A privacy protection method, system and electronic device in a vehicle networking network

Abstract

The application relates to the field of smart grid information security and information security technology, and discloses a privacy protection method and system in a vehicle networking network and electronic equipment, the method is applied to a vehicle networking system comprising a central authorization agency and a power grid enterprise, the privacy protection method comprises the following steps: the central authorization agency receives service request information of an electric vehicle; the central authorization agency determines a first privacy budget for perturbing charging / discharging capacity and a second privacy budget for perturbing arrival time and departure time based on the service request information and a current state of the vehicle networking system through a pre-defined perturbation strategy; the central authorization agency adopts a truncated Laplace mechanism to perturb the charging / discharging capacity, and adopts a truncated discrete Laplace mechanism to perturb the arrival time and the departure time respectively; and the central authorization agency uploads data containing the perturbed charging / discharging capacity, the arrival time and the departure time to the power grid enterprise. The method has the advantages of simple steps, high protection flexibility and good protection effect.

Description

Technical Field

[0001] This invention relates to the field of smart grid information security and information security technology, and in particular to a privacy protection method, system and electronic device in a vehicle network. Background Technology

[0002] The rapid development of smart cities and intelligent electric vehicles has driven the widespread application of V2G (Vehicle-to-Grid) interaction systems. V2G technology uses electric vehicles (EVs) as distributed energy storage units to stabilize the power grid through peak shaving and valley filling. In this system, a central authority (CA) needs to collect real-time charging and discharging data from EVs and provide it to the power grid company (GE) for dispatching and allocation. However, this process introduces significant privacy risks: the power grid company can analyze the collected service data to infer sensitive user behavior patterns, such as commuting habits, daily activity locations, and specific travel purposes. Therefore, striking a balance between protecting user privacy and ensuring the data utility required to maintain stable power grid operation is the core challenge facing V2G systems.

[0003] In recent years, the privacy leakage problem in V2G systems has attracted the attention of scholars at home and abroad, but most privacy protection research has certain limitations. At present, some related studies use encryption and blockchain technology to protect privacy, such as ([1] Wang, B.; Yang, Y.; Liu, W.; Xu, LA Privacy-Preserving Scheme for Charging Reservations and Subsequent Deviation Settlements for Electric Vehicles Basedon a Consortium Blockchain. World Electr. Veh. J. 2025, 16, 243. https: / / doi.org / 10.3390 / wevj16050243) which builds a secure privacy environment by integrating consortium blockchain and homomorphic encryption, but this method has huge computational overhead and high maintenance cost.

[0004] Meanwhile, some studies have used pseudonym allocation and session key protocols to protect user privacy. The paper ([2] W.Hou,Y.Sun,D.Li,Z.Guan and J.Liu, "Lightweight and Privacy-PreservingCharging Reservation Authentication Protocol for 5G-V2G," in IEEETransactions on Vehicular Technology,vol.72,no.6,pp. 7871-7883,June 2023,doi:10.1109 / TVT.2023.3241324.) studies a lightweight and scalable reserved authentication and key protocol and embeds a physically unclonable function (PUF) to resist physical attacks. This mechanism takes into account the computing power in V2G devices and effectively achieves privacy protection by using a lightweight processing method. However, the tiny data associations after pseudonymization may still expose the privacy of user behavior patterns.

[0005] Furthermore, differential privacy, as a lightweight technology, provides strong privacy protection and is very suitable for V2G systems. Differential privacy protection technology based on deep reinforcement learning (RL) can effectively achieve lightweight strict privacy protection ([3] S. Lee and D.-H. Choi, "Multilevel DeepReinforcement Learning for Secure Reservation-Based Electric Vehicle Chargingvia Differential Privacy and Energy Storage System," in IEEE Transactions on Vehicular Technology, vol. 73, no. 8, pp. 11097-11109, Aug. 2024, doi:10.1109 / TVT.2024.3372517.). However, in practical scenarios, traditional differential privacy has its limitations. First, the content of V2G system contains discrete time data and continuous power data, and a single DP perturbation method is difficult to handle this heterogeneity. In addition, the unrestricted noise added to ensure privacy will seriously impair the accuracy of the data and may affect the normal operation and scheduling of the power grid. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a privacy protection method, system, and electronic device in a vehicle-to-everything (V2X) network. The steps are simple and effective, real-time disturbance of user service request information aligns with speculative attack patterns targeting service information, thus ensuring the privacy and security of service data.

[0007] The first aspect of this invention provides a privacy protection method in a vehicle-to-everything (V2X) network, applicable to V2X systems including centrally authorized agencies and power grid companies. The privacy protection method includes: The centrally authorized agency receives service request information from electric vehicles, including arrival time, departure time, and charge / discharge amount. Based on service request information and the current state of the vehicle-to-everything (V2X) system, the centrally authorized agency determines a first privacy budget for perturbing the charging / discharging amount and a second privacy budget for perturbing the arrival and departure times through a predefined perturbation strategy. Based on the first privacy budget, the centrally authorized agency uses a truncated Laplace mechanism to perturb the charge / discharge quantities, obtaining the perturbed charge / discharge quantities; Based on the second privacy budget, the central authorized agency uses the truncated discrete Laplace mechanism to perturb the arrival and departure times respectively, and obtains the perturbed arrival and departure times; The centrally authorized agency will upload data including the amount of charge / discharge after the disturbance, arrival time, and departure time to the power grid company.

[0008] Optionally, the current status of the vehicle-to-everything (V2X) system also includes the current electricity price.

[0009] Optionally, the perturbation strategy is a policy model trained through reinforcement learning. The steps for determining the first privacy budget and the second privacy budget include: inputting a state vector constructed based on service request information and the current state of the vehicle network system into the policy model, and having the policy model output an action vector containing the first privacy budget and the second privacy budget.

[0010] Optionally, the policy model is trained based on a dual-delay deep deterministic policy gradient algorithm.

[0011] Optionally, the method for training the policy model includes the following steps: In the simulation environment, the central authorizing agency is treated as an intelligent agent, and the following iterative process is executed: In each training time slot, the status of the vehicle-to-everything (V2X) system is observed and input into the policy network to obtain candidate privacy budgets; The simulated electric vehicle service request information is perturbed and uploaded based on the candidate privacy budget; Instant rewards are calculated based on the privacy protection effect, data utility, and user cost generated by the perturbed data. Store the state, action, reward, and next state as experience. The parameters of the policy network and value network are updated by sampling from experience. Repeat the iteration until the policy converges to obtain the perturbation policy.

[0012] Optionally, the privacy protection effect is evaluated by calculating the KL divergence between the noisy Laplace distribution and the uniform distribution; the data utility is quantified by calculating the error in the user's charging / discharging power before and after the perturbation.

[0013] Optionally, instant rewards are obtained using the following formula: , in, EOPP For the degree of privacy protection, DU For data utility, C For user cost, instant rewards Calculated based on a predefined optimization objective, the optimization objective is a reward. G At its maximum, privacy budget is adjusted through deep reinforcement learning algorithms, enabling... G maximum; When users value privacy ; When users value the utility of data ; When users value the utility of data .

[0014] A second aspect of this invention provides a privacy protection system in a vehicle-to-everything (V2X) network, applied to a V2X system, the V2X system including a central authorization agency, the central authorization agency including: The request receiving module is used to receive service request information from electric vehicles. The strategy decision module stores perturbation policies, which are used to determine the privacy budget based on service request information and the current state of the vehicle network system. The data perturbation module is used to perturb the charge / discharge data using a truncated Laplace mechanism according to the privacy budget, so as to obtain the perturbed charge / discharge data. The arrival time and departure time are perturbed using a truncated discrete Laplace mechanism respectively, so as to obtain the perturbed arrival time and departure time. The communication module is used to upload the disturbed data to the power grid company.

[0015] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0017] The technical solution provided by the embodiments of the present invention has the following advantages compared with the prior art: This invention provides a privacy protection method, system, and electronic device for vehicle-to-everything (V2X) networks. It employs a truncated Laplace mechanism to perturb continuous battery data, a classic differential privacy mechanism suitable for continuous numerical data, which achieves privacy protection by adding noise conforming to a Laplace distribution. A truncated discrete Laplace mechanism is used to perturb discrete time data, specifically designed for integer or discrete data. This better preserves the discrete characteristics of time data and avoids invalid time values ​​due to perturbation. By adding noise to arrival and departure times, the precise correlation between data is broken, making it impossible for attackers to accurately reconstruct the user's true time behavior pattern. Through heterogeneous data processing strategies, the characteristics of different data types are adapted, ensuring privacy while more effectively maintaining data availability and business logic rationality (e.g., time remains an integer). By truncating both the Laplace and discrete Laplace mechanisms, the added noise is limited to a certain range, avoiding extreme perturbation values. Through predefined perturbation strategies, a first privacy budget and a second privacy budget are allocated to battery and time data respectively, applying different levels of protection to data with different sensitivities and avoiding excessive perturbation. It can control noise amplitude, ensuring that the perturbed data remains within a reasonable range, thus achieving a better balance between privacy protection and data utility and reducing the negative impact on the accuracy of power grid dispatch. Employing differential privacy as the core technology, compared to cryptographic methods such as homomorphic encryption, differential privacy performs data perturbation at the central authority (CA), resulting in lower computational overhead and eliminating the need for complex multi-party computations or key management. It requires no changes to the existing communication architecture; the CA directly performs perturbation processing after collecting data before uploading it to the power grid company (GE), simplifying the process and making it easy to deploy. It implements a lightweight and scalable privacy protection scheme suitable for the real-time processing needs of massive electric vehicle data in V2G systems. Through the rigorous mathematical definition of differential privacy, it ensures that even if an attacker possesses some background knowledge, they cannot reliably infer the existence of an individual in the dataset from the perturbed data, thereby protecting user behavioral privacy. It provides strong privacy guarantees, more effectively resisting behavioral pattern analysis based on data associations than simple pseudonymization. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of a privacy protection method and system in a vehicle network provided by an embodiment of the present invention. Detailed Implementation

[0019] The following detailed description of a specific embodiment of the present invention is provided in conjunction with the accompanying drawings. However, it should be understood that the scope of protection of the present invention is not limited to the specific embodiment.

[0020] The present invention will be described below through several specific embodiments. To keep the following description of the embodiments of the present invention clear and concise, detailed descriptions of known functions and known components may be omitted.

[0021] The first part of this invention provides a privacy protection method in a vehicle-to-everything (V2X) network, applicable to V2X systems including centrally authorized agencies and power grid companies. The privacy protection method includes: The centrally authorized agency receives service request information from electric vehicles, including arrival time, departure time, and charge / discharge amount. Based on service request information and the current state of the vehicle-to-everything (V2X) system, the centrally authorized agency determines a first privacy budget for perturbing the charging / discharging amount and a second privacy budget for perturbing the arrival and departure times through a predefined perturbation strategy. Based on the first privacy budget, the centrally authorized agency uses a truncated Laplace mechanism to perturb the charge / discharge quantities, obtaining the perturbed charge / discharge quantities; Based on the second privacy budget, the central authorized agency uses the truncated discrete Laplace mechanism to perturb the arrival and departure times respectively, and obtains the perturbed arrival and departure times; The centrally authorized agency will upload data including the amount of charge / discharge after the disturbance, arrival time, and departure time to the power grid company.

[0022] Optionally, the current status of the vehicle-to-everything (V2X) system also includes the current electricity price.

[0023] Optionally, the perturbation strategy is a policy model trained through reinforcement learning. The steps for determining the first privacy budget and the second privacy budget include: inputting a state vector constructed based on service request information and the current state of the vehicle network system into the policy model, and having the policy model output an action vector containing the first privacy budget and the second privacy budget.

[0024] Optionally, the policy model is trained based on a dual-delay deep deterministic policy gradient algorithm.

[0025] Optionally, the method for training the policy model includes the following steps: In the simulation environment, the central authorizing agency is treated as an intelligent agent, and the following iterative process is executed: In each training time slot, the status of the vehicle-to-everything (V2X) system is observed and input into the policy network to obtain candidate privacy budgets; The simulated electric vehicle service request information is perturbed and uploaded based on the candidate privacy budget; Instant rewards are calculated based on the privacy protection effect, data utility, and user cost generated by the perturbed data. Store the state, action, reward, and next state as experience. The parameters of the policy network and value network are updated by sampling from experience. Repeat the iteration until the policy converges to obtain the perturbation policy.

[0026] Optionally, the privacy protection effect is evaluated by calculating the KL divergence between the noisy Laplace distribution and the uniform distribution; the data utility is quantified by calculating the error in the user's charging / discharging power before and after the perturbation.

[0027] Optionally, instant rewards are obtained using the following formula: , in, EOPP For the degree of privacy protection, DU For data utility, C For user cost, instant rewards Calculated based on a predefined optimization objective, the optimization objective is a reward. G At its maximum, privacy budget is adjusted through deep reinforcement learning algorithms, enabling... G maximum; When users value privacy ; When users value the utility of data ; When users value the utility of data .

[0028] A privacy protection method in a vehicle-to-everything (V2G) network is proposed, based on data security in the V2G system. The Central Authorization Authority (CA), acting as a DRL (Data Transfer Level) agent, uses the arrival and departure times of users (EVs) and their charging / discharging needs as states. The current state is then perturbed and transmitted to the power grid company (GE) using a perturbation strategy based on a dual-delay deep deterministic policy gradient. KLD (Knowledge, Dispersion, and Divergence) divergence is used to assess the degree of privacy protection, and the degree of data perturbation is used to evaluate data validity. Determined by the perturbation noise generated by the Laplace mechanism and These represent disturbances in departure and arrival times, respectively. Disturbances representing electrical charge levels. It is the power distribution error, which is the difference between the power before and after the disturbance.

[0029] , , in, It is the power before the disturbance. It is the power after the disturbance. This is the actual charge and discharge volume. That's the actual departure time. This is the actual arrival time. It is the amount of disturbance due to the departure time. It is the disturbance amount of arrival time. It is the disturbance of electrical quantity. The power before the disturbance is equal to the actual charge / discharge amount divided by the duration of the time. The power after the disturbance is equal to the amount of electricity after the disturbance divided by the duration of the disturbance.

[0030] The greater the perturbation, the lower the data utility, so DU is negative. KLD measures the correlation between two distributions; KLD is a value greater than 0. The closer KLD is to 0, the greater the correlation between the two distributions. Therefore, KLD is used to calculate the correlation between a noisy Laplace distribution and a uniform distribution. The closer the noisy Laplace distribution is to a uniform distribution, the more uniform the perturbation, and the better the privacy protection. Because one is a continuous distribution and the other is a discrete distribution, KLD is divided into the following two types:

[0031] , , in, and These represent the truncated discrete Laplace distribution and the truncated Laplace distribution, respectively. Let be a discrete uniform distribution, and let be a discrete set of points. It is a continuous uniform distribution, and its values ​​are continuous variables.

[0032] , This is sensitivity, usually set to 1. It is a privacy budget, which is learned and selected by the reinforcement learning algorithm. The cutoff length is set to 20. It is the independent variable.

[0033] It is a privacy budget, which is learned and selected by the reinforcement learning algorithm. The cutoff length is set to 20. It is the independent variable.

[0034] The current time-slot system benefit is calculated. This benefit comprises data utility, privacy protection effectiveness, and cost. The specific expression is: The system constructs the state of the next time slot through a Markov transition process; based on user service requests, a dynamic perturbation strategy is adopted to protect user service data privacy; through experience pool replay, the system network state of the current time slot and the next time slot, the dynamic perturbation strategy, and the evaluation results are stored as experience in the experience pool, and the data privacy protection strategy is continuously optimized by updating the weight parameters of the actor network and the critic network in the perturbation strategy; combining the degree of data privacy protection and data validity, the system benefits are obtained. Users continuously update the weight parameters of the actor network and the critic network until a stable perturbation strategy is obtained. The simulation result curve reaches stability and no longer rises or falls, ultimately achieving real-time data privacy protection in the V2G system.

[0035] like Figure 1 As shown, the specific operation steps are as follows: S1. Initialize the V2G system environment settings, based on the perturbation policy parameters of the dual-delay deep deterministic policy gradient network TD3; S2, in time The central authority (CA) of the deep reinforcement learning DRL agent observes the V2G system to obtain service request information from electric vehicles (EVs), including arrival times. Departure time Charge / discharge amount and current electricity prices Construct the current state vector .

[0036] S3 and CA will display the current observation status. Input to policy network Actor network In the middle; the Actor network outputs an action based on the current policy. , It's a privacy budget for controlling disturbance power. This is the privacy budget for controlling the disturbance time. This action represents the privacy budget selected for the continuous charge and discharge data and the discrete arrival and departure time data of the disturbance, respectively. The privacy budget is between 0.5 and 5, and is selected according to the general scenario requirements and the need to weigh the value.

[0037] S4, CA uses the privacy budget selected in step S3. The heterogeneous data of EVs is perturbed using a hybrid differential privacy (DP) mechanism: the discrete arrival and departure times of EVs are perturbed. Disturbance application Truncated Discrete Laplace Mechanism for Privacy Budget , This is sensitivity, usually set to 1. It is a privacy budget, which is learned and selected by the reinforcement learning algorithm. The cutoff length is set to 20. It is the independent variable, that is, adding noise generated by the truncated discrete Laplace mechanism to the real data, affecting the continuous charge and discharge quantities. Disturbance application Truncated Laplace mechanism for privacy budget , in, This is sensitivity, usually set to 1. It is a privacy budget, which is learned and selected by the reinforcement learning algorithm. The cutoff length is set to 20. It is the independent variable.

[0038] S5 and CA upload the perturbed data to the power grid company (GE) and obtain immediate rewards from the V2G environment. The reward The optimization objective, calculated based on a predefined goal, is to maximize the reward G. A deep reinforcement learning algorithm is used to adjust the privacy budget to maximize G, which is then used to comprehensively balance the extent of privacy protection (EOPP), data utility (DU), and user cost (C). These three weighted parameters are derived based on user requirements. For example, if the user prioritizes privacy protection, the EOPP parameter is larger; if the user prioritizes cost, the C parameter is larger; when the user prioritizes privacy... .

[0039] When users value the utility of data When users value data utility, At the same time, the system transitions to the state of the next moment. ;

[0040] S6 and CA will use this "state-action-reward-next state" experience tuple. Store in the experience replay pool R; S6, the state is the state of EV observed at time k obtained in S2. The action is that S3 takes the state input from the policy network and obtains the result. The reward is calculated according to the formula in S5. Then, the state at time k+1 is obtained based on the state transition probability matrix. Treat these physical quantities as an empirical tuple Store it in the experience replay pool R.

[0041] S7. Randomly sample a minimum batch of experience from the experience replay pool R. The minimum batch size is the size used for experience replay in the neural network, generally called batchsize. In this invention, batchsize is set to 128. The target policy network is then used as the next state. Select target action (And noise is added, meaning the TD3 algorithm will add some noise to the action to increase the exploratory nature); then, based on the TD3 dual-commentator structure, the target value is calculated using the one with the smaller Q-value in the two target commentator networks. : ,

[0042] in, It is the target Q-value at time k, obtained from the target critic network. The reward at time k is calculated in S5. It is a discount factor (artificially set) used to weigh the importance of "current rewards" and "future rewards". It is the Q-value obtained by the target commentator network based on the state and action at the next moment.

[0043] S8. Using the gradient descent algorithm, minimize the evaluation values ​​output by the two value networks (Critic1, Critic2). Compared with the target value calculated in step S7 Mean square error between , Update the two value networks separately. and The parameters; where, It is the predicted Q-value of the current critic (value) network. It is the target Q value. The instantaneous loss value of the commentator network; the smaller the instantaneous loss value, the more accurate the prediction. This represents the network parameters for the critics (values). This formula is used to find the network parameters that minimize the loss of batch data.

[0044] S9. According to the delayed update strategy of TD3, every... Each time step The value is typically set between 2 and 10, adjusted based on simulation results, and the optimal value is selected. The deterministic policy gradient, calculated using a formula, is used to update the policy network. Parameters; It is the policy objective function Policy network parameters gradient, It is the average gradient of the batch of samples. Q-value pairs of actions gradient

[0045] S10. After the policy network is updated, use the soft update method. Slowly update the target policy network. and two target value networks , The parameters are adjusted to maintain the stability of the training process; S11, Update the status to Repeat steps S3 to S10; continue this process until the maximum number of episodes is reached. The maximum number of episodes is set manually, and here it is set to 400. Finally, a stable perturbation strategy that can dynamically balance privacy protection and data utility is learned.

[0046] The second part of the embodiments of the present invention provides a privacy protection system in a vehicle network, applied to a vehicle network system, the vehicle network system including a central authorization agency, the central authorization agency including: The request receiving module is used to receive service request information from electric vehicles. The strategy decision module stores perturbation policies, which are used to determine the privacy budget based on service request information and the current state of the vehicle network system. The data perturbation module is used to perturb the charge / discharge data using a truncated Laplace mechanism according to the privacy budget, so as to obtain the perturbed charge / discharge data. The arrival time and departure time are perturbed using a truncated discrete Laplace mechanism respectively, so as to obtain the perturbed arrival time and departure time. The communication module is used to upload the disturbed data to the power grid company.

[0047] The third embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method.

[0048] The fourth embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0049] Example 1: A privacy protection method in vehicle-to-grid (V2G) networks is proposed. This method models the privacy budget allocation problem as a Markov decision process (MDP). A central server acts as an agent, adaptively generating perturbation strategies (such as adjusting privacy budget parameters) based on user charging needs and time requirements. Noise is added to the data before uploading it to the power grid company, and the model is continuously optimized based on feedback from privacy protection effectiveness and data utility.

[0050] The specific steps are as follows: 1. Construct a simulated V2G environment including electric vehicle (EV) users, charging stations, a central server (CA), and the power grid company (GE). Set the hyperparameters of the reinforcement learning algorithm, including the learning rate, discount factor, experience replay buffer size, and batch size. Simultaneously, randomly initialize the weights of the policy network (actor) and the value network (critic).

[0051] 2. The central server (CA) collects data in real time at the current moment. User status information Including arrival time (Time when the user needs to connect to the charging station), departure time (Time when the user needs to pick up the vehicle and leave the charging station), charge / discharge amount (The amount of electricity the user expects to charge or discharge) and the current electricity price. Construct the current state vector .

[0052] 3. The central server (CA) will display the current observation status. Input to actor network In the middle; the actor network outputs an action based on the current strategy. This action represents the privacy budget selected for the continuous charge / discharge data and the discrete arrival and departure time data, respectively. The privacy budget determines how much noise to add to the real data.

[0053] 4. The central server CA is configured according to the privacy budget selected in step 3. A hybrid differential privacy (DP) mechanism is used to add noise to the heterogeneous data of electric vehicles. The calculation formula is as follows: , Noise generated according to a privacy budget. Discrete arrival and departure times of electric vehicles (EVs) Disturbance application Truncated Discrete Laplace Mechanism for Privacy Budget For continuous charge and discharge quantities Disturbance application Truncated Laplace mechanism for privacy budget Then the central service will process the perturbation data ( The data is published and uploaded to the power grid company for power grid dispatching.

[0054] 5. The central server CA will calculate rewards based on the feedback from the V2G environment. The reward Calculated based on predefined optimization objectives, used to comprehensively balance privacy protection effectiveness (EOPP), data utility (DU), and user cost (C).

[0055] 6. CA will store this "state-action-reward-next state" experience tuple. Store the data in the experience replay pool R; sample from the pool and use the gradient descent algorithm to update the weights of the policy network and value network, so that the model evolves in the direction of maximizing long-term cumulative reward.

[0056] The specific process is as follows: [Initialization]: The server has loaded the pre-trained reinforcement learning model.

[0057] [Observation Status]: The user has submitted a charging request to the central server. The arrival time is 7 PM and the departure time is 7 AM the next day. The user needs to charge 30 kWh. (Implicit Status: It is currently the evening peak of the power grid, the electricity price is high, and it is easy to expose Zhang San's work and rest patterns).

[0058] [Policy Generation]: After analyzing the current state, the model concludes that this is a peak period, and directly uploading the actual demand (30kWh) would severely expose user privacy (e.g., revealing that Zhang San just arrived home and someone is home). The policy network outputs actions. (Privacy Budget) (smaller), which means that more noise needs to be added to mask the true intention.

[0059] [Generate Disturbance Information]: The central server, based on... A random noise (e.g., +5.2 kWh, between 7:20 PM and 6:50 AM the next day) is generated using a truncated Laplace mechanism. Data upload: The server reports to the power grid company: "The user's demand is 35.2 kWh, arriving between 7:20 PM and 6:50 AM the next day."

[0060] [Assessment and Feedback]: Privacy Assessment: A third-party attacker attempted to infer whether Zhang San was at home based on the 35.2 kWh data, but the inference failed due to high noise levels. Data Utility: The power grid distributes electricity based on 35.2 kWh, which deviates from the actual demand of 30 kWh by 5.2 kWh, resulting in some scheduling redundancy costs.

[0061] [Model Update]: The model has "learned" that adding more noise is a "high-scoring" strategy during the evening rush hour (7:00 PM) and when the dwell time is long (until 7:30 AM the next day) because it protects privacy very well, even though it sacrifices some accuracy.

[0062] [Loop]: At 02:00 the next day (off-peak electricity period), the model makes another decision. At this time, the policy network may choose the option with less noise (privacy budget). (Large-scale) data is uploaded to help the power grid accurately complete the final charging task, as the risk of privacy breach is low at this time.

[0063] The above inventions are merely a few specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A privacy protection method in a vehicle-to-everything (V2X) network, characterized in that, The privacy protection method, applied to vehicle-to-everything (V2X) systems including those of centrally authorized agencies and power grid companies, includes: The central authorized agency receives service request information from electric vehicles, including arrival time, departure time, and charge / discharge amount. Based on the service request information and the current state of the vehicle network system, the central authorization agency determines a first privacy budget for perturbing the charging / discharging amount and a second privacy budget for perturbing the arrival time and departure time through a predefined perturbation strategy. The central authorized agency, based on the first privacy budget, uses a truncated Laplace mechanism to perturb the charge / discharge amount, thereby obtaining the perturbed charge / discharge amount; The central authorized agency, based on the second privacy budget, uses a truncated discrete Laplace mechanism to perturb the arrival and departure times respectively, to obtain the perturbed arrival and departure times; The central authorized agency will upload data including the charge / discharge amount, arrival time, and departure time after the disturbance to the power grid company.

2. The privacy protection method in the vehicle network as described in claim 1, characterized in that, The current status of the vehicle-to-everything (V2X) system also includes the current electricity price.

3. The privacy protection method in the vehicle network as described in claim 1, characterized in that, The perturbation strategy is a strategy model obtained through reinforcement learning. The step of determining the first privacy budget and the second privacy budget includes: inputting a state vector constructed based on the service request information and the current state of the vehicle network system into the strategy model, and having the strategy model output an action vector containing the first privacy budget and the second privacy budget.

4. The privacy protection method in a vehicle-to-everything (V2X) network as described in claim 3, characterized in that, The policy model is trained based on the dual-delay deep deterministic policy gradient algorithm.

5. The privacy protection method in a vehicle-to-everything (V2X) network as described in claim 3, characterized in that, The method for training the policy model includes the following steps: In the simulation environment, the central authorization agency is treated as an intelligent agent, and the following iterative process is executed: In each training time slot, the status of the vehicle-to-everything (V2X) system is observed and input into the policy network to obtain candidate privacy budgets; The simulated electric vehicle service request information is perturbed and uploaded based on the candidate privacy budget; Instant rewards are calculated based on the privacy protection effect, data utility, and user cost generated by the perturbed data. Store the state, action, reward, and next state as experience. The parameters of the policy network and value network are updated by sampling from experience. Repeat the iteration until the policy converges to obtain the perturbation policy.

6. The privacy protection method in a vehicle-to-everything (V2X) network as described in claim 5, characterized in that, The privacy protection effect is evaluated by calculating the KL divergence between the noisy Laplace distribution and the uniform distribution; the data utility is quantified by calculating the error in the user's charging / discharging power before and after the disturbance.

7. The privacy protection method in a vehicle-to-everything (V2X) network as described in claim 5, characterized in that, The instant reward is obtained using the following formula: ， in, EOPP For the degree of privacy protection, DU For data utility, C For user cost, the instant reward Calculated based on a predefined optimization objective, the optimization objective is a reward. G At its maximum, privacy budget is adjusted through deep reinforcement learning algorithms, enabling... G maximum; When users value privacy ; When users value the utility of data ; When users value the utility of data .

8. A privacy protection system in a vehicle-to-everything (V2X) network, applied to a V2X system, characterized in that, The vehicle networking system includes a central authorization agency, which includes: The request receiving module is used to receive service request information from electric vehicles. The strategy decision module stores perturbation strategies and is used to determine the privacy budget based on the service request information and the current state of the vehicle network system. The data perturbation module is used to perturb the charge / discharge data using a truncated Laplace mechanism according to the privacy budget, to obtain the perturbed charge / discharge data, and to perturb the arrival time and departure time using a truncated discrete Laplace mechanism, to obtain the perturbed arrival time and departure time respectively. The communication module is used to upload the disturbed data to the power grid company.

9. An electronic 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 program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images