A chance-constrained MEC task offloading method

By constructing a joint optimization model and a hybrid offloading algorithm, the problem of inaccurate offloading decisions in multi-access edge computing environments is solved, achieving efficient, low-cost, and secure task offloading, and optimizing task allocation among terminals, edge servers, and cloud computing centers.

CN121396989BActive Publication Date: 2026-11-10XIAN UNIV OF POSTS & TELECOMM
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202511783311.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-12-01
Publication Date
2026-11-10
Estimated Expiration
2045-12-01

AI Technical Summary

Technical Problem

In multi-access edge computing environments, the environmental uncertainties caused by the dynamic changes in task characteristics and network resources, the heterogeneity of multiple network accesses, and the need for personalized privacy protection are difficult to effectively address, leading to inaccurate offloading decisions and increased terminal pressure and data transmission consumption.

Method used

By constructing a latency-energy consumption model, a fuzzy privacy model, and a chance-constrained model, a joint optimization model is established. An improved hybrid offloading algorithm combining adaptive genetics, Monte Carlo simulation, and neural networks is used to solve the task offloading strategy and network selection strategy, thereby optimizing the allocation of tasks across terminals, edge servers, and cloud computing centers.

Benefits of technology

It enables the reduction of latency and energy consumption costs for offloading decisions, the improvement of task processing efficiency, the protection of privacy, the reduction of network transmission occupancy, and the improvement of transmission speed and security in multi-access edge computing environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121396989B_ABST
    Figure CN121396989B_ABST
Patent Text Reader

Abstract

This invention discloses a chance-constrained MEC task offloading method, relating to the field of cloud computing technology. The method includes: constructing a latency-energy consumption model, a fuzzy privacy model, and a chance-constrained model using the fuzzy data input and fuzzy CPU cycle count of the fuzzy task to be processed; establishing a joint optimization model using the latency-energy consumption model, the fuzzy privacy model, and the chance-constrained model; solving the joint optimization model to obtain a task offloading strategy and a network selection strategy; allocating the fuzzy task to be processed to offloading locations according to the task offloading strategy; and performing resource transmission according to the network selection strategy. This invention solves the joint optimization model using a designed offloading algorithm, obtaining optimal solutions for the task offloading location strategy and the network data transmission selection strategy. Tasks are allocated to corresponding offloading locations. According to the network data transmission selection strategy, task data is proportionally allocated to different networks for data transmission.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, and in particular to a method for offloading MEC tasks based on opportunity constraints. Background Technology

[0002] MEC, short for Multi-Access Edge Computing, is a new computing approach that provides users with resources across a three-tiered system: cloud servers, edge servers, and terminals. Multi-access edge computing increases available terminal resources and improves task processing efficiency by offloading computing tasks from the terminal to edge and cloud servers. However, users need to make accurate decisions when deciding whether to execute tasks locally on the terminal or offload them to edge or cloud servers. Incorrect offloading decisions not only fail to reduce terminal load but can also lead to increased resource consumption due to additional data transmission.

[0003] Existing technologies still have some problems when dealing with complex and ever-changing network environments and uncertain task requirements, such as environmental uncertainty caused by dynamic changes in task characteristics and network resources, heterogeneity of multiple network accesses, and personalized privacy protection requirements. Summary of the Invention

[0004] This invention provides an opportunity-constrained MEC task offloading method to address the problems of environmental uncertainty caused by dynamic changes in task characteristics and network resources, heterogeneity of multiple network accesses, and personalized privacy protection requirements in the prior art.

[0005] On one hand, embodiments of the present invention provide a MEC task offloading method based on chance constraints, including: constructing a latency-energy consumption model, a fuzzy privacy model, and a chance constraint model by using the fuzzy data input amount and fuzzy CPU cycle number of the fuzzy task to be processed; A joint optimization model is established using the aforementioned time-delay energy consumption model, fuzzy privacy model, and chance-constraint model. Solving the joint optimization model yields the task offloading strategy and the network selection strategy; The fuzzy pending tasks are assigned to the uninstallation location according to the task uninstallation strategy; Resource transfer is performed according to the network selection strategy.

[0006] In one possible implementation, the offloading location includes a terminal, an edge server, and a cloud computing center, wherein the cloud computing center is equipped with multiple cloud servers.

[0007] In one possible implementation, the latency and energy consumption model is used to obtain the latency and energy consumption costs of the task to be processed during data transmission and processing at the offloading location. The latency and energy consumption model enables data transmission between the terminal and the edge server, as well as between the edge server and the cloud server, through multiple wireless networks simultaneously.

[0008] In one possible implementation, the fuzzy privacy model includes fuzzy privacy level requirements for the fuzzy task to be processed and data privacy entropy quantization. The fuzzy privacy level requirements are used to constrain the degree of privacy protection of the fuzzy task to be processed during transmission, and the data privacy entropy quantization is used to obtain the probability and level of data leakage of the fuzzy task to be processed during data transmission.

[0009] In one possible implementation, the opportunity constraint model is used for the conversion between the determinization of the fuzzy task data volume and the number of CPU cycles and the determinization of the opportunity constraint form of the fuzzy privacy model.

[0010] In one possible implementation, the latency-energy consumption model includes a communication model and a computation model, wherein the communication model is used to obtain the latency cost of the fuzzy task to be processed, and the computation model obtains the energy consumption cost of the task to be processed.

[0011] In one possible implementation, solving the joint optimization model to obtain the task offloading strategy and network selection strategy includes: The joint optimization model is solved using a hybrid unloading algorithm combining improved adaptive genetics, Monte Carlo simulation, and neural networks.

[0012] The opportunity-constrained MEC task offloading method of this invention has the following advantages: (1) By transforming the multi-access edge computing task offloading problem into a joint optimization model based on communication model, computing model, privacy entropy model and opportunity constraint model, the problem of insufficient offloading decision to meet high task processing efficiency, large local terminal processing pressure, and high latency and energy consumption cost in the process of additional data transmission is solved from the perspectives of uncertain environment, latency and energy consumption measurement and privacy protection.

[0013] (2) The joint optimization model was solved using the designed unloading algorithm, yielding optimal solutions for the task unloading location strategy and the network data transmission selection strategy. Based on the task unloading strategy, tasks were assigned to corresponding unloading locations. Based on the network data transmission selection strategy, task data was proportionally allocated to different networks for data transmission. This solves the problems of minimizing latency and energy consumption during task unloading and protecting privacy during data transmission, achieving high efficiency, low cost, and security in task unloading. Furthermore, this application enables multi-terminal transmission between edge servers and other terminals, reducing network occupancy and improving transmission speed. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating an opportunity-constrained MEC task offloading method provided in this application; Figure 2 A schematic diagram of the HTOA-IMN algorithm, a hybrid unloading algorithm based on improved adaptive genetics, Monte Carlo simulation, and neural networks, provided for embodiments of this application; Figure 3 The different computing power at different unloading locations provided by the unloading method of this application in the embodiments of this application ( A diagram illustrating the target value and task distribution ratio under ( ); Figure 4 Different computing power at the local terminal uninstallation location using the uninstallation method of this application, as provided in the embodiments of this application. A diagram illustrating the target value and task distribution ratio under ( ); Figure 5 Different computing power at the edge server unloading location using the unloading method of this application, as provided in the embodiments of this application. A diagram illustrating the target value and task distribution ratio under ( ); Figure 6 Different computing power at the cloud server uninstallation location using the uninstallation method of this application, as provided in the embodiments of this application. A diagram illustrating the target value and task distribution ratio under ( ); Figure 7 This is a schematic diagram illustrating the impact of changes in the number of multiple network accesses on the target value using the unloading method of this application, provided in an embodiment of this application. Figure 8A schematic diagram illustrating the target value of using the uninstallation method of this application in a stage where the number of tasks is between 10 and 40, as provided in the embodiments of this application; Figure 9 A schematic diagram illustrating a target value of 60 to 90 tasks using the uninstallation method of this application, provided for an embodiment of this application. Figure 10 A schematic diagram illustrating the impact of the uninstallation method of this application on the system target value under low, medium, and high privacy and security requirements, provided for embodiments of this application. Figure 11 A schematic diagram illustrating the impact of the uninstallation method of this application on the task uninstallation ratio at three locations—terminal, edge server, and cloud service center—under different privacy and security requirements, provided for embodiments of this application. Figure 12 A line graph illustrating the performance comparison of the unloading method of this application with other technologies at different iteration numbers, provided for embodiments of this application; Figure 13 Box plot showing the performance comparison of the unloading method of this application with other techniques at different iteration numbers, provided for embodiments of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Figure 1 This is a flowchart illustrating a chance-constrained MEC task offloading method according to an embodiment of the present invention. The embodiment of the present invention provides a chance-constrained MEC task offloading method, including: A latency-energy consumption model, a fuzzy privacy model, and a chance-constraint model are constructed by using the fuzzy data input amount and the number of fuzzy CPU cycles of the fuzzy task to be processed. A joint optimization model is established using the aforementioned time-delay energy consumption model, fuzzy privacy model, and chance-constraint model. Solving the joint optimization model yields the task offloading strategy and the network selection strategy; The fuzzy pending tasks are assigned to the uninstallation location according to the task uninstallation strategy; Resource transfer is performed according to the network selection strategy.

[0018] The unloading locations include terminals, edge servers, and cloud computing centers, with the cloud computing centers equipped with multiple cloud servers.

[0019] The latency and energy consumption model is used to obtain the latency and energy consumption cost of the task to be processed during data transmission and processing at the offloading location. The latency and energy consumption model enables data transmission between the terminal and the edge server, as well as between the edge server and the cloud server, through multiple wireless networks simultaneously.

[0020] The fuzzy privacy model includes fuzzy privacy level requirements for the fuzzy task to be processed and data privacy entropy quantization. The fuzzy privacy level requirements are used to constrain the degree of privacy protection of the fuzzy task to be processed during transmission, and the data privacy entropy quantization is used to obtain the probability and level of data leakage of the fuzzy task to be processed during data transmission.

[0021] The opportunity constraint model is used for the conversion between the determination of the amount of fuzzy task data and the number of CPU cycles, and the determination of the opportunity constraint form of the fuzzy privacy model.

[0022] The latency and energy consumption model includes a communication model and a computation model. The communication model is used to obtain the latency cost of the fuzzy task to be processed, and the computation model obtains the energy consumption cost of the task to be processed.

[0023] Solving the joint optimization model yields the task offloading strategy and network selection strategy, including: The joint optimization model is solved using a hybrid unloading algorithm combining improved adaptive genetics, Monte Carlo simulation, and neural networks.

[0024] For example, the unloading locations in this application include terminals, edge servers, and cloud computing centers, respectively via... l, s, c This indicates that terminals and edge servers can connect in parallel via multiple wireless networks, while edge servers and cloud computing centers can connect via wireless or wired networks.

[0025] The terminals can be industrial equipment such as intelligent monitoring devices and intelligent robots. The wireless networks can be wireless wide area networks (WANs), wireless local area networks (LANs), wireless metropolitan area networks (MANs), and wireless personal area networks (PANs), and the network transmission protocols can be Bluetooth, 5G, Wi-Fi, and ZigBee, etc. Edge servers are considered edge computing centers that are closer to the terminal side. The cloud computing center layer includes multiple cloud servers or high-performance servers.

[0026] The information for the task to be processed includes the amount of fuzzy input data and the number of fuzzy CPU cycles required to process the task.

[0027] In this embodiment of the application, the information of the task to be processed is shown in the following formula.

[0028]

[0029] In the formula, This represents the tasks to be processed, where the index n indicates the task number in the set. The amount of fuzzy input data for the task to be processed. The number of fuzzy CPU cycles required to process the task to be processed is here. , To pessimistically estimate the lower bound, , To optimistically estimate the upper bound, and This forms the confidence interval for the actual value. This parameterized model allows decision-makers to reflect differentiated assessments of task attributes by adjusting the interval boundary parameters, both of which are fuzzy numbers.

[0030] A joint optimization model is established using a time-delay energy consumption model, a fuzzy privacy model, and a chance-constrained model.

[0031] In this embodiment of the application, the latency-power consumption model can support the simultaneous transmission of task data to the offloading location using multiple wireless networks. The offloading location includes the terminal, edge server, and cloud computing center server.

[0032] Therefore, the communication model supports the simultaneous use of multiple wireless networks for data transmission between terminals and edge servers, as well as between edge servers and cloud servers.

[0033] The process of constructing the time-delay energy consumption model is as follows: When pending tasks are offloaded to edge servers, the data uplink transmission latency cost The calculation is as follows: .

[0034] in, This indicates that the data is unloaded from the local terminal to the edge terminal. Indicates that the terminal is in a multi-network set The maximum delay value in; This represents the data transmission allocation strategy for the nth task; This represents the data transmission ratio of the nth task on the mth wireless network, multiplied by the total amount of fuzzy data for the nth task. get , representing the amount of fuzzy data transmitted in the m-th network for this task; the denominator comes from the Shannon formula, where , , and These represent the bandwidth, transmission power, channel gain, and noise power spectral density of the m-th wireless network, respectively.

[0035] Meanwhile, the energy consumption of data transmission during task offloading to the edge server This equals the cumulative energy consumption of all wireless networks, and can be expressed as: .

[0036] When a task is unloaded to the cloud server, the task data must first be transmitted to the access point (located at the same time as the edge server) via a wireless network, and then transmitted to the cloud via the fiber optic backbone network. The data transmission latency for unloading a task to the cloud is expressed as: .

[0037] in, This indicates that the task has been unloaded from the local terminal to the cloud; This represents the round-trip propagation delay of the fiber optic backbone network. Considering that fiber optic transmission is more energy efficient than wireless transmission, negligible energy consumption is ignored, and the data transmission energy consumption is expressed as... .

[0038] In addition to the latency and energy consumption costs of the task unloading process mentioned above, the corresponding computational latency and energy consumption for processing tasks at different unloading locations are calculated as follows: , , in, Indicates the CPU power consumption coefficient. This refers to the three unloading locations. l :local s Edge server c cloud server This indicates the computing power of the unloading location.

[0039] The above-mentioned latency and energy consumption calculation based on the latency and energy consumption model yields the total latency and total energy consumption cost at each unloading location, which can be obtained from the latency and energy consumption calculation of the task being unloaded from the local terminal to the edge terminal. For example, the total latency cost of unloading the task from the local terminal to the edge terminal. Total energy consumption cost They can be calculated separately as and In addition, it can also minimize task processing latency and energy consumption.

[0040] This application also provides a fuzzy privacy model, including fuzzy privacy level requirements for tasks and data privacy entropy quantification.

[0041] To assess channel security, we divide the transmitted data into several segments of length [missing information]. A block of bits. Let... This indicates that the task is performed through the m-th wireless network. The number of data blocks transmitted. The calculation formula is: , in Let represent the upper bound function. Let represent the probability of leakage of k data blocks on the m-th wireless network. Assume that privacy leakage events follow a Poisson distribution. The calculation formula is: , in Represents the average privacy leakage on the m-th wireless network, where... It is the base of the natural logarithm. Through the m-th wireless network and the task... Privacy entropy associated with data transmission Calculated using the following formula: .

[0042] Fuzzy task privacy level requirements are used to constrain the degree of privacy protection during task transmission. The requirements for data privacy and security protection during transmission are divided into three levels: "low," "medium," and "high." Different security levels correspond to different fuzzy entropy thresholds. , and By calculating the privacy entropy threshold and tasks Privacy entropy The difference quantifies user satisfaction with privacy protection, and is expressed as... Negative values ​​indicate that the user's privacy and security level meets the threshold requirements, while positive values ​​indicate insufficient privacy protection.

[0043] The fuzzy task privacy level requirement is used to constrain the degree of privacy protection for the task during transmission; the data privacy entropy is used to obtain the probability and level of data leakage of the task to be processed during data transmission. Different privacy protection levels, different specific locations, and different multi-network transmission strategies set by the user will result in different probabilities of information leakage, thus affecting the final offloading strategy.

[0044] The joint optimization model consists of an objective function and multiple constraints. The objective function aims to minimize task processing time and minimize task processing energy consumption, and is expressed as: .

[0045] in and Representing tasks Different uninstall locations i ( i∈l,s,cThe latency and energy consumption costs of ) Indicates task At the corresponding uninstallation location i The decision to uninstall N This indicates the total number of tasks.

[0046] The constraints of the joint optimization model are as follows: (c1) This indicates that user privacy levels are strictly controlled within a predefined threshold, thereby ensuring user privacy security.

[0047] (c2) This means imposing mathematical constraints on the offloading strategy to ensure that each task is executed only once on the mobile device, at the edge, or in the cloud.

[0048] (c3) This indicates that mathematical constraints are imposed on the allocation strategy, meaning that no data transmission occurs when the task is executed locally, and when the task is unloaded, the wireless network collaborates to complete all data transmission.

[0049] (c4) This indicates the uninstallation decision at different uninstallation locations. It is a binary value, which can only be 0 or 1.

[0050] (c5) The network transmission data allocation strategy for the nth task on the mth wireless network. Limits to any value between 0 and 1.

[0051] This application also provides an opportunity constraint model, including the determination of the amount of fuzzy task data and the number of CPU cycles, the objective function, and the transformation of the opportunity constraint form of the fuzzy privacy model.

[0052] Opportunity Constraints It is a probability measure function, whose value quantifies the degree of likelihood that the input fuzzy conditions will be met. Under fuzzy chance constraints, the optimization objective is transformed into minimizing... Meanwhile, the objective function based on fuzzy chance constraints and the privacy entropy constraint are respectively transformed into , .

[0053] in, and These are preset confidence levels, used to measure the confidence level of whether the privacy constraints and objective function constraints are satisfied. Subsequently, the joint optimization model can be transformed into the following form: .

[0054] Under fuzzy chance constraints, a solution is admissible if the probability of satisfying the constraints remains at least at a pre-specified confidence threshold, while occasional violations of the constraints are allowed. Through fuzzy chance constraint transformation, the solution process is converted into finding the minimum objective function that achieves the confidence level while satisfying the constraints.

[0055] The model transformation converts the original uncertain optimization problem into a deterministic equivalent with provable confidence level guarantees. The transformed model retains the physical interpretation and structural characteristics of the original task unloading problem. It transforms the original uncertain optimization problem, which involved complex membership function processing, into an equivalent deterministic problem. This transformation provides a mathematical basis for replacing complex fuzzy analysis derivations with numerical computation, thereby reducing the difficulty of solving the task unloading problem.

[0056] To obtain the optimal task unloading and data allocation strategy from the transformed optimization model, a hybrid task unloading algorithm based on an improved adaptive genetic algorithm, Monte Carlo simulation, and neural networks is designed and proposed.

[0057] An adaptive genetic operator with dynamically adjusted evolutionary parameters and an improved repair mechanism are combined to handle infeasible solutions while preserving their genetic information. To further improve the efficiency of the task offloading algorithm, a neural network is used instead of computationally intensive Monte Carlo simulation. In the offline phase, samples generated by Monte Carlo simulation are used to train the neural network, thereby accurately approximating the input-output mapping relationship of the fuzzy chance constraints. In the online phase, the trained neural network is integrated into the algorithm's optimization process to estimate the fuzzy chance constraint values ​​in real time.

[0058] like Figure 2 As shown, the hybrid task offloading algorithm based on improved adaptive genetic algorithm, Monte Carlo simulation and neural network to solve the joint optimization model mainly involves six steps, summarized as follows: Initialize the random population, calculate individual fitness, and set algorithm parameters.

[0059] Select a subset of individuals based on the selection strategy set by the algorithm, perform crossover and mutation among individuals using adaptive parameters, and repair gene segments in the offspring chromosomes that violate constraints.

[0060] Greedy inheritance operations are performed on non-genetic segments of the offspring chromosomes.

[0061] The offspring chromosomes were re-evaluated using a trained neural network.

[0062] Update the optimal solution for the population, and regenerate new individuals to join the population at random intervals.

[0063] Repeat the above steps until the iteration meets the termination criterion, and output the global optimal solution and fitness value.

[0064] Population individuals are represented as ,in . This indicates the uninstallation location selection for an individual device; 0, 1, and 2 represent the local device, edge device, and cloud device, respectively. This indicates the proportion of data distribution across multiple wireless networks.

[0065] The calculation of an individual's fitness depends on the fitness function, expressed as: , in, It is a penalty coefficient that balances objective optimization and constraint satisfaction by scaling the magnitude of the penalty for violating constraints. The penalty function for chromosomes that violate privacy entropy constraints is defined as: , When the privacy entropy condition of the fuzzy chance constraint is met, the penalty function returns zero, and no penalty is imposed on the chromosome. Otherwise, as the value of the fuzzy chance constraint decreases, the fitness of the chromosome gradually decreases.

[0066] In calculating the fitness value, Monte Carlo sampling is combined with fuzzy simulation to obtain accurate chance constraint evaluation values. The proposed dual simulation method minimizes the systematic error achieved by the traditional single method, while enhancing the consistency of the uncertainty spectrum results.

[0067] Calculating the fuzzy chance constraint values ​​requires an initial probability estimate for the fuzzy chance constraints during the initialization process. This can begin with a conservative (deliberately as small as possible) initial estimate. During the iteration process, fuzzy guided Monte Carlo sampling generates input parameters reflecting the system's uncertainties. Under simulation conditions, the probability of the fuzzy chance constraints is calculated using membership functions.

[0068] While Monte Carlo-based fuzzy simulation effectively solves the computational problem of fuzzy chance constraints, it requires independent computation for each sample and numerous iterations, leading to prohibitively high computational costs. To overcome this efficiency bottleneck, a neural network is integrated to estimate the fuzzy chance constraint values. To ensure the trained neural network accurately approximates the objective function value and privacy entropy constraint, a training dataset is first generated through Monte Carlo-based fuzzy simulation. This dataset serves as the basis for supervised learning, ensuring the neural network captures the latent input-output relationship between the fuzzy objective function and privacy entropy constraint. The trained neural network is then integrated into the iterative process of a genetic algorithm, replacing the original Monte Carlo-based fuzzy simulation process. By providing real-time estimation of the fuzzy chance constraint values, the optimization speed is significantly accelerated while maintaining solution quality.

[0069] Subsequently, crossover variations of individual gene segments are represented using an adaptive strategy, as follows: .

[0070] This represents the maximum fitness value in the population. This represents the average fitness value of the population in each generation. This represents the larger fitness value among the two individuals to be crossed. This represents the fitness value of the individual to be mutated. and Then, they represent the adaptive crossover and mutation probabilities, respectively, where , , and These are four constants. To prevent the population from rapidly converging to a local optimum in the early stages of evolution, the crossover and mutation probabilities of fitter individuals, after calculation using the above formula, will not tend towards a minimum value, but will instead increase to a higher value. and This can, to some extent, ensure that the population does not mature too early.

[0071] Since each chromosome encodes task offloading decisions and assignment ratios, applying standard crossover and mutation operations to the entire chromosome would significantly impair the progeny population's ability to explore the solution space. Therefore, we perform crossover and mutation operations on all gene segments simultaneously. The system model imposes a constraint requiring the sum of all task assignment ratios to equal 1. However, after random crossover and mutation operations, the resulting sub-chromosomes may violate this constraint, thus hindering the generation of feasible solutions within the search space. To restore feasibility, all assignment ratios are scaled proportionally to a normalized sum equal to 1 while preserving their original relative proportions.

[0072] Individual uninstallation strategy The fact that the crossover is a single discrete value fundamentally limits its effectiveness. To avoid crossover degenerating into pattern-destroying value swaps, a greedy genetic operation is proposed to replace the traditional crossover operation. The specific steps are as follows: Step 1: Use a greedy strategy to sequentially modify the unloading location of individuals. Step 2: Recalculate and compare fitness values ​​at different unloading locations. Step 3: Modifications will only be retained if the fitness value is improved; otherwise, the original position will be retained.

[0073] While greedy genetic operations effectively accelerate convergence, their strong goal-driven nature can lead to chromosomes getting trapped in local optima. Adaptive mutation is employed as a complementary mechanism, where the offspring position is randomly perturbed with probability after genetic operations, thus achieving a balance between exploitation and exploration.

[0074] After determining the task offloading strategy and network allocation strategy, the proportion of tasks to be processed in the offloading location is determined according to the offloading strategy, and the proportion of task data transmitted simultaneously on each network is determined according to the network allocation strategy.

[0075] In this embodiment of the application, the parameter settings of the multi-access edge computing system are shown in Table 1: Table 1 Parameter Table of Multi-Access Edge Computing System

[0076] like Figures 3 to 13 As shown, the task unloading method provided in this application is used to perform task unloading, and the results and effects are as follows.

[0077] like Figure 3 As shown, in order to evaluate the impact of the changes in computing power at each proposed unloading location on the objective function during the unloading process, the changes in the objective value and the task unloading distribution at different unloading locations and under different computing power are studied by adjusting the percentage of computing power at each unloading location. Figure 3 This demonstrates the impact of changes in computing power on the target value. Figure 4 , 5 Figures 6 and 6 respectively demonstrate the task offloading and allocation of local terminals, edge servers, and cloud computing centers under conditions of changing computing power.

[0078] Figure 3 The change in computing power at each location corresponds to a broken line, showing how the target value of the local terminal changes with computing power ( Local computing power Figure 3 The local terminal (referred to as the "local terminal") gradually increases and then monotonically decreases. And... (edge ​​server computing power, Figure 3 (referred to as edge end) and (Computing power of cloud computing centers) Figure 3 When the computing power (hereinafter referred to as cloud computing power) reaches the 20% to 100% stage, the target values ​​of edge servers and cloud computing centers show a trend of first decreasing and then slowly rising to a stable level. This phenomenon is because when computing power is too low, higher latency accounts for the majority of the target value; while when computing power exceeds an appropriate level, the energy consumption caused by high computing power also increases, thus affecting the target value. In the stage exceeding 100% (i.e., basic computing power), the target values ​​of edge servers and cloud servers show slight fluctuations, reflecting the complex relationship between latency and energy consumption. Figure 4 As computing power increases, local terminals are able to complete tasks with lower target values, thus gradually acquiring more tasks. The task ratio increased from an initial 16% to 60%, and even when computing power reached 140%-160%, all tasks remained executed locally. Clearly, the increase in computing power brought significant benefits to the local machine. Figure 5 In the process, the number of tasks on edge servers showed a trend of first increasing and then decreasing, while the number of tasks on local terminals fluctuated slightly. At 40% computing power, edge servers could handle nearly 60% of the tasks, while the number of tasks in the cloud computing center accounted for only 20%, mainly due to privacy and security constraints and computing power limitations. As computing power continued to increase, the energy consumption of edge servers increased faster than their latency reduction, leading to a rise in target values. This caused tasks to tend to be processed on local terminals and in the cloud computing center. Ultimately, when computing power reached 160%, only 3% of tasks were offloaded to edge servers. Figure 6 It exhibits a similar trend to edge servers, which will not be elaborated here.

[0079] It can be seen that continuously increasing computing power has a positive effect because local computing eliminates the overhead of the unloading process. However, for edge servers and cloud computing centers, maintaining computing power at an appropriate level is crucial to optimize task unloading efficiency and system target values.

[0080] To evaluate the multi-network transmission communication model during the unloading process. Figure 7 The impact of changes in the number of network access points on the target value was compared, among which... Figure 8 and 9 They were shown respectively Figure 7 The details at different stages are as follows. It can be observed that as the number of tasks increases, the target value also rises, a trend consistent with real-world application scenarios. When the number of tasks is low, the system load is small, and the target value changes relatively smoothly. However, as the number of tasks increases, the number of tasks the system needs to process surges, and the target value inevitably rises.

[0081] It is worth noting that, in Figure 8In the phase where the number of tasks ranges from 10 to 40, the change of the target value does not follow the monotonic decreasing rule of the number of access networks, and the following sequence appears: Net=3<Net=4<Net=5<Net=2. This indicates that under the same number of tasks, an increase in the number of access networks does not always lead to a better target value. This non-monotonic phenomenon can be attributed to multiple factors, such as the differences between networks, the requirements for ensuring privacy and security, and the trade-off between energy consumption and delay. In the experimental design, the bandwidth of multiple networks fluctuates, resulting in unstable channel quality between different networks. When the number of tasks is small, although the number of network accesses is increased, unstable fluctuations in bandwidth quality may lead to an increase in energy consumption. In addition, in a fluctuating network environment, maintaining the security of data privacy becomes more complex, which may cause more tasks to choose to remain for local computation, thereby increasing the system burden and additional overhead. Although in theory multi-network access helps to share the task load, excessive network access may lead to higher energy consumption, especially in the case of poor channel quality, the additional overhead may exceed the benefits brought by transmission delay.

[0082] As Figure 9 shown, as the number of tasks further increases (in the 60-90 phase), the relationship of target values between different numbers of networks tends to be stable, indicating that under a certain task scale, more network accesses mean smaller transmission overhead. At this time, the impact of the number of access networks on the target value begins to show a positive effect, demonstrating that in a high-load environment, a reasonable number of networks can effectively reduce the system burden.

[0083] The influence of the degree of privacy and security requirements on the system target value and task offloading decision is analyzed, and the influence on the above indicators under different numbers of tasks is observed by changing the privacy requirements. The results are as Figure 10 shown, in Figure 10 , the impact of low, medium and high privacy and security requirements on the system target value is demonstrated, Figure 11 the impact of different privacy and security requirements on the task offloading ratios at three locations: the terminal, the edge server and the cloud service center is demonstrated.

[0084] As the number of tasks increases, the system target values under different privacy requirements generally show an increasing trend. This indicates that under the condition of the same number of tasks, the higher the privacy and security requirements, the higher the system target value. This is because higher privacy requirements will restrict more tasks from being offloaded to the cloud server, causing tasks to stay for local computation, thereby increasing the overall overhead of the system. At the same time, since the cloud server is far away, privacy leakage is more likely to occur during transmission, so stricter privacy requirements prompt the system to reduce its dependence on the cloud server and increase the proportion of tasks processed by local intelligent terminals.

[0085] When privacy requirements are low, most tasks can meet transmission privacy requirements and tend to be offloaded to edge servers and cloud servers with higher computing power. Due to the distance advantage of edge servers compared to cloud service centers, tasks are primarily offloaded to edge servers. With increased privacy requirements, the proportion of tasks remaining on local terminals rises significantly, even exceeding the number offloaded to cloud service centers. As the number of tasks increases, the proportion offloaded to cloud service centers gradually decreases, reflecting the increased demand for local computing due to heightened privacy and security concerns. In scenarios with high privacy requirements, almost all tasks do not choose to offload to cloud service centers, and edge servers may also face attack risks. In this case, the vast majority of tasks choose to remain on local terminals for computation, although this leads to longer latency and higher system target values, but it satisfies privacy and security requirements.

[0086] It can be seen that increased privacy requirements significantly affect the offloading rate of tasks and the system target value. With low privacy requirements, tasks tend to be offloaded to servers with stronger computing power, while high privacy requirements encourage more tasks to remain for local computation, leading to an increase in the system target value.

[0087] Figure 12 , 13 This diagram illustrates the performance comparison of the unloading method proposed in this application with other techniques at different iteration numbers, provided for embodiments of this application. The algorithm proposed in this application (HTOA-IMN) was comprehensively compared and evaluated with newer existing benchmark algorithms: (1) Task unloading algorithm based on whale optimization algorithm (TOA-WOA), (2) Task unloading algorithm based on bat algorithm (TOA-BA), (3) Task unloading algorithm based on genetic algorithm (TOA-GA), and (4) Task unloading algorithm based on simulated annealing algorithm (TOA-SA). The results are as follows: Figure 12-13 As shown.

[0088] As the number of iterations increases, the HTOA-IMN algorithm exhibits relatively rapid convergence, stabilizing around 40 iterations, and its objective value is significantly superior to other algorithms. This indicates that the HTOA-IMN algorithm can effectively search the solution space and quickly find a better solution during the optimization process. In comparison, the TOA-SA and TOA-WOA algorithms also show good convergence speed, both reaching convergence after 60 iterations. However, despite their faster convergence speed, their final objective values ​​are still lower than those of HTOA-IMN, showing that HTOA-IMN performs better in specific problem contexts. This may be related to the design and mechanism of the HTOA-IMN algorithm, making it more advantageous in handling complex optimization tasks. The TOA-GA and TOA-BA algorithms show a positive trend in the convergence process; however, their convergence speed is relatively slow, failing to reach the performance level of other algorithms within the same number of iterations. This indicates that although TOA-GA and TOA-BA can find feasible solutions, their optimization efficiency is low, and they may require more iterations to reach the ideal objective value under certain parameter settings. In summary, the convergence characteristics of the various algorithms show significant differences. The HTOA-IMN algorithm has a significant advantage in optimization performance, while TOA-SA and TOA-WOA, although possessing certain convergence capabilities, still cannot match HTOA-IMN.

[0089] from Figure 12-13 As can be seen, the algorithm HTOA-IMN proposed in this application has the smallest objective value and a faster convergence speed, and its performance is better than other algorithms in the prior art.

[0090] While this application provides the method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in this embodiment is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the methods shown in this embodiment or the accompanying drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0091] The various embodiments described in this specification are presented in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. All or part of this application can be used in numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, mobile communication terminals, multiprocessor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0092] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0093] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for offloading MEC tasks based on opportunity constraints, characterized in that, include: A latency-energy consumption model, a fuzzy privacy model, and a chance-constraint model are constructed by using the fuzzy data input amount and the number of fuzzy CPU cycles of the fuzzy task to be processed. A joint optimization model is established using the aforementioned time-delay energy consumption model, fuzzy privacy model, and chance-constraint model. Solving the joint optimization model yields the task offloading strategy and the network selection strategy; The fuzzy pending tasks are assigned to the uninstallation location according to the task uninstallation strategy; Resource transmission is performed according to the network selection strategy described above; The unloading locations include terminals, edge servers, and cloud computing centers, with the cloud computing centers equipped with multiple cloud servers; The fuzzy privacy model includes fuzzy privacy level requirements for the fuzzy task to be processed and data privacy entropy quantization. The fuzzy privacy level requirements are used to constrain the degree of privacy protection of the fuzzy task to be processed during transmission, and the data privacy entropy quantization is used to obtain the probability and level of data leakage of the fuzzy task to be processed during data transmission. The joint optimization model is solved using a hybrid unloading algorithm combining improved adaptive genetics, Monte Carlo simulation, and neural networks. The fuzzy task information to be processed is shown in the following formula: ; In the formula, Indicates pending tasks, subscript n Indicates the task number in the set. The amount of fuzzy input data for the task to be processed. The number of fuzzy CPU cycles required to process the task to be processed is here. , To pessimistically estimate the lower bound, , To optimistically estimate the upper bound, and This constitutes a confidence interval for the actual value. Opportunity Constraints It is a probability measure function, whose value quantifies the degree of probability that the input fuzzy conditions are satisfied; under fuzzy chance constraints, the optimization objective is transformed into minimizing Meanwhile, the objective function based on fuzzy chance constraints and the privacy entropy constraint are respectively transformed into: , ; in, and These are preset confidence levels, used to measure the confidence level of whether the privacy constraints and objective function constraints are satisfied; subsequently, the joint optimization model can be transformed into the following form: ; The transmitted data is divided into several segments of length. A block of bits; let This indicates that the task is performed through the m-th wireless network. The number of data blocks transmitted; The calculation formula is: , in Let represent the upper limit function; This represents the probability of leakage of k data blocks on the m-th wireless network; This represents the data transmission ratio of the nth task on the mth wireless network; assuming privacy leakage events follow a Poisson distribution. The calculation formula is: , in Represents the average privacy leakage on the m-th wireless network, where... It is the base of the natural logarithm; through the m-th wireless network and the task Privacy entropy associated with data transmission Calculated using the following formula: ; Fuzzy task privacy level requirements are used to constrain the degree of privacy protection during task transmission. The requirements for data privacy and security protection during transmission are divided into three levels: "low," "medium," and "high." Different security levels correspond to different fuzzy entropy thresholds. , and By calculating the privacy entropy threshold and tasks Privacy entropy The difference quantifies user satisfaction with privacy protection, and is expressed as... Negative values ​​indicate that the user's privacy and security level meets the threshold requirements, while positive values ​​indicate insufficient privacy protection.

2. The MEC task offloading method based on opportunity constraints according to claim 1, characterized in that, The latency and energy consumption model is used to obtain the latency and energy consumption cost of the task to be processed during data transmission and processing at the offloading location. The latency and energy consumption model enables data transmission between the terminal and the edge server, as well as between the edge server and the cloud server, through multiple wireless networks simultaneously.

3. The MEC task offloading method based on opportunity constraints according to claim 1, characterized in that, The opportunity constraint model is used for the conversion between the determination of the amount of fuzzy task data and the number of CPU cycles, and the determination of the opportunity constraint form of the fuzzy privacy model.

4. The MEC task offloading method based on opportunity constraints according to claim 1, characterized in that, The latency and energy consumption model includes a communication model and a computation model. The communication model is used to obtain the latency cost of the fuzzy task to be processed, and the computation model obtains the energy consumption cost of the task to be processed.

Citation Information

Patent Citations

  • Task unloading method for uncertain MEC environment

    CN118509920A

  • MEC task unloading method and device combining resource allocation and privacy protection

    CN118921700A