehicle’s real position information during transmission. By perturbing the vehicle’s real position through the planar Laplace mechanism, it not only increases the difficulty for attackers to obtain accurate position information, but also meets the needs of privacy protection, providing users with stronger location privacy protection.

[0060] (3) Based on the application of Markov Decision Process (MDP) and GRPO algorithm, this invention enables the system to make the optimal computational offloading decision dynamically according to the current environmental state, taking into account not only the immediate task requirements, but also the long-term performance goals. Attached Figure Description

[0061] Figure 1 This is a flowchart of a vehicle-to-everything (V2X) computing offloading method based on information age and privacy protection, according to an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of a vehicle-to-everything (V2X) environment according to an embodiment of the present invention;

[0063] Figure 3 This is an information age model diagram for a task in an embodiment of the present invention;

[0064] Figure 4 This is a flowchart illustrating the training process of the GRPO algorithm according to an embodiment of the present invention.

[0065] Figure 5 This is a diagram of a vehicle-to-everything (V2X) computing offloading system based on information age and privacy protection, according to an embodiment of the present invention. Detailed Implementation

[0066] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0067] like Figure 1 As shown, the present invention provides a vehicle-to-everything (V2X) computing offloading method based on information age and privacy protection, comprising:

[0068] S1. The current vehicle environment is abstracted into a vehicle network model, and a vehicle task model and a vehicle position disturbance model are constructed. Based on the vehicle task model, a task latency and energy consumption calculation model is constructed. Based on the task latency calculation model, an information age model of the vehicle task is constructed. Based on the information age model and the task energy consumption calculation model, an optimization problem is constructed.

[0069] The task latency and energy consumption calculation model is used to calculate the total latency and total energy consumption generated in completing a task.

[0070] Total delay in completing the task The calculation formula is as follows:

[0071] ;

[0072] Among them, parameters Used to describe the unloading status of a task; This refers to the processing latency incurred when tasks are processed locally. This indicates the number of CPU cycles required for the current task to be computed locally. This indicates the CPU computing frequency of the local vehicle. The processing latency incurred by the task being computed on the edge server. This indicates the number of CPU cycles required for tasks offloaded to the edge server. This indicates the CPU computing frequency of the edge server; The transmission latency caused by sending tasks to the edge server for computation. Indicates the amount of task data. Indicates the transmission rate;

[0073] Total energy consumption generated in completing the task The calculation formula is as follows:

[0074] ;

[0075] in, For vehicle equipment operating power, Indicates the operating power of the edge server. This refers to the transmission power of the local vehicle.

[0076] Specifically, in S1, the vehicle position disturbance model satisfies geographic indistinguishability, which means that within a specified range, the vehicle's actual position and the disturbance position are indistinguishable in geographic coordinates.

[0077] Generating disturbance location The probability of satisfies the following formula:

[0078] ;

[0079] in, Privacy budgets that influence the strength of privacy protection The smaller the value, the stronger the privacy protection. The actual location of the vehicle and The Euclidean distance between them; This indicates the actual location of the vehicle. Input the vehicle position disturbance model M to obtain the disturbance position ; This indicates the actual location of the vehicle. Input the vehicle position disturbance model M to obtain the disturbance position ; Represents probability;

[0080] The vehicle position perturbation model uses a planar Laplace mechanism to perturb the vehicle's true position while satisfying geographical indistinguishability. The probability density function of the planar Laplace mechanism is as follows:

[0081] ;

[0082] in, The Euclidean distance between the true location and the perturbation location; The angle between the line connecting the true location and the perturbation location and the polar coordinate axis; These are normalization parameters; This represents the probability density function of the plane Laplace mechanism.

[0083] Specifically, in S1, an information age model for vehicle tasks is constructed based on the task latency calculation model. This includes: assuming the current vehicle-to-everything (V2X) environment is a discrete time-slot system, considering the evolution of information age under zero-wait conditions, and deriving the formula based on the task latency model to obtain the information age of the [missing information]. Information age for each task ,in This represents the total delay in completing the nth task.

[0084] Specifically, in S1, the optimization problem is defined as follows:

[0085] ;

[0086] in, and This indicates the weighting of information age and energy consumption in a balanced manner; This indicates that a constraint is imposed on the latency of the task. This indicates the maximum time constraint for processing this task. It is a constant representing the penalty term when the task's latency exceeds the task's maximum constraint time; A privacy leakage parameter is defined to characterize the extent to which disturbed locations compromise vehicle location privacy.

[0087] Specifically, in S1, the set of vehicles in the vehicle-to-everything (V2X) environment is set as , the set of edge server nodes is , and the set of tasks generated by the vehicles is denoted as . Each vehicle task is represented as , consisting of the CPU revolutions required to process the task, the amount of data contained in the task, and the maximum time constraint for processing the task; the transmission rate between the vehicles and edge servers involved in the unloading process is . ,in For transmission channel bandwidth, The vehicle's signal transmission power. This represents the channel gain between the vehicle and the edge server. The Gaussian white noise power of the transmission channel between the vehicle and the edge server; tasks generated locally by the vehicle can be processed locally on the vehicle or offloaded to the edge server for processing. Processing is done on the vehicle side, requiring no data transmission during the process. Processing latency is limited by the number of CPU cycles required for the current task to be computed locally. and local vehicles CPU computing frequency Impact, Obtaining the Task Processing latency caused by local processing for Energy consumption is affected by the operating power of vehicle equipment. The processing latency impact of tasks being computed locally, then the task The processing energy consumption generated by local computation is If the task Offloading processing to edge servers requires consideration of latency and energy consumption during the transmission process. Processing latency is affected by the number of CPU cycles required for the task computation on the offloaded edge server. and edge servers CPU computing frequency Impact, task Processing latency caused by computation on edge servers Represented as: ,Task The process of offloading to the edge server is affected by the amount of task data. and transmission rate The impact is on the transmission latency of the task. for ;Task Processing is offloaded to edge servers, where energy consumption is affected by the operating power of the edge servers. The processing latency impact of tasks computed on edge servers, then the task The processing power consumption generated by computing on the edge server is... The energy consumption generated during the process of offloading tasks to edge servers is affected by the transmission power of the local vehicle. And the impact of transmission latency caused by transmitting tasks to edge servers, tasks Transmission energy consumption for Total delay in completing the task for Total energy consumption generated in completing the task Represented as , where parameters Used to describe the unloading status of a task. This indicates that the task is calculated locally on the vehicle. This indicates that the task is offloaded to an edge server for processing.

[0088] Specifically, the vehicle position perturbation model uses a planar Laplace mechanism to perturb the vehicle's true position while satisfying geographical indistinguishability. The probability density function of the one-dimensional Laplace mechanism is transformed into the probability density function of the planar Laplace mechanism in polar coordinates as follows:

[0089] ;

[0090] in, The Euclidean distance between the true location and the perturbation location. From the formula Seeking, For the -1 branch of the Lambert W function, the parameter , The angle between the line connecting the true location and the perturbation location and the polar coordinate axis. , These are normalization parameters;

[0091] To characterize the extent to which disturbed locations compromise vehicle location privacy, a privacy leakage parameter is defined. ,in , and For parameters, Let the Euclidean distance between the perturbation location and the true location be within a radius of . The expectation within the circular region is given by the formula have to, The cosine similarity between the perturbation position and the true position at a radius of The expectation within the circular region is given by the formula Furthermore, based on the task latency calculation model, an information age model for vehicle tasks is constructed. Assuming the current vehicle-to-everything (V2X) environment is a discrete time-slot system, considering the information age evolution under zero-wait conditions, the information age at any given time is obtained. Information age is ,in The latest status update time is represented as , To complete the processing of the first The time to update the status after each task, the first The information age of each task is given by the formula calculate.

[0092] like Figure 2 As shown in this embodiment, the vehicle generates a large amount of data that requires computation and processing. This data is modeled as tasks, which can be processed locally or offloaded to an edge server. If offloaded to the edge server, the vehicle generating the task needs to be within the communication range of the edge server, and the agent needs to perturb the vehicle's position during the offloading process. That is, a point is selected as the perturbation location within the perturbation area to reduce the possibility of location privacy leakage.

[0093] S2 treats vehicles in the vehicle-to-everything (V2X) model as intelligent agents, defines the computational offloading problem faced by vehicles when generating tasks as a Markov decision process, and uses a vehicle position perturbation model to perturb the vehicle's real position information to generate perturbed vehicle position information during the task offloading process.

[0094] Specifically, in S2, the computational offloading problem faced by the vehicle when generating a task is defined as a Markov decision process. The Markov decision process specifically includes:

[0095] The agent interacts with the environment, observes the current state of the environment, selects and executes an action from the action space, and then receives a corresponding reward. The environment then transitions to the state of the next moment with a certain probability.

[0096] The formula for calculating the reward function is as follows:

[0097] ;

[0098] in, Represents the action space; This represents the reward function.

[0099] Specifically, the steps for generating perturbed vehicle position information by perturbing the vehicle's true position information using a vehicle position perturbation model are as follows:

[0100] S21, from uniform distribution The sampling angle is used as the angle of the perturbation position in the polar coordinate system. ;

[0101] S22, from uniform distribution Generate random numbers ;

[0102] S23, make the random number generated in step S22 Equal to the radius of the disturbance position in polar coordinates cumulative distribution function ;

[0103] S24, Equation obtained based on cumulative distribution function. And the radius of the disturbance location in the polar coordinate system is obtained. ;

[0104] S25, based on the angle of the disturbance location in polar coordinates and radius Generate disturbance location information ;in, Indicates the location of the disturbance; Indicates the initial position.

[0105] Specifically, in S2, the task unloading problem is modeled as a Markov decision process, which mainly includes the current time series environmental state. Action space reward function and the state of the next time sequence The state space of the current time series is defined as follows: ,in, For the current time series, The task for processing the current time series is represented as , The calculation frequency is denoted as the local CPU frequency of the vehicle, and the calculation frequency of the edge server is denoted as... , generate tasks vehicles The Euclidean distance between the edge server and the edge server is denoted as . , and These represent the operating power and transmission power of the vehicle's local equipment, respectively. The operating power of the edge server is denoted as... The action space of the previous time series is represented as: , This indicates that the task is processed locally on the vehicle. This indicates that the task is offloaded from the local vehicle to the edge server for computation; the reward function is defined as follows: The Markov decision process is as follows: the agent interacts with the state of the environment, performs actions and receives corresponding rewards, and the environment will change to the state of the next moment with a certain probability.

[0106] like Figure 3 As shown, an information age model for vehicle tasks is constructed based on a task latency calculation model. Assuming the current vehicle-to-everything (V2X) environment is a discrete time-slot system, the evolution of information age under zero-wait conditions is considered. The information age increases linearly and decreases sharply when the state is updated, resulting in the information age at any given time. Information age is ,in The latest status update time is represented as , To complete the processing of the first The time to update the status after each task, the first The information age of each task is given by the formula calculate.

[0107] S3. Based on the Markov decision process and vehicle disturbance location information, the GRPO algorithm is used to solve the optimization problem to obtain the optimal agent unloading strategy, and the task is unloaded based on the optimal task unloading strategy.

[0108] Specifically, such as Figure 4 As shown, the GRPO algorithm is used to solve the optimization problem based on Markov decision process and vehicle disturbance location information to obtain the optimal task offloading strategy, which specifically includes:

[0109] S301, Initialize the policy network for the GRPO algorithm. Obtain vehicle network environment status information ;

[0110] S302, Execute actions based on the current time sequence of tasks and environmental states. And receive a reward ;

[0111] S303, store the current state, action, reward and next state into the experience replay pool;

[0112] S304, Update current state ;

[0113] S305, if the experience replay pool is full, the number of samples per group is... The action is recorded as Otherwise, execute S302;

[0114] S306, calculate each sampled action based on the reward function. Rewards ;

[0115] S307, normalize the reward for the set of actions performed by the agent at time sequence t, and the agent performs the actions. The advantage reward obtained is expressed as ,in, This represents the average reward within the group. The standard deviation of the group's rewards;

[0116] S308, Calculate the current policy network The policy network obtained from the previous training round Unbiased estimation of KL divergence between ;

[0117] S309, Update the objective function of the GRPO algorithm, the calculation formula is as follows:

[0118] ;

[0119] in, Describe the objective function. Indicates the policy network parameters; For hyperparameters; This indicates the current state based on the strategy obtained in the previous round. The agent performs actions The probability of; Indicates the current state based on the current round's strategy. The agent performs actions The probability of; Represents the mathematical expectation; Indicates the number of sampling actions. Indicates the size of the set. This indicates that the parameters are being clipped; These are parameters used to control the cutting range;

[0120] S310, Calculate the gradient of the objective function. ;

[0121] S311, Update the policy network parameters. , The learning rate;

[0122] S312, determine whether the total number of training rounds has been reached. If the total number of training rounds has been reached, output the optimal unloading strategy; otherwise, reset the environment and execute S302.

[0123] In summary, this invention employs a geographic indistinguishability mechanism, using a planar Laplace mechanism to perturb the vehicle's true location while satisfying geographic indistinguishability, thus avoiding location privacy leaks during the unloading process. Introducing information age better characterizes the timeliness of information. When optimizing the algorithm's policy network, the algorithm adopts an intra-group relative reward mechanism and a lightweight single-network architecture, utilizing KL divergence to constrain the magnitude of policy updates, thereby reducing training costs while ensuring training stability.

[0124] like Figure 5 As shown, this embodiment also discloses a vehicle-to-everything (V2X) computing offloading system based on information age and privacy protection, including:

[0125] The optimization problem construction module 51 abstracts the current vehicle environment into a vehicle network model and constructs a vehicle task model and a vehicle position disturbance model. Based on the vehicle task model, it constructs a task latency and energy consumption calculation model. Based on the task latency calculation model, it constructs a vehicle task information age model. Based on the vehicle task information age model and the task energy consumption calculation model, it constructs an optimization problem.

[0126] The perturbation module 52 treats the vehicles in the vehicle-to-everything (V2X) model as intelligent agents, defines the computational offloading problem faced by the vehicles when generating tasks as a Markov decision process, and uses the vehicle position perturbation model to perturb the real position information of the vehicles to generate perturbed position information of the vehicles during the task offloading process.

[0127] The unloading module 53 uses the GRPO algorithm based on the Markov decision process and vehicle disturbance location information to solve the optimization problem, obtain the optimal task unloading strategy, and unload the task based on the optimal task unloading strategy.

[0128] The specific implementation of the vehicle network computing offloading method and system based on information age and privacy protection is the same as that of the vehicle network computing offloading method based on information age and privacy protection, and will not be described again in this embodiment.

[0129] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A vehicle-to-everything (V2X) computing offloading method based on information age and privacy protection, characterized in that, Includes the following steps: S1. The current vehicle environment is abstracted into a vehicle network model, and a vehicle task model and a vehicle position disturbance model are constructed. Based on the vehicle task model, a task latency and energy consumption calculation model is constructed. Based on the task latency calculation model, an information age model of the vehicle task is constructed. Based on the information age model and the task energy consumption calculation model, an optimization problem is constructed. S2 treats vehicles in the vehicle-to-everything (V2X) model as intelligent agents, defines the computational offloading problem faced by vehicles when generating tasks as a Markov decision process, and uses a vehicle position perturbation model to perturb the vehicle's real position information to generate perturbed vehicle position information during the task offloading process. S3. Based on the Markov decision process and vehicle disturbance location information, the GRPO algorithm is used to solve the optimization problem to obtain the optimal task unloading strategy, and the task is unloaded based on the optimal task unloading strategy.

2. The vehicle network computing offloading method based on information age and privacy protection according to claim 1, characterized in that, In S1, the task latency and energy consumption calculation model is used to calculate the total latency and total energy consumption generated in completing the task. Total delay in completing the task The calculation formula is as follows: ； Among them, parameters Used to describe the unloading status of a task; This refers to the processing latency incurred when tasks are processed locally. This indicates the number of CPU cycles required for the current task to be computed locally. This indicates the CPU computing frequency of the local vehicle. The processing latency incurred by the task being computed on the edge server. This indicates the number of CPU cycles required for tasks offloaded to the edge server. This indicates the CPU computing frequency of the edge server; The transmission latency caused by sending tasks to the edge server for computation. Indicates the amount of task data. Indicates the transmission rate; Total energy consumption generated in completing the task The calculation formula is as follows: ； in, For vehicle equipment operating power, Indicates the operating power of the edge server. This refers to the transmission power of the local vehicle.

3. The vehicle network computing offloading method based on information age and privacy protection according to claim 1, characterized in that, In S1, the vehicle position disturbance model satisfies the geographical indistinguishability, which specifically means that within a specified range, the vehicle's actual position and the disturbance position are indistinguishable in geographical coordinates. Generating disturbance location The probability of satisfies the following formula: ； in, Privacy budgets that influence the strength of privacy protection The smaller the value, the stronger the privacy protection. The actual location of the vehicle and The Euclidean distance between them; This indicates the actual location of the vehicle. Input the vehicle position disturbance model M to obtain the disturbance position ; This indicates the actual location of the vehicle. Input the vehicle position disturbance model M to obtain the disturbance position ; Represents probability; The vehicle position perturbation model uses a planar Laplace mechanism to perturb the vehicle's true position while satisfying geographical indistinguishability. The probability density function of the planar Laplace mechanism is as follows: ； in, The Euclidean distance between the true location and the perturbation location; The angle between the line connecting the true location and the perturbation location and the polar coordinate axis; These are normalization parameters; This represents the probability density function of the plane Laplace mechanism.

4. The vehicle network computing offloading method based on information age and privacy protection according to claim 1, characterized in that, In S1, an information age model for vehicle tasks is constructed based on the task delay calculation model. Specifically, this includes: assuming the current vehicle-to-everything (V2X) environment is a discrete time-slot system, considering the evolution of information age under zero-wait conditions, and deriving the formula based on the task delay model to obtain the information age of the [missing information]. Information age for each task ,in This represents the total delay in completing the nth task.

5. The vehicle network computing offloading method based on information age and privacy protection according to claim 4, characterized in that, In S1, the optimization problem is defined as follows: ； in, and This indicates the weighting of information age and energy consumption in a balanced manner; This indicates that a constraint is imposed on the latency of the task. This indicates the maximum time constraint for processing this task. It is a constant representing the penalty term when the task's latency exceeds the task's maximum constraint time; A privacy leakage parameter is defined to characterize the extent to which disturbed locations compromise vehicle location privacy.

6. The vehicle network computing offloading method based on information age and privacy protection according to claim 5, characterized in that, In S2, the computational offloading problem faced by the vehicle when generating a task is defined as a Markov decision process. The Markov decision process specifically includes: The agent interacts with the environment, observes the current state of the environment, selects and executes an action from the action space, and then receives a corresponding reward. The environment then transitions to the state of the next moment with a certain probability. The formula for calculating the reward function is as follows: ； in, Represents the action space; This represents the reward function.

7. The vehicle network computing offloading method based on information age and privacy protection according to claim 1, characterized in that, In S2, the specific steps for generating perturbed vehicle position information by perturbing the vehicle's true position information using the vehicle position perturbation model are as follows: S21, from uniform distribution The sampling angle is used as the angle of the perturbation position in the polar coordinate system. ; S22, from uniform distribution Generate random numbers ; S23, make the random number generated in step S22 Equal to the radius of the disturbance position in polar coordinates cumulative distribution function ; S24, Equation obtained based on cumulative distribution function. And the radius of the disturbance location in the polar coordinate system is obtained. ; S25, based on the angle of the disturbance location in polar coordinates and radius Generate disturbance location information ;in, Indicates the location of the disturbance; Indicates the initial position.

8. The vehicle network computing offloading method based on information age and privacy protection according to claim 1, characterized in that, In S3, the GRPO algorithm is used to solve the optimization problem based on the Markov decision process and vehicle disturbance location information to obtain the optimal task unloading strategy, specifically including: S301, Initialize the policy network for the GRPO algorithm. Obtain vehicle network environment status information ; S302, Execute actions based on the current time sequence of tasks and environmental states. And receive a reward ; S303, store the current state, action, reward and next state into the experience replay pool; S304, Update current state ; S305, if the experience replay pool is full, the number of samples per group is... The action is recorded as Otherwise, execute S302; S306, calculate each sampled action based on the reward function. Rewards ; S307, normalize the reward for the set of actions performed by the agent at time sequence t, and the agent performs the actions. The advantage reward obtained is expressed as ,in, This is the average of the rewards within the group. The standard deviation of the group's rewards; S308, Calculate the current policy network The policy network obtained from the previous training round Unbiased estimation of KL divergence between ; S309, Update the objective function of the GRPO algorithm, the calculation formula is as follows: ； in, Describe the objective function. Indicates the policy network parameters; For hyperparameters; This indicates the current state based on the strategy obtained in the previous round. The agent performs actions The probability of; Indicates the current state based on the current round's strategy. The agent performs actions The probability of; Represents the mathematical expectation; Indicates the number of sampling actions. Indicates the size of the set. This indicates that the parameters are being clipped; These are parameters used to control the cutting range; S310, Calculate the gradient of the objective function. ; S311, Update the policy network parameters. , The learning rate; S312, determine whether the total number of training rounds has been reached. If the total number of training rounds has been reached, output the optimal unloading strategy; otherwise, reset the environment and execute S302.

9. A vehicle-to-everything (V2X) computing offloading system based on information age and privacy protection, characterized in that, include: The optimization problem construction module abstracts the current vehicle environment into a vehicle network model and constructs a vehicle task model and a vehicle position disturbance model. Based on the vehicle task model, a task latency and energy consumption calculation model is constructed. Based on the task latency calculation model, an information age model of the vehicle task is constructed. Based on the information age model and the task energy consumption calculation model, the optimization problem is constructed. The perturbation module treats vehicles in the vehicle-to-everything (V2X) model as intelligent agents, defines the computational offloading problem faced by vehicles when generating tasks as a Markov decision process, and uses a vehicle position perturbation model to perturb the vehicle's real position information to generate perturbed vehicle position information during the task offloading process. The unloading module uses the GRPO algorithm based on Markov decision process and vehicle disturbance location information to solve the optimization problem, obtain the optimal task unloading strategy, and unload the task based on the optimal task unloading strategy.

Images