A task offloading method and system for vehicle heterogeneous services

Abstract

This invention discloses a task offloading method and system for heterogeneous vehicle services, relating to the field of computer system architecture technology. On a large time scale, service deployment and content deployment decisions are determined based on the storage capacity constraints of roadside units. On a small time scale, task offloading and computational resource allocation decisions for heterogeneous vehicles are determined based on these decisions. A weighted index that integrates content service assurance and computational offloading gain is introduced to measure reliability loss. An energy consumption deficit queue and a latency overflow queue are established. A K-step Lyapunov drift reduction framework is used to reconstruct the optimization problem into decomposable subproblems, and these subproblems on different time scales are solved alternately until convergence. After updating the queue state, the system moves to the next time slot. This invention, employing the above-mentioned task offloading method and system for heterogeneous vehicle services, can reduce system energy consumption and latency while ensuring service reliability.

Description

Technical Field

[0001] This invention relates to the field of computer system architecture technology, and in particular to a task offloading method and system for heterogeneous services in vehicles. Background Technology

[0002] The development of 6G technology has spurred numerous computationally intensive and latency-sensitive applications, such as autonomous driving, collision warning systems, and dynamic path planning engines, placing higher demands on in-vehicle terminals with limited computing resources. By combining vehicle-to-everything (V2X) computing with mobile edge computing, vehicles can offload computing tasks to nearby roadside units (RSUs) for processing, effectively reducing latency caused by local computing or remote cloud computing. Existing research on task offloading in vehicular edge computing (VEC) largely assumes that vehicles generate only a single type of computing task, failing to fully consider the heterogeneity of task requests. With the continuous emergence of in-vehicle entertainment applications such as in-vehicle video and virtual reality, the demand for content delivery in V2X is increasing, and content requests are gradually becoming as important as computing requests. Furthermore, due to the high-speed mobility of vehicles, some research has begun to introduce service migration mechanisms to dynamically deploy the computing services required by the vehicle to appropriate RSUs. For computing requests and content requests, it is typically necessary to deploy corresponding computing service instances and content caches separately to support the processing of vehicle tasks. However, RSUs have limited storage and computing resources. How to reasonably coordinate the deployment of computing services and content caching under resource constraints to effectively support the heterogeneous task requests of vehicles remains a key issue that needs to be addressed.

[0003] When a vehicle moves from the communication coverage area of ​​one Roadside Unit (RSU) to that of another, a communication access handover process is inevitable. During this process, if relevant services or content caches fail to migrate to the target RSU in a timely manner, task execution will be interrupted or delayed, leading to service performance degradation or even resource loss. Furthermore, the comprehensive benefits of dynamic service and content deployment strategies under different network conditions and vehicle movement scenarios lack effective quantitative evaluation metrics. Existing research generally does not establish clear metrics for measuring performance losses during migration and access handover. In addition, existing solutions often focus on short-term gain optimization, optimizing energy consumption and latency only from a single time slot or short-term perspective. While this may achieve some performance improvement in specific scenarios, it lacks adaptability in highly dynamic and complex network environments, making it difficult to guarantee the long-term stability and sustainable performance optimization of the system. Summary of the Invention

[0004] The purpose of this invention is to provide a task offloading method and system for heterogeneous vehicle services, thereby solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a task offloading method for heterogeneous vehicle services, comprising the following steps: S1. On a large time scale, determine service deployment decisions and content deployment decisions based on the storage capacity constraints of roadside units; S2. At a small time scale, determine the task offloading decision and computing resource allocation decision for vehicle heterogeneity based on service deployment decisions and content deployment decisions; S3. Introduce the vehicle service collaborative benefit evaluation function CHSD, which integrates content service assurance and calculation offloading gain, to measure the reliability loss of service and content migration. S4. Establish an energy consumption deficit queue and a time delay overflow queue. Use the K-step Lyapunov drift penalty reduction framework to refactor the optimization problem into decomposable subproblems. Solve the subproblems of different time scales alternately until convergence. Update the queue state and enter the next time slot.

[0006] Preferably, S1 includes: S11. Linearly relax the integer programming problem of service deployment and content deployment; S12. The random rounding algorithm based on Chernoff bounds is used to solve the problem and obtain the integer deployment decision that satisfies the storage capacity constraint.

[0007] Preferably, the vehicle heterogeneous task in S2 includes a computation request and a content request; The computational requests require computing resources from the roadside unit or the cloud to complete the processing. Content requests require the roadside unit to pre-cache the requested content file and respond with the transmission.

[0008] Preferably, S2 includes: S21. Solve the integer programming problem using the linear relaxation plus random rounding method for the computation request and the content request respectively, and obtain the task unloading decision. S22. Allocate computing resources, establish the Lagrange function, derive the closed-form solution using the KKT conditions, and solve for the optimal Lagrange multiplier through binary search.

[0009] Preferably, solving for the optimal Lagrange multipliers in step S22 includes: When Lagrange multipliers At that time, the optimal allocation of computing resources is ,in It is the extreme point under the corresponding piecewise condition; otherwise, , which represents the maximum available computing resources for the roadside unit; The termination precision of the binary search is set to The maximum number of iterations is set to 200.

[0010] Preferably, in step S3, the vehicle service collaborative benefit evaluation function CHSD is constructed, and its expression is: ; in, This is a dynamic bias parameter used to adjust the optimization focus between content service assurance and computational task processing based on the current vehicle network service status. This represents the content service assurance level, which measures the ability of edge nodes to provide real-time support for vehicle content access requests within a time slot. Its value is determined by the ratio of the number of successfully responded content requests to the total number of content requests. The calculation of the offloading gain is used to reflect the latency optimization benefit of the task after collaborative offloading compared to the pure local execution mode. Its value is defined as the normalized difference between the baseline local execution latency and the actual completion latency. The vehicle service collaborative benefit evaluation function CHSD is used to uniformly characterize the collaborative optimization effect of cache resource configuration, service deployment decisions and task offloading strategies in the vehicle heterogeneous service scenario, and measure the reliability loss.

[0011] Preferably, the energy-deficient queue in S4 and delayed overflow queue The update method is as follows: ; ; in, For time slots The actual energy consumption inside, As a long-term budget ceiling for energy consumption, For vehicles The actual delay As the upper limit of the extended budget, The number of time slots contained within a time frame.

[0012] Preferably, the alternating solution in S4 includes: At the beginning of each time frame, S1 is executed first to obtain deployment decisions on a large time scale, and then S2 is executed to obtain offloading and resource allocation decisions on a small time scale. Substitute the result of S2 back into S1 to re-optimize the deployment decision, and repeat this process iteratively until the change in the objective function value is less than the preset threshold.

[0013] A system for a task offloading method for heterogeneous vehicle services includes a roadside unit, a remote cloud server, and a controller; Multiple roadside units are provided, and each roadside unit is equipped with storage resources and computing resources; The remote cloud server is communicatively connected to the roadside unit; The controller is deployed on the roadside unit or the remote cloud; The controller is configured to: On a large time scale, service deployment decisions and content deployment decisions are determined based on the storage capacity constraints of roadside units; on a small time scale, vehicle heterogeneous task offloading decisions and computing resource allocation decisions are determined based on the service deployment decisions and content deployment decisions; the CHSD index is introduced to measure reliability loss; a queue is established and an alternating solution is performed using a K-step Lyapunov drift reduction penalty framework.

[0014] Preferably, the controller is used to perform linear relaxation and Chernoff bound-based random rounding algorithm to solve deployment decisions at the beginning of each time frame; and to perform Lagrange duality and binary search to solve computational resource allocation within each time frame.

[0015] Therefore, the above-mentioned task offloading method and system for vehicle heterogeneous services has the following beneficial effects: (1) On a large time scale, this method establishes a service and content deployment framework based on storage sensitivity and uses it as the basic framework for task unloading in that time frame; on a small time scale, it dynamically formulates task unloading and resource allocation methods based on the service and content deployment framework to continuously adapt to the high-speed movement of vehicles and meet the needs of heterogeneous services.

[0016] (2) This method introduces the vehicle service collaborative benefit evaluation function CHSD, which integrates the content service guarantee degree and the calculation of the offloading gain degree, and sets a dual weight factor, which, together with the long-term energy consumption-delay constraint, serves as a system stability and reliability constraint, and can more accurately measure the reliability loss caused by service / content migration and high mobility.

[0017] (3) This method models the task unloading problem as a multi-stage mixed integer nonlinear programming problem with two time scales. The problem is decomposed by the improved Lyapunov method, and the optimal solution is obtained by alternating the solution of the problem from different time scales using a hybrid algorithm with two time scales.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a system framework diagram of a task offloading method and system for heterogeneous vehicle services according to the present invention; Figure 2 This is the dual-timescale model of the present invention; Figure 3 This is an iterative diagram of the dual-timescale algorithm of the present invention; Figure 4This is an iterative diagram of the small-timescale algorithm of the present invention; Figure 5 For the different invention Changes in energy consumption and latency; Figure 6 For the different invention Changes in performance indicators; Figure 7 To demonstrate the convergence of energy consumption and time delay in this invention; Figure 8 These are the performance metrics for different algorithms of this invention; Figure 9 A comparison of the energy consumption and latency of different algorithms in this invention; Figure 10 Energy efficiency convergence of different algorithms in this invention; Figure 11 This refers to the time-delay convergence of different algorithms in this invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0022] Please see Figures 1-11 This invention provides a task offloading method for heterogeneous vehicle services, comprising the following steps: S1. On a large time scale, specifically at the beginning of each time frame, the controller performs a joint deployment decision on candidate service instances and content objects based on the remaining storage resource status of each roadside unit, thereby forming a storage capacity-aware deployment strategy oriented towards the heterogeneous service needs of vehicles. The service deployment decision and content deployment decision must satisfy the following storage resource constraints: ; in, RSU Storage capacity, and .when and At that time, they respectively represent services With content In time frame Deployed at RSU Otherwise, it is not deployed. By performing the above joint deployment decision at the time frame level, the roadside unit's ability to support dynamic vehicle service requests can be improved under limited storage resources, providing a resource foundation for subsequent small-scale task offloading and scheduling.

[0023] The specific steps of S1 include: S11. Linearly relax the integer programming problem of service deployment and content deployment; S12. The random rounding algorithm based on Chernoff bounds is used to solve the problem and obtain the integer deployment decision that satisfies the storage capacity constraint.

[0024] S2. At a small time scale (each time slot), determine the task offloading decision and computing resource allocation decision for vehicle heterogeneity based on service deployment decisions and content deployment decisions.

[0025] In heterogeneous service scenarios within vehicles, business requests generated by in-vehicle terminals are categorized into two types based on resource requirements: computational processing tasks and content retrieval tasks. Computational processing tasks require computing power from roadside unit edge computing nodes or cloud computing nodes to offload and execute the task. Content retrieval tasks rely on the roadside unit's pre-deployment and caching capabilities for target content objects, responding through an edge content distribution mechanism. To ensure that a single business request has a unique service attribute within the same decision-making cycle, the following request type constraints are established: ; The specific steps of S2 include: S21. For computation requests and content requests, solve integer programming problems using linear relaxation and random rounding methods to obtain task unloading decisions.

[0026] S22. For computational resource allocation, establish the Lagrange function, derive the closed-form solution using the KKT conditions, and solve for the optimal Lagrange multipliers using a binary search. The solution for the optimal Lagrange multipliers includes: When Lagrange multipliers At that time, the optimal allocation of computing resources is ,in It is the extreme point under the corresponding piecewise condition; otherwise, This represents the maximum available computing resources for the roadside unit.

[0027] The termination precision of the binary search is set to The maximum number of iterations is set to 200.

[0028] S3. Introduce the vehicle service collaborative benefit evaluation function CHSD, which integrates content service assurance and calculation offloading gain, and dynamically adjust parameters. An adaptive trade-off is made between the optimization contributions of computing services and content services to uniformly measure the overall performance of the current service deployment and task offloading strategies.

[0029] Content service guarantee and calculate unloading gain Calculated using the following formula: ; ; in, Indicates the vehicle in the time slot The baseline task processing latency when using a fully local processing mode. This indicates the actual task completion delay under the current joint scheduling strategy. Indicates time slot The set of vehicles that generate content requests within the system, and This indicates the number of content requests successfully responded to by at least one roadside unit. Used to characterize the edge node's ability to provide immediate support for vehicle content access requests. This is used to characterize the normalized latency gains of the collaborative unloading mechanism compared to the purely local processing method.

[0030] The Vehicle Service Collaboration Benefit Evaluation Function (CHSD) is used to uniformly characterize the collaborative optimization effect of cache resource configuration, service deployment decisions, and task offloading strategies in heterogeneous vehicle service scenarios, and to measure reliability loss. The Vehicle Service Collaboration Benefit Evaluation Function (CHSD) is defined as follows: ; in, This is a dynamic bias parameter used to adjust the optimization focus between content service assurance and computational task processing based on the current vehicle network service status. When the vehicle network is in a content-intensive request state, it can improve... To enhance the impact of cache resource allocation on the decision-making process; when the vehicle network is in a computationally intensive state, to reduce To enhance the role of computational unloading benefits in joint optimization; This represents the content service assurance level, which measures the ability of edge nodes to provide real-time support for vehicle content access requests within a time slot. Its value is determined by the ratio of the number of successfully responded content requests to the total number of content requests. This represents the calculated offloading gain, used to reflect the latency optimization benefit of a task after cooperative offloading relative to the pure local execution mode. Its value is defined as the normalized difference between the baseline local execution latency and the actual completion latency. When When this condition is met, it indicates that the current resource deployment strategy has reached its theoretical optimal state, meaning that all content requests are effectively responded to, and all computing tasks achieve maximum latency optimization benefits. By combining storage resource constraints and computing resource constraints, collaborative optimization of service deployment strategies and task offloading strategies can be achieved, thereby improving the overall system performance in heterogeneous vehicle service scenarios.

[0031] S4. Establish an energy consumption deficit queue and a time delay overflow queue. Ensure the sustainability of the system through long-term constraints. Use the K-step Lyapunov drift reduction penalty framework to refactor the optimization problem into decomposable subproblems. Solve the subproblems of different time scales alternately until convergence. Update the queue state and enter the next time slot.

[0032] The long-term constraint is: ; ; In task-intensive regions, uneven loads can lead to system instability, resulting in high energy consumption and long latency. Compared to short-term controls, long-term constraints on energy consumption and latency help keep overall consumption within acceptable limits, ensuring system sustainability and supporting real-time response.

[0033] Energy loss queue and delayed overflow queue The update method is as follows: ; ; in, For time slots The actual energy consumption inside, As a long-term budget ceiling for energy consumption, For vehicles The actual delay As the upper limit of the extended budget, The number of time slots contained within a time frame.

[0034] Alternating solution involves: at the beginning of each time frame, first execute S1 to obtain deployment decisions on a large time scale, then execute S2 to obtain offloading and resource allocation decisions on a small time scale; substitute the result of S2 back into S1 to re-optimize the deployment decisions, and repeat this iterative process until the change in the objective function value is less than a preset threshold.

[0035] The system described above, which is a task offloading method for heterogeneous vehicle services, includes a roadside unit, a remote cloud server, and a controller.

[0036] Multiple roadside units are set up, and each roadside unit is equipped with storage and computing resources; a remote cloud server communicates with the roadside units; the controller is deployed on the roadside units or the remote cloud.

[0037] The controller is configured to: determine service deployment and content deployment decisions based on the storage capacity constraints of roadside units at a large time scale; determine vehicle heterogeneous task offloading and computing resource allocation decisions based on service deployment and content deployment decisions at a small time scale; introduce the CHSD metric to measure reliability loss; establish a queue and use a K-step Lyapunov drift reduction penalty framework for alternating solutions.

[0038] The controller is also used to perform linear relaxation and Chernoff bound-based random rounding algorithms to solve deployment decisions at the beginning of each time frame; and to perform Lagrange duality and bisection search to solve computational resource allocation within each time slot.

[0039] Example 1 Please see Figure 1 The application scenarios for this method can be: This embodiment considers a Vehicle-Road-Cloud Convergence System (VRCIS) consisting of vehicles, Roadside Units (RSUs), and a remote cloud. Consider a densely trafficked scenario consisting of four orthogonal bidirectional roads, each with eight lanes and a length of 800 m. This includes a remote cloud server, four RSUs, and 240 vehicles. The four RSUs are deployed at four intersections, each with a coverage radius of 200 m. In this architecture, moving vehicles may leave the coverage area of ​​one RSU and enter the coverage area of ​​another. This embodiment assumes that each vehicle can generate two types of requests: compute requests and content requests, and that time is divided into large-scale time frames and small-scale time slots.

[0040] Based on this scenario, this embodiment proposes a dual-time-scale precision sensing energy consumption-delay control scheme. The dual-time-scale model is shown below. Figure 2 .

[0041] 1) Large time scale. Within a time frame Initially, RSU will deploy services and content caching based on storage resources and reliability metrics. The energy consumption and latency of its service deployment can be expressed as: ; ; in, and These represent the unit deployment energy consumption and latency cost, respectively. RSU In time frame The energy consumption and latency of the initial deployment can be expressed as: ; ; 2) Small time scale. Within a time frame. The total energy consumption of the system can be expressed as: ; Wherein, represents the energy consumption for local vehicle processing of computing tasks, represents the energy consumption for the vehicle to offload computing tasks to the RSU, represents the energy consumption for the RSU to process vehicle computing tasks, represents the energy consumption for the remote cloud to process vehicle computing tasks, and represents the energy consumption for the RSU to transmit content to the vehicle. Correspondingly, the total vehicle latency can be expressed as: ; To enhance the system's adaptability to network dynamics and high loads, a long-term constraint is introduced, the expression of which is: ; ; However, the above problems are difficult to solve directly. Therefore, by introducing the Lyapunov optimized architecture, this embodiment defines an energy-loss queue. and delay overflow queue The task scheduling problem is modeled in its final form: ; ; Wherein, C1 is the vehicle request constraint, C2 is the RSU storage capacity constraint, C3' and C4' are the energy consumption and time extension constraints, respectively, C5 is the computing resource constraint, and C6 is the RSU communication range constraint.

[0042] See Figure 3 This embodiment addresses the original problem. The problem is decomposed into two subproblems at different time scales, and solved alternately until convergence. The system state is then updated, and the process moves to the next time slot. Given a small-scale variable... and In the case of large time-scale variables, and By dividing the problem into subproblems, we can obtain subproblems on a large time scale. : ; ; question The goal is to jointly optimize service deployment at the beginning of each time frame. and content deployment Given a large timescale variable and In the case of small time-scale variables, and By dividing the problem into smaller time scales, we can obtain subproblems. : ; ; The following section provides solutions for each decision variable at different time scales.

[0043] 1) Large time scale. This is to address integer programming problems. This embodiment fixes the small-scale variable to obtain an approximate optimal solution for the large-scale variable. Simultaneously, this method introduces a random rounding technique based on Chernoff bounds to ensure constraint awareness and theoretically supported adaptive sampling. The specific process is as follows: First, the decision variables and Relaxation is performed, and the relaxed problem is solved in polynomial time using a linear programming solver. This yields continuous solutions to the problem. Next, the probability of target failure is set to... The sample size was determined based on the Chernoff bound. This embodiment performs... Several independent experiments were conducted to obtain the optimal solution to the problem. Finally, for each experiment... This allows us to obtain the total storage resource usage. And only retain those that meet the requirements. Feasible experiments. For each feasible experiment By assessing the problem To calculate its objective value, the optimal solution is obtained by selecting the experiment with the smallest objective value. and .

[0044] 2) Small time scale. For a given resource allocation decision... This embodiment solves the computation request decision respectively. Content request decision Subsequently, resource allocation decisions are further optimized based on the obtained solutions. Repeat this process until convergence. The iterative solution process is as follows: Figure 4 As shown, after obtaining the current optimal solution, the system state will be updated and the next time slot will begin.

[0045] a) Content requests and computation requests. Given a large timescale variable and computational resource allocation decisions, when a vehicle sends a content request, i.e. ,have: ; ; When the vehicle sends a calculation request, that is ,have: ; ; ; Similarly, this embodiment uses relaxation and random rounding techniques to solve the integer programming problem. and To obtain the computational request decision Content request decision .

[0046] b) Computational resource allocation. Based on existing large-scale timescale decisions. and and small-scale decision-making and Substitute them into the problem In the middle, we get: ; ; By solving the problem RSUs can be obtained Optimal allocation of computing resources .because This is a linear convex programming problem, and its Lagrangian function can be defined as: ; in, These are Lagrange multipliers. According to the KKT conditions and... Optimal allocation of computing resources and optimal Lagrange multipliers The following equation must be satisfied: ; in, It is to satisfy The conditional solution obtained at that time.

[0047] It can be observed that It is about It is a monotonic function. Also, because... That is, when When decreasing, Lagrange multipliers Increase, therefore the optimal Lagrange multiplier This can be efficiently obtained using a binary search method. The binary search interval is... The upper boundary satisfy The lower realm satisfy .

[0048] ; When satisfied When the time reaches 1, the binary search algorithm terminates, where Indicates the accuracy of the algorithm.

[0049] This method targets seamless heterogeneous services on high-speed mobile nodes. Based on storage-sensitive service deployment and content caching, it formulates dynamic service / content deployment and task scheduling strategies at different time scales. Furthermore, to quantify the reliability loss caused by service migration, this paper introduces a dual-weighting factor, CHSD, which, along with energy consumption and latency control, serves as a long-term constraint. To verify the effectiveness of this method, its performance is compared with various benchmark algorithms, and a brief analysis is as follows: Figure 5 and Figure 6 For follow The changes in the performance of the proposed TAED algorithm system as the energy increases include total energy consumption, average latency, and CHSD. Figure 5 It can be seen that, with As the energy level increases, total energy consumption gradually decreases while average latency gradually increases. Subsequently, the magnitude of change in total energy consumption and average latency gradually decreases and becomes relatively stable. Figure 6 With The increase in content service security Calculate the unloading gain. and CHSD index It has declined, and in The surrounding area is stable. This is because of the problem. The CHSD item was removed from the text, on the one hand, as... As the energy density increases, the weights of total energy consumption and average latency in the objective function increase, causing TAED to focus more on energy consumption and latency, thus transitioning from a fluctuating state to a stable state. On the other hand, as the weight of CHSD decreases, TAED relaxes its constraints on CHSD during optimization, leading to a reduction in CHSD values. With a balance between energy consumption and latency, the overall system performance stabilizes. Therefore, in Near the target area, various performance indicators gradually approached a balanced state, and the overall performance stabilized. Therefore, based on the simulation results, this paper selected [the appropriate parameters] for subsequent experiments. As a reference parameter.

[0050] To ensure fairness, the number of vehicles is set to 240, and the task density is 10. , Figure 7 This shows the convergence results of the TAED method proposed in this paper. It can be seen that in the initial time slots, both the system's energy consumption and latency exhibit significant fluctuations; as energy consumption increases, latency decreases, indicating a dynamic balance between the two. This also demonstrates that the TAED algorithm operates under dual-timescale co-optimization, where deployment and unloading occur at different time scales. In the initial stage of the algorithm, it is still in the exploratory phase, with large performance fluctuations. However, as time progresses, the algorithm gradually converges, and the system performance stabilizes.

[0051] To verify whether TAED is effective, Figure 8 The content service assurance levels of TAED and five benchmark schemes are presented. Calculate the unloading gain. and CHSD indicators A comparison of aspects. As can be seen from the figure, TAED's... These are 2.2 times, 2.2 times, 4.3 times, and 3.6 times that of MCPSA, TACO, Greedy, and Random, respectively, similar to MBAT. (Higher) This indicates a better ability to fulfill content requests, meaning TAED outperforms most comparable solutions in meeting content needs. Meanwhile, MBAT... Its performance is similar to TAED, but at the cost of higher energy consumption, as will be demonstrated in subsequent comparisons. The calculated offloading gain of TAED, MCPSA, TACO, MBAT, Greedy, and Random... The values ​​are 0.28, 0.26, 0.30, 0.12, 0.14, and 0.27, respectively. The higher values... The improved latency reduction capability indicates that TAED can effectively reduce the processing latency of computational tasks, thus better meeting latency-sensitive requirements. Looking at the CHSD metrics, TAED is 1.9 times, 1.8 times, 3.7 times, and 2.7 times faster than MCPSA, TACO, Greedy, and Random, respectively. This demonstrates that TAED can balance... and This relationship allows for better coordination of services and content in a dynamically deployed environment, and reduces performance losses due to mobility and migration.

[0052] Figure 9 , Figure 10 and Figure 11 The convergence of total energy consumption and average delay for each scheme is described. Figure 9The total energy consumption and average latency of each scheme are described. As can be seen from the figure, although the MBAT scheme achieves a higher CHSD, the cost is much higher than the energy consumption of other schemes. Furthermore, due to its poor long-term stability and excessive energy consumption, MBAT will no longer be included in the comparison in subsequent subplots. Figure 10 The graph shows the change in total energy consumption over 1000 time slots. As can be seen, TAED converges with the lowest energy consumption level, exhibiting a relatively stable overall trend. In contrast, the MCPSA and TACO schemes show larger fluctuations but similar average levels, which is also evident in the box plot. Figure 11 In the results, the average latency of each scheme is roughly concentrated around 0.5 s. The box plot shows that TAED's distribution is more concentrated compared to MCPSA, TACO, and Random, indicating better convergence and robustness in dynamic network environments. The plot also shows that MCPSA and TACO exhibit significant energy consumption fluctuations. MCPSA divides task offloading into many time scales, making coordination between related variables impossible. Furthermore, it focuses on optimizing latency stability while neglecting energy consumption coordination control, thus causing energy consumption fluctuations. In contrast, TACO's piecewise McCormick envelope method accumulates approximation errors for non-convex objective functions, leading to deviations from the global optimum and affecting overall energy consumption performance. TAED, on the other hand, provides long-term joint stability control of energy consumption and latency, resulting in stronger stability under dynamic conditions.

[0053] Therefore, the present invention adopts the above-mentioned task offloading method and system for heterogeneous vehicle services, which can reduce system energy consumption and latency while ensuring service reliability, and its performance is better than the comparison algorithm.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A task offloading method for heterogeneous vehicle services, characterized in that, Includes the following steps: S1. On a large time scale, determine service deployment decisions and content deployment decisions based on the storage capacity constraints of roadside units; S2. At a small time scale, determine the task offloading decision and computing resource allocation decision for vehicle heterogeneity based on service deployment decisions and content deployment decisions; S3. Introduce the vehicle service collaborative benefit evaluation function CHSD, which integrates content service assurance and calculation offloading gain, to measure the reliability loss of service and content migration. S4. Establish an energy consumption deficit queue and a time delay overflow queue. Use the K-step Lyapunov drift penalty reduction framework to refactor the optimization problem into decomposable subproblems. Solve the subproblems of different time scales alternately until convergence. Update the queue state and enter the next time slot.

2. The task offloading method for heterogeneous vehicle services according to claim 1, characterized in that, S1 includes: S11. Linearly relax the integer programming problem of service deployment and content deployment; S12. The random rounding algorithm based on Chernoff bounds is used to solve the problem and obtain the integer deployment decision that satisfies the storage capacity constraint.

3. The task offloading method for heterogeneous vehicle services according to claim 2, characterized in that: The vehicle heterogeneous task in S2 includes computation requests and content requests; The computational requests require computing resources from the roadside unit or the cloud to complete the processing. Content requests require the roadside unit to pre-cache the requested content file and respond with the transmission.

4. The task offloading method for heterogeneous vehicle services according to claim 3, characterized in that, S2 includes: S21. Solve the integer programming problem using the linear relaxation plus random rounding method for the computation request and the content request respectively, and obtain the task unloading decision. S22. Allocate computing resources, establish the Lagrange function, derive the closed-form solution using the KKT conditions, and solve for the optimal Lagrange multiplier through binary search.

5. A task offloading method for heterogeneous vehicle services according to claim 4, characterized in that, Solving for the optimal Lagrange multipliers in S22 includes: When Lagrange multipliers At that time, the optimal allocation of computing resources is ,in It is the extreme point under the corresponding piecewise condition; otherwise, , which represents the maximum available computing resources for the roadside unit; The termination precision of the binary search is set to The maximum number of iterations is set to 200.

6. A task offloading method for heterogeneous vehicle services according to claim 5, characterized in that: In step S3, the vehicle service collaborative benefit evaluation function CHSD is constructed, and its expression is: ； in, This is a dynamic bias parameter used to adjust the optimization focus between content service assurance and computational task processing based on the current vehicle network service status. This represents the content service assurance level, which measures the ability of edge nodes to provide real-time support for vehicle content access requests within a time slot. Its value is determined by the ratio of the number of successfully responded content requests to the total number of content requests. The calculation of the offloading gain is used to reflect the latency optimization benefit of the task after collaborative offloading compared to the pure local execution mode. Its value is defined as the normalized difference between the baseline local execution latency and the actual completion latency. The vehicle service collaborative benefit evaluation function CHSD is used to uniformly characterize the collaborative optimization effect of cache resource configuration, service deployment decisions and task offloading strategies in the vehicle heterogeneous service scenario, and measure the reliability loss.

7. A task offloading method for heterogeneous vehicle services according to claim 6, characterized in that: The energy loss queue in S4 and delayed overflow queue The update method is as follows: ； ； in, For time slots The actual energy consumption inside, As a long-term budget ceiling for energy consumption, For vehicles The actual delay As the upper limit of the extended budget, The number of time slots contained within a time frame.

8. A task offloading method for heterogeneous vehicle services according to claim 7, characterized in that, The alternating solution in S4 includes: At the beginning of each time frame, S1 is executed first to obtain deployment decisions on a large time scale, and then S2 is executed to obtain offloading and resource allocation decisions on a small time scale. Substitute the result of S2 back into S1 to re-optimize the deployment decision, and repeat this process iteratively until the change in the objective function value is less than the preset threshold.

9. A system applied to the task offloading method for heterogeneous vehicle services as described in claim 8, characterized in that: This includes roadside units, remote cloud servers, and controllers; Multiple roadside units are provided, and each roadside unit is equipped with storage resources and computing resources; The remote cloud server is communicatively connected to the roadside unit; The controller is deployed on the roadside unit or the remote cloud; The controller is configured to: On a large time scale, service deployment decisions and content deployment decisions are determined based on the storage capacity constraints of roadside units; on a small time scale, vehicle heterogeneous task offloading decisions and computing resource allocation decisions are determined based on the service deployment decisions and content deployment decisions; the CHSD index is introduced to measure reliability loss; a queue is established and an alternating solution is performed using a K-step Lyapunov drift reduction penalty framework.

10. The system for a task offloading method for heterogeneous vehicle services according to claim 9, characterized in that: The controller is used to perform linear relaxation and Chernoff bound-based random rounding algorithms to solve deployment decisions at the beginning of each time frame; and to perform Lagrange duality and bisection search to solve computational resource allocation within each time slot.

Images