A vehicle task scheduling method resistant to prior privacy attacks
By dividing time slots in the vehicle edge computing environment and optimizing the offloading strategy using Lyapunov optimization and Markov approximation algorithms, the problem of vehicle location privacy leakage is solved, and system cost is minimized while privacy is protected.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FUJIAN NORMAL UNIV
- Filing Date
- 2023-06-09
- Publication Date
- 2026-07-21
AI Technical Summary
In vehicle edge computing environments, attackers can infer a vehicle's location and driving trajectory by monitoring its access patterns, leading to location privacy leaks. Furthermore, existing technologies struggle to optimize system costs while protecting privacy.
An on-board task scheduling method is adopted, which calculates location privacy loss and system cost by dividing time slots, optimizes the offloading strategy by using Lyapunov optimization and Markov approximation algorithm, and designs an objective function to minimize long-term system cost and privacy loss.
It effectively protects vehicle location privacy and minimizes system energy consumption and computing latency while ensuring privacy, thereby optimizing vehicle system costs.
Smart Images
Figure CN116782202B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of mobile edge computing, privacy protection and optimization algorithms, and in particular to an in-vehicle task scheduling method that resists prior privacy attacks. Background Technology
[0002] With the development of intelligent transportation systems and intelligent vehicles, vehicles are being given more functions, such as autonomous driving and pattern recognition, while also generating a large amount of data that needs to be processed. However, traditional vehicles cannot provide the computing and storage resources required to process this data, especially when facing latency-sensitive and highly complex applications. How to perform rapid computation is a huge challenge. Mobile edge computing, as a new research field that reduces transmission distance and the burden of cloud computing, integrates with traditional vehicle networks to bring available computing resources in the cloud closer to the end vehicle, providing vehicles with low-latency and high-bandwidth computing services.
[0003] While the combination of mobile edge computing and vehicle-to-everything (V2X) networks improves user experience, it also presents a location privacy issue in V2X environments. To achieve better communication performance, task requesters tend to offload computation to the nearest edge server. Attackers can exploit this offloading pattern by monitoring the location of the access server to infer the requester's true location, leading to the leakage of their privacy information. Although some researchers have addressed this issue, their studies generally focus on privacy protection at single locations, neglecting the impact of long-term accumulated access information on current location leakage.
[0004] Over a historical period, information related to vehicles accessing edge servers while driving in urban areas is stored on the servers. The more times a vehicle accesses an edge server, the more access information is stored, known as prior information. Prior information includes the accessing vehicle ID, access time, and task. Attackers can use this information to infer the vehicle's past movement trajectory and unloading pattern, and predict the vehicle's current driving pattern based on these patterns. Furthermore, attackers can calculate the vehicle user's decision preferences at each location. Generally, vehicles tend to choose corresponding unloading strategies at fixed locations during task requests, thus creating a correlation between location and unloading strategy. The higher the frequency of implementing the corresponding strategy, the stronger this correlation. Therefore, attackers can predict more accurate vehicle locations based on historical vehicle trajectories and unloading strategies. Based on the above analysis, this paper studies vehicle location protection from two perspectives. However, since protecting vehicle location increases the overall system's energy consumption and reduces computational efficiency, achieving a combined optimization of these three factors through task scheduling is challenging. Summary of the Invention
[0005] The purpose of this invention is to provide an in-vehicle task scheduling method that resists prior privacy attacks and minimizes system costs while ensuring vehicle location privacy.
[0006] The technical solution adopted in this invention is:
[0007] A method for vehicle-mounted task scheduling that resists prior privacy attacks includes the following steps:
[0008] Step 1: Divide the current vehicle's travel time into equal-length time slots. Where Γ is the set of time slots and T is the number of time slots; the vehicle unloading strategy for time slot t is defined as z(t), z(t)∈Z={z1, z2,..., z M}, where Z is the set of unloading strategies and M is the total number of schedulable strategies;
[0009] Step 2: Calculate the location privacy loss Loss(t) caused by prior information in the vehicle-to-everything (V2X) mobile edge computing in each time slot t. The location privacy loss Loss(t) consists of the location privacy loss Loss1(t) caused by access prior information and the privacy loss Loss2(t) caused by offloading strategy prior information.
[0010] Step 3: Calculate the system cost Cost(t) in each time slot t. Cost(t) includes the delay cost Delay(t) and the energy cost Energy(t).
[0011] Step 4: Establish the system objective function;
[0012] Step 5: Transform the objective function into a real-time privacy cost-pair stability control problem based on Lyapunov optimization; use the Markov approximation algorithm to calculate the optimal decision for the current time slot t. .
[0013] Furthermore, as a further improvement of the present invention, the location privacy loss Loss(t):
[0014]
[0015] in,
[0016]
[0017]
[0018]
[0019]
[0020]
[0021] Where Q = {Q1,..,Q} N} represents the set of vehicle trajectories guessed by the attacker, where N is the number of guessed trajectories; p(Q f ) represents guessing the trajectory Q in time slot t. f The probability that the trajectory is exactly the true trajectory; L=[l1, l2, ..., l t [l] represents the published trajectory of the vehicle at the cutoff time slot t. t The function h(X) represents the position point; the function h(X) represents the trajectory X after removing the last position point; c(q) represents the trajectory X after removing the last position point. ft ,l t ) represents q ft and l t The loss function between two locations, q ft Indicates guessing Q f The t-th position; when q ft =l t At that time, q ft The value of q is 0, otherwise q ft The value is 1; p(z(t)) represents the probability of invoking strategy z(t).
[0022] Furthermore, the specific steps of step 4 are as follows:
[0023] Step 41: Define long-term location privacy loss constraints: Introduce ε to represent time slot t i The privacy loss threshold must be met, and the average privacy loss over a long period of time must be less than this value:
[0024]
[0025] Step 42: Minimize the long-run system cost, with objective function P1:
[0026]
[0027]
[0028] Where x(z(t))∈{0,1}, As a constraint, it means that only one offloading strategy can be selected for time slot t.
[0029] Furthermore, as a further improvement of the present invention, step 5 includes the following steps:
[0030] Step 51: Define the virtual queue as a size exceeding the prior privacy threshold, assuming an initial queue backlog of 0, as follows:
[0031]
[0032] Where Q(t) is the length of the virtual queue at time slot t, representing the privacy cost that exceeds the threshold when the task is completed in time slot t;
[0033] Step 52: Define the quadratic Lyapunov function, as follows:
[0034]
[0035] in, A scalar measure of congestion in the privacy cost queue, when A very small value means that the queue backlog is also small;
[0036] Introducing the conditional Lyapunov drift function, it is expressed as follows:
[0037]
[0038] Here, the drift function Δ(Θ(t)) represents the change in the access privacy cost queue from one time slot to the next Lyapunov function.
[0039] Step 53: Incorporate queue stability into the computational cost by defining a drift-plus-penalty function:
[0040]
[0041] Here, V is a non-negative penalty parameter.
[0042] Step 54: Transform the problem into minimizing the upper bound of the real-time drift plus penalty function, that is, minimizing the right side of the following formula:
[0043]
[0044] That is, solve problem P2:
[0045]
[0046] Using D(z(t) i Replace the objective function of problem P2 with )) and transform problem P2 into the following problem:
[0047]
[0048]
[0049] in, This represents the probability that the unloading strategy z(t) is adopted in the current time slot t.
[0050] Introducing the long-sum-exp function y β (x):
[0051]
[0052] Problem P2 is transformed into an approximate optimization problem P3 with redundant entropy terms, as follows:
[0053]
[0054]
[0055] Here, β is a positive constant that affects the accuracy of the approximation.
[0056] The optimal solution to problem P3 is obtained based on the KKT conditions;
[0057]
[0058] Step 55: The formula for the state transition probability is shown below:
[0059]
[0060] in, To transform from policy z to policy The state transition probability, σ is a positive constant, and β is a positive constant that affects the approximation accuracy; ;
[0061] Step 56: Input the task set Task and available server set S for time slot t; randomly assign a computing server to each task to initialize the policy z(t); while the system has not converged, execute the following loop:
[0062] First, a task is randomly selected, and then another computing server is randomly chosen to generate a new offload strategy for that task. Calculate And calculate the state transition probability. Update the uninstallation strategy The smallest recorded so far The process is repeated until the system converges; finally, the system outputs the calculated optimal strategy. .
[0063] This invention employs the above technical solutions, designs a method for measuring location privacy loss caused by prior information, proposes a system cost model for vehicle edge computing, and designs an objective function. Since solving the long-term optimization problem is computationally very complex, it is transformed into a real-time queue stability control problem based on Lyapunov optimization. This remains an NP-hard problem; therefore, a Markov approximation algorithm is used to find an approximate optimal solution in constant time, achieving minimization of long-term system cost. This invention not only effectively protects vehicle location privacy but also meets the requirement of minimizing long-term vehicle system cost. Attached Figure Description
[0064] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments;
[0065] Figure 1 This is a schematic diagram of the task scheduling model of the present invention;
[0066] Figure 2 This is a flowchart illustrating the steps of the in-vehicle task scheduling method for resisting prior privacy attacks according to the present invention.
[0067] Figure 3 This is a flowchart of the Markov approximation algorithm of the present invention;
[0068] Figure 4 This is an algorithm diagram based on the Markov approximation algorithm of this invention. Detailed Implementation
[0069] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.
[0070] like Figures 1 to 4 As shown in the figure, this invention discloses an in-vehicle task scheduling method that resists prior privacy attacks, used to solve the problem of allocating computing tasks between the vehicle's local machine and edge servers during vehicle operation, including the following steps:
[0071] Step 1: Divide the current vehicle's travel time into equal-length time slots. Where Γ is the set of time slots and T is the number of time slots; the vehicle unloading strategy for time slot t is defined as z(t), z(t)∈Z={z1, z2,..., z M}, where Z is the set of unloading strategies and M is the total number of schedulable strategies;
[0072] Step 2: In each time slot t, calculate the location privacy loss Loss(t) caused by prior information in the vehicle-to-everything (V2X) mobile edge computing. It consists of the location privacy loss Loss1(t) caused by access prior information and the privacy loss Loss2(t) caused by offloading policy prior information.
[0073] in,
[0074]
[0075]
[0076]
[0077]
[0078]
[0079] Where Q = {Q1,..,Q} N} represents the set of vehicle trajectories guessed by the attacker, where N is the number of guessed trajectories; p(Q f ) represents guessing the trajectory Q in time slot t. f The probability that the trajectory is exactly the true trajectory; L=[l1, l2, ..., l t [l] represents the published trajectory of the vehicle at the cutoff time slot t. t The position point is represented by h(X). The function h(X) represents the remaining trajectory of trajectory X after removing the last position point; c(q) represents the position point. ft ,l t ) represents q ft and l t The loss function between two locations, q ft Indicates guessing Q f The t-th position; when q fi =l t When the value is 0, its value is 0; otherwise, it is 1. p(z(t)) represents the probability of invoking strategy z(t).
[0080] Step 3: Calculate the system cost Cost(t) in each time slot t.
[0081] Step 4: Establish the system objective function.
[0082] Step 4 includes the following steps:
[0083] Step 41: Define long-term location privacy loss constraints: Introduce ε to represent the privacy loss threshold for time slot t. The average privacy loss over a long period must be less than this value.
[0084]
[0085] Step 42: Minimize the long-run system cost, with objective function P1:
[0086]
[0087]
[0088] Where x(z(t))∈{0,1}, As a constraint, it means that only one offloading strategy can be selected for time slot t.
[0089] Step 5: Transform the objective function into a real-time privacy cost-pair column stability control problem based on Lyapunov optimization. Solve the problem using the Markov approximation algorithm to calculate the current time slot t. i Optimal decision .
[0090] Step 5 includes the following steps:
[0091] Step 51: Define the virtual queue as a size exceeding the prior privacy threshold, assuming an initial queue backlog of 0, as follows:
[0092]
[0093] Where Q(t) is the length of the virtual queue at time slot t, representing the privacy cost that exceeds when the task is completed in time slot t.
[0094] Step 52: Define the quadratic Lyapunov function, as follows:
[0095]
[0096] in, A scalar measure of congestion in the privacy cost queue, when A very small value means that the queue backlog is also small.
[0097] Introducing the conditional Lyapunov drift function, it is expressed as follows:
[0098]
[0099] Here, the drift function Δ(Θ(t)) represents the change in the access privacy cost queue from one time slot to the next Lyapunov function.
[0100] Step 53: Incorporate queue stability into the computational cost by defining a drift-plus-penalty function:
[0101]
[0102] Here, V is a non-negative penalty parameter.
[0103] Step 54: Transform the problem into minimizing the upper bound of the real-time drift plus penalty function, that is, minimizing the right side of the following formula:
[0104]
[0105] That is, solve problem P2:
[0106]
[0107] Replace the objective function of problem P2 with D(z(t)) and transform problem P2 into the following problem:
[0108]
[0109]
[0110] in, This represents the probability that the unloading strategy z(t) is adopted in the current time slot t.
[0111] Introducing the long-sum-exp function y β (x):
[0112]
[0113] Problem P2 is transformed into an approximate optimization problem P3 with redundant entropy terms, as follows:
[0114]
[0115]
[0116] Here, β is a positive constant that affects the accuracy of the approximation.
[0117] The optimal solution to problem P3 is obtained based on the KKT conditions;
[0118]
[0119] Step 55: The formula for the state transition probability is shown below:
[0120]
[0121] Where σ is a positive constant.
[0122] Step 56: Input the task set Task for time slot t and the available server set S. Randomly assign a computing server to each task to initialize the policy z(t). If the system does not converge, execute the following loop: First, randomly select a task, then randomly select another computing server for that task to generate a new offloading policy. Calculate And calculate the state transition probability. Update the uninstallation strategy The smallest recorded so far The process continues until the system converges. Finally, the system output shows the optimal policy calculated. .
[0123] This invention employs the above technical solutions, designs a method for measuring location privacy loss caused by prior information, proposes a system cost model for vehicle edge computing, and designs an objective function. Since solving the long-term optimization problem is computationally very complex, it is transformed into a real-time queue stability control problem based on Lyapunov optimization. This remains an NP-hard problem; therefore, a Markov approximation algorithm is used to find an approximate optimal solution in constant time, achieving minimization of long-term system cost. This invention not only effectively protects vehicle location privacy but also meets the requirement of minimizing long-term vehicle system cost.
[0124] Obviously, the described embodiments are only a part of the embodiments of this application, not all of them. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. The components of the embodiments of this application described and illustrated herein can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
Claims
1. A vehicle-mounted task scheduling method resistant to prior privacy attacks, characterized in that: It includes the following steps: Step 1: Divide the current vehicle's travel time into equal-length time slots. Where Γ is the set of time slots and T is the number of time slots; the vehicle unloading strategy for time slot t is defined as z(t), z(t)∈Z={z1, z2,..., z M }, where Z is the set of unloading strategies and M is the total number of schedulable strategies; Step 2: In each time slot t, calculate the location privacy loss Loss(t) caused by prior information in the vehicle-to-everything (V2X) mobile edge computing. The location privacy loss Loss(t) consists of the location privacy loss Loss1(t) caused by access prior information and the privacy loss Loss2(t) caused by offloading policy prior information. The expression for the location privacy loss Loss(t) is as follows: ； in, ； ； ； ； ； Where Q = {Q1,..,Q} N } represents the set of vehicle trajectories guessed by the attacker, where N is the number of guessed trajectories; p(Q f ) represents guessing the trajectory Q in time slot t. f The probability that the trajectory is exactly the true trajectory; L=[l1, l2, ..., l t [l] represents the published trajectory of the vehicle at the cutoff time slot t. t The function h(X) represents the position point; the function h(X) represents the trajectory X after removing the last position point; c(q) represents the trajectory X after removing the last position point. ft ,l t ) represents q ft and l t The loss function between two locations, q ft Indicates guessing Q f The t-th position; p(z(t)) represents the probability of invoking strategy z(t); Step 3: Calculate the system cost Cost(t) in each time slot t. Cost(t) includes the delay cost Delay(t) and the energy cost Energy(t). Step 4: Establish the system objective function; Step 5: Transform the objective function into a real-time privacy cost-pair stability control problem based on Lyapunov optimization; use the Markov approximation algorithm to calculate the optimal decision for the current time slot t. .
2. The in-vehicle task scheduling method for resisting prior privacy attacks according to claim 1, characterized in that: When q ft =l t At that time, q ft The value of q is 0, otherwise q ft The value is 1.
3. The in-vehicle task scheduling method resisting prior privacy attacks according to claim 1, characterized in that: The specific steps for step 4 are as follows: Step 41: Define long-term location privacy loss constraints: Introduce ε to represent the privacy loss threshold for time slot t. The average privacy loss over a long period must be less than this value. ； Step 42: Minimize the long-run system cost, with objective function P1: ； ； Where x(z(t))∈{0,1}, As a constraint, it means that only one offloading strategy can be selected for time slot t.
4. The in-vehicle task scheduling method resisting prior privacy attacks according to claim 1, characterized in that: Step 5 includes the following steps: Step 51: Define the virtual queue as a size exceeding the prior privacy threshold, assuming an initial queue backlog of 0, as follows: ； Where Q(t) is the length of the virtual queue at time slot t, represents the privacy cost exceeded when the task is completed at time slot t, and ε represents the privacy loss threshold at time slot t; Step 52: Define the quadratic Lyapunov function, as follows: ； in, A scalar measure of congestion in the privacy cost queue. The value is directly proportional to the queue backlog; Introducing the conditional Lyapunov drift function, it is expressed as follows: ； Here, the drift function Δ(Θ(t)) represents the change in the access privacy cost queue of the Lyapunov function from one time slot to the next time slot; Step 53: Incorporate queue stability into the computational cost by defining a drift-plus-penalty function: ； Where V is a non-negative penalty parameter; Step 54: Transform the problem into minimizing the upper bound of the real-time drift plus penalty function, that is, minimizing the right side of the following formula: ； Step 55: The formula for the state transition probability is shown below: ； in, To transform from policy z to policy The state transition probability, σ is a positive constant, and β is a positive constant that affects the approximation accuracy; ; Step 56: Input the task set Task and available server set S for time slot t; randomly assign a computing server to each task to initialize the policy z(t); while the system has not converged, execute the following loop: First, a task is randomly selected, and then another computing server is randomly selected for that task to generate a new offload strategy. Calculate Then, the state transition probability is calculated. Update the uninstallation strategy The smallest recorded so far The process is repeated until the system converges; finally, the system outputs the calculated optimal strategy. .
5. The in-vehicle task scheduling method resisting prior privacy attacks according to claim 4, characterized in that: Step 54 transforms the problem into minimizing the upper bound of the real-time drift plus penalty function, specifically including the following steps: Step 54-1, solve problem P2: ； Using D(z(t) i Replace the objective function of problem P2 with )) and transform problem P2 into the following problem: ； ； in, This represents the probability that the offloading strategy z(t) is adopted in the current time slot t; Introducing the long-sum-exp function y β (x): ； Step 54-2: Transform problem P2 into an approximate optimization problem P3 with a redundant entropy term. The expression for P3 is as follows: ； ； Where β is a positive constant that affects the accuracy of the approximation; Step 54-3: Obtain the optimal solution to problem P3 based on the KKT conditions; 。