A method and device for MEC task offloading oriented to interval uncertainty

Abstract

The application discloses a kind of MEC task unloading methods and devices for interval uncertainty, it is related to edge computing technical field.The method includes: obtaining the task parameter of task to be handled, terminal resource parameter and MEC resource parameter, and the uncertain parameter in it is expressed as interval parameter;According to unloading decision variable and interval parameter, construct interval uncertain optimization model including energy consumption, time delay and user side cost;Interval energy consumption target is converted into interval center value target and interval radius value target based on interval sequence relation, and interval constraint is converted into deterministic constraint based on interval possibility, to obtain deterministic multi-objective optimization model;Dual-layer optimization unloading algorithm is used to solve, to obtain target unloading strategy.The application can consider energy consumption, time delay, cost and resource constraint in the case where parameter exists interval uncertainty, improve the robustness of task unloading.

Description

Technical Field

[0001] This application relates to the field of edge computing technology, and in particular to a task offloading decision-making technology based on interval uncertainty modeling. Background Technology

[0002] Multi-access edge computing (MEC) is a computing paradigm that provides computing and storage capabilities closer to the terminal to reduce round-trip latency, alleviate core network pressure, and improve energy efficiency. For computationally intensive and latency-sensitive mobile applications, relying solely on the terminal for execution is often costly due to battery capacity and heat dissipation limitations. Therefore, a decision must typically be made between "local execution" and "offloading to edge servers for execution." This decision is influenced by multiple factors, including wireless bandwidth, task data volume, return traffic, and the processing frequency of both the terminal and the edge.

[0003] However, real-world network environments exhibit widespread parameter fluctuations and uncertainties: key quantities such as task data volume, required Central Processing Unit (CPU) cycles, transmission bandwidth, and edge server computing resource allocation are difficult to obtain precisely and dynamically change with service load and wireless environment. Modeling these quantities as deterministic constants often fails to meet energy budget requirements in complex scenarios; simultaneously, end-to-end latency constraints and pay-as-you-go billing constraints further complicate the offloading decision problem. These types of problems are inherently combinatory and nonlinear, making them highly difficult to solve.

[0004] Existing technologies mainly follow two paths: one is to model and solve for the joint optimization of energy consumption and latency under deterministic settings with known parameters; the other is to introduce stochastic or fuzzy paradigms to handle uncertainties. However, in real deployments, the fluctuations of the above key parameters mostly stem from non-probabilistic uncertainties such as measurement errors, environmental changes, and configuration changes, making it difficult to obtain reliable probability distributions or accurately characterize them using membership functions.

[0005] Therefore, there is an urgent need to provide a method for offloading edge computing tasks by using interval quantities to characterize uncertain parameters, in order to solve the above-mentioned problems in the existing technology. Summary of the Invention

[0006] This application provides a method and apparatus for MEC task offloading oriented to interval uncertainty, which is used to solve the problem that existing MEC task offloading methods usually rely on deterministic parameters, probability distributions or membership functions for modeling when there are fluctuations in task parameters and resource parameters. This makes it difficult to adapt to non-probabilistic uncertainties with unknown statistical distributions, which leads to the offloading strategy easily failing to meet end-to-end latency constraints and reducing energy consumption optimization effect in real network environments.

[0007] In a first aspect, embodiments of this application provide a method for offloading MEC tasks accommodating interval uncertainty, including: The task parameters, terminal resource parameters, and MEC resource parameters of the task to be processed are obtained, and the uncertain parameters in the task parameters, terminal resource parameters, and MEC resource parameters are represented as interval parameters. The uncertain parameters include at least the task data volume, the number of CPU cycles required by the task, the transmission bandwidth, and the CPU frequency allocated to the task by the edge server. The terminal resource parameters include the local CPU frequency of the terminal, which is a deterministic parameter used to calculate the local execution latency and local execution energy consumption of the task when it is executed locally on the terminal. Based on the offloading decision variables and the interval parameters, an interval uncertain optimization model is constructed. The interval uncertain optimization model includes the energy consumption and latency corresponding to local execution, and the energy consumption, latency and user-side cost corresponding to offloading to the edge server for execution. The interval uncertain optimization model aims to minimize system energy consumption and includes end-to-end latency constraints, user-side cost constraints and MEC resource constraints. Based on the interval order relationship, the interval energy consumption target in the interval uncertain optimization model is converted into the interval center value target and the interval radius value target, and the interval constraints in the interval uncertain optimization model are converted into deterministic constraints based on the interval possibility degree, thus obtaining a deterministic multi-objective optimization model; The deterministic multi-objective optimization model is solved using a two-layer optimization unloading algorithm. The outer layer of the two-layer optimization unloading algorithm performs a multi-objective search for candidate unloading decisions based on the interval center value objective and the interval radius value objective. The inner layer of the two-layer optimization unloading algorithm calculates the upper and lower bounds of the objective function and the constraint function in the interval domain corresponding to the uncertain parameters, and feeds back the upper and lower bounds to the outer layer. Based on the search results of the outer layer, a target unloading strategy that satisfies the deterministic constraint is determined. The target unloading strategy is used to instruct the task to be processed to be executed locally on the terminal or unloaded to an edge server for execution.

[0008] In one possible implementation, the interval parameters include a task data volume interval, a CPU cycle count interval, a bandwidth interval, and an edge CPU frequency interval. The task parameters may also include the amount of result data to be sent after task execution, and the MEC resource parameters may also include at least one of the following: terminal uplink transmit power, edge server downlink transmit power, uplink channel gain, downlink channel gain, and noise power spectral density.

[0009] In one possible implementation, when constructing the interval uncertainty optimization model, the local execution latency and local execution energy consumption of the task to be processed when it is executed locally on the terminal can be determined according to the unloading decision variables, as well as the upload latency, edge server receiving latency, edge computing latency, result return latency, user result reception latency, and unloading execution energy consumption when the task to be processed is unloaded to the edge server for execution. The end-to-end latency is determined based on the local execution latency, the upload latency, the edge server receiving latency, the edge computing latency, the result return latency, and the user result reception latency. The user-side cost is determined based on the computing resources allocated by the edge server for the task to be processed unloaded to the edge server for execution. The unloading execution energy consumption includes at least one of the energy consumption generated by the terminal uploading the task, the energy consumption generated by the edge server receiving the task, the energy consumption generated by the edge server executing the task, the energy consumption generated by the edge server returning the result, the energy consumption generated by the terminal receiving the result, and the terminal standby energy consumption. The local execution energy consumption includes at least one of the energy consumption generated by the terminal executing the task and the edge server standby energy consumption.

[0010] In one possible implementation, the MEC resource constraints may include at least one of the following: edge server CPU frequency resource constraints, total data volume constraints for offloaded tasks, local CPU cycle count constraints, edge CPU cycle count constraints, and transmission bandwidth resource constraints.

[0011] In one possible implementation, when obtaining a deterministic multi-objective optimization model, the upper and lower bounds of the interval energy consumption target can be determined within the interval domain corresponding to the uncertain parameters, and the interval center value target and interval radius value target can be determined based on the upper and lower bounds of the interval energy consumption target; alternatively, the upper and lower bounds of the constraint function corresponding to each interval constraint can be determined within the interval domain corresponding to the uncertain parameters, and the deterministic constraint corresponding to the interval constraint satisfying the preset possibility level can be determined based on the upper and lower bounds of the constraint function and the preset possibility level.

[0012] In one possible implementation, the outer layer of the bi-layer optimized unloading algorithm can employ an improved non-dominated sorting genetic algorithm to perform multi-objective search for candidate unloading decisions, while the inner layer can employ an improved adaptive genetic algorithm to calculate the upper and lower bounds of the objective function and constraint functions. Further, the inner layer can use enhanced optimal Latin hypercube sampling to generate initial candidate samples or initialize the population, and enhance the sampling capability for the boundary regions of uncertain parameter intervals through cosine mapping; the inner layer can also use a residual neural network surrogate model to approximate the objective function and constraint functions.

[0013] In one possible implementation, the improved non-dominated sorting genetic algorithm may include at least one of a phased hybrid crossover mechanism, an adaptive mutation probability mechanism, a decoding mechanism, a soft constraint penalty mechanism, and an elite re-evolution mechanism; the improved adaptive genetic algorithm may include at least one of an adaptive crossover mechanism, an adaptive mutation mechanism, and an elite boundary preservation strategy.

[0014] Secondly, embodiments of this application provide a MEC task offloading device for interval uncertainty, including a parameter acquisition module, a model building module, a model transformation module, a solution module, and a strategy determination module. The parameter acquisition module acquires task parameters, terminal resource parameters, and MEC resource parameters of the task to be processed, and represents the uncertain parameters among the task parameters, terminal resource parameters, and MEC resource parameters as interval parameters. The model building module constructs an interval uncertainty optimization model based on the offloading decision variables and the interval parameters. The model transformation module performs deterministic processing on the interval uncertainty optimization model to obtain a deterministic multi-objective optimization model. The solution module solves the deterministic multi-objective optimization model using a two-layer optimization offloading algorithm. The strategy determination module determines a target offloading strategy that satisfies deterministic constraints based on the search results from the outer layer.

[0015] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: This application's embodiments represent uncertain parameters such as task data volume, number of CPU cycles required for the task, transmission bandwidth, and CPU frequency allocated by the edge server to the task as interval parameters. This enables the characterization of non-probabilistic uncertainties in the MEC task offloading process without prior knowledge of parameter probability distributions or membership functions, thereby improving the adaptability of the task offloading method to parameter fluctuations in real network environments.

[0016] This application embodiment constructs an interval uncertainty optimization model based on offloading decision variables and interval parameters. The interval uncertainty optimization model simultaneously considers the energy consumption, latency, and user-side costs corresponding to local execution and offloading to the edge server for execution. This enables the task offloading decision to take into account system energy consumption, end-to-end latency, user-side costs, and MEC resource constraints, thereby improving the engineering feasibility of the offloading strategy.

[0017] This application's embodiments are based on interval order relations, transforming interval energy consumption targets into interval center value targets and interval radius value targets, and transforming interval constraints into deterministic constraints based on interval possibility, thus obtaining a deterministic multi-objective optimization model; at the same time, a two-layer optimization offloading algorithm is adopted to combine the outer layer multi-objective search with the inner layer upper and lower bound evaluation, thereby reducing system energy consumption and improving the robustness of the target offloading strategy while satisfying deterministic constraints. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This application provides a schematic diagram of an MEC task unloading application scenario. Figure 2 A schematic flowchart of a MEC task offloading method for interval uncertainty provided in an embodiment of this application; Figure 3 This application provides a schematic diagram of a conversion process for an interval uncertainty optimization model. Figure 4 This application provides a schematic diagram of a two-layer optimized unloading algorithm. Figure 5 A schematic diagram of the inner layer upper and lower bound calculation process provided in this application embodiment; Figure 6 This is a schematic diagram of a MEC task offloading device for interval uncertainty provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are all within the protection scope of this application.

[0021] It should be noted that in the description of the embodiments of this application, the terms "first," "second," etc., are used only for distinguishing descriptive purposes and should not be construed as indicating or implying relative importance. Unless otherwise expressly defined, the terms "connection," "communication connection," etc., should be interpreted broadly; for example, they can refer to a direct connection or an indirect connection through an intermediate device or network. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0022] Figure 1 This is a schematic diagram illustrating an application scenario for MEC task unloading provided in an embodiment of this application. For example... Figure 1As shown, this application scenario may include a terminal device 100, an MEC server 200, and a wireless access network 300. The terminal device 100 can communicate with the MEC server 200 through the wireless access network 300 to enable the local execution or offloading execution of the task 101 to be processed.

[0023] For example, the terminal device 100 can be a mobile phone, tablet computer, laptop computer, vehicle terminal, industrial control terminal, drone terminal, or other electronic device with computing and communication capabilities, and this application embodiment does not specifically limit it. The terminal device 100 may contain one or more tasks 101 to be processed. The tasks 101 to be processed can be computationally intensive tasks, latency-sensitive tasks, or tasks that simultaneously possess computationally intensive and latency-sensitive characteristics.

[0024] In one possible implementation, the task 101 to be processed can be executed locally on the terminal device 100 or offloaded to the MEC server 200 via the wireless access network 300. When the task 101 to be processed is executed locally on the terminal device 100, the task processing mainly occupies the local computing resources of the terminal device 100; when the task 101 to be processed is offloaded to the MEC server 200 for execution, the task processing may involve tasks such as uploading task data, receiving data on the edge server, performing calculations on the edge server, transmitting calculation results back, and receiving calculation results on the terminal device 100.

[0025] The MEC server 200 can be deployed on the side close to the terminal device 100, such as on the base station side, access network side, or edge equipment room side. The MEC server 200 may include computing resources 201 and storage resources 202. The computing resources 201 can be used to execute the pending tasks 101 that are offloaded to the MEC server 200, and the storage resources 202 can be used to store task data, calculation results, intermediate data, or configuration information required during the task offloading process.

[0026] Figure 1 In This is used to represent multiple pending tasks or multiple task units in a task queue in terminal device 100, where n represents the number of pending tasks. Figure 1The transmission bandwidth is used to characterize the communication bandwidth resources used for task data upload and calculation result feedback between the terminal device 100 and the MEC server 200. The calculation frequency resources are used to characterize the CPU frequency resources that the MEC server 200 can allocate to the unloaded tasks. The target unloading strategy is used to determine whether each task 101 to be processed adopts a local execution mode or an unloaded execution mode to the MEC server 200. When the target unloading strategy indicates local execution, the corresponding task 101 to be processed is executed on the terminal device 100 side. When the target unloading strategy indicates unloaded execution, the corresponding task 101 to be processed is transmitted to the MEC server 200 side for execution via the wireless access network 300.

[0027] In this embodiment, the wireless access network 300 can provide a data transmission channel between the terminal device 100 and the MEC server 200. Whether the task 101 to be processed is executed locally on the terminal device 100 or offloaded to the MEC server 200 for execution can be determined according to the target offloading strategy. Thus, even when there is uncertainty in the task parameters and MEC resource parameters, the corresponding execution method for the task 101 to be processed can be determined.

[0028] like Figure 2 As shown in the embodiments of this application, a method for offloading MEC tasks oriented towards interval uncertainty may include the following steps.

[0029] S101, obtain the task parameters, terminal resource parameters and MEC resource parameters of the task to be processed, and represent the uncertain parameters in the task parameters, terminal resource parameters and MEC resource parameters as interval parameters.

[0030] In this embodiment of the application, the task to be processed can be a task generated by the terminal device that requires computational processing. For the user... i The j For each pending task, the corresponding task parameters can be obtained. The task parameters may include at least one of the following: task data volume, number of CPU cycles required for the task, and amount of result data to be sent after the task is executed.

[0031] In this embodiment, terminal resource parameters and MEC resource parameters can also be obtained. The terminal resource parameters may include the terminal's local CPU frequency, which is a deterministic parameter used to calculate the local execution latency and local execution energy consumption of the task to be processed when it is executed locally on the terminal. The MEC resource parameters may include at least one of the following: transmission bandwidth, CPU frequency allocated to the task to be processed by the edge server, terminal uplink transmit power, edge server downlink transmit power, uplink channel gain, downlink channel gain, and noise power spectral density.

[0032] Considering that in real-world MEC application scenarios, parameters such as task data volume, number of CPU cycles required for a task, transmission bandwidth, and CPU frequency allocated by the edge server to the task being processed may be affected by factors such as changes in service content, wireless link fluctuations, multi-user concurrent tasks, and edge server load status, and are therefore difficult to determine precisely, in this embodiment, the uncertain parameters mentioned above can be represented as interval parameters.

[0033] For example, the task data volume can be represented as a task data volume range, the number of CPU cycles required by the task can be represented as a CPU cycle range, the transmission bandwidth can be represented as a bandwidth range, and the CPU frequency allocated by the edge server to the task to be processed can be represented as an edge CPU frequency range. Here, the task data volume range represents the range between the minimum and maximum possible values ​​of the task data volume; the CPU cycle range represents the range between the minimum and maximum possible values ​​of the number of CPU cycles required by the task; the bandwidth range represents the range between the minimum and maximum possible values ​​of the transmission bandwidth; and the edge CPU frequency range represents the range between the minimum and maximum possible values ​​of the CPU frequency allocated by the edge server to the task to be processed.

[0034] Through the above methods, the embodiments of this application can characterize the non-probabilistic uncertainties in the task parameters and MEC resource parameters of the task to be processed without relying on the parameter probability distribution or membership function, providing a parameter basis for the subsequent construction of an interval uncertainty optimization model.

[0035] It should be noted that the upper and lower bounds of the above-mentioned interval parameters can be determined based on historical task data, real-time measurement data, network status estimation results, terminal device operating status, edge server load status, or preset engineering experience values. This application embodiment does not impose specific limitations on this. By representing the above-mentioned task parameters and MEC resource parameters as interval parameters, the parameter fluctuation range can be preserved in the subsequent modeling process, thereby improving the adaptability of the task offloading strategy to uncertain environments.

[0036] S102, construct an interval uncertainty optimization model based on the unloading decision variables and the interval parameters.

[0037] For users The A pending task can have its unload decision variables set. ,in, This indicates that the task to be processed is executed locally on the terminal device. This indicates that the pending task will be unloaded to the MEC server for execution. This is achieved by unloading decision variables. This allows for the unification of local execution and unloaded execution methods into the same optimization model.

[0038] In one possible implementation, a latency model, an energy consumption model, and a user-side cost model for the task to be processed can be constructed based on the offloading decision variables and interval parameters, respectively. An interval uncertainty optimization model can then be constructed based on these models. The interval uncertainty optimization model can have the goal of minimizing system energy consumption and includes end-to-end latency constraints, user-side cost constraints, and MEC resource constraints.

[0039] The following provides an exemplary illustration of the delay model.

[0040] If user Execute the task locally, the first The execution time for each task is:

[0041] in, Indicates user The The local execution latency range of each pending task when it is executed locally on the terminal; Indicates user The The range of CPU cycles required for each pending task; and These represent the lower and upper bounds of the CPU cycle count interval, respectively. Indicates user The local CPU frequency of the corresponding terminal device.

[0042] Since the number of CPU cycles required for a task is difficult to determine precisely due to the influence of task type, data size and running status, the number of CPU cycles required for the task is represented as an interval parameter when making a task unloading decision.

[0043] In one possible implementation, the uplink transmission rate between the terminal device and the MEC server, the uplink receiving rate of the edge server, the downlink transmission rate of the edge server, and the downlink receiving rate of the terminal device can be determined based on the transmission bandwidth, transmission power, channel gain, and noise power spectral density.

[0044]

[0045] in, Indicates user The uplink transmission rate range corresponding to the terminal device, This indicates that the MEC server is for users. The uplink receiving rate range; This indicates that the MEC server is for users. downlink transmission rate range Indicates user This corresponds to the downlink receive rate range of the terminal device. Without distinguishing between protocol overhead differences, within the same uplink... and Equal, in the same downlink and equal. and These represent the allocation to users. The lower bound and upper bound of the channel bandwidth; and These are allocated to the terminal devices. The ratio of uplink to downlink bandwidth resources; Indicates user Uplink transmit power; This indicates that the edge server is for the user. Downlink transmit power; This indicates that the edge server receives user data. Uplink receive power during task upload; Indicates user The downlink receiving power when the corresponding terminal device receives the calculation results; and Representing users respectively The uplink channel gain and downlink channel gain; It is the noise power spectral density.

[0046] If user Offload the task to the edge server for execution, and upload the first... The time required for each task is:

[0047] in, Indicates user i Uploaded j The upload latency interval required for each pending task to be sent to the edge server; Indicates allocation to user The channel bandwidth range; Indicates user of The range of task data volume for the first task to be processed; and These represent the lower bound and the upper bound of the task data volume, respectively.

[0048] Edge server receives the first The time for each task is:

[0049] in, This indicates that the edge server receives the first... The required reception delay interval for each pending task; Task The execution time on the edge server is:

[0050] in, Indicates the first The computational latency interval of each pending task when executed on the edge server side; This indicates that the edge server is assigned to the user. The CPU frequency range for each pending task and These represent the lower and upper bounds of the CPU frequency allocated by the edge server to the task to be processed, respectively.

[0051] Edge server transmission The time required for the results of each task is:

[0052] in, This indicates that the edge server is sending data to the user. i Return j The result return latency interval required for each task's calculation results; User receives the The time required to calculate the results of each task is:

[0053] in, Indicates user i The j The time required to calculate the results of each task; Indicates user i The j The amount of result data that needs to be sent to the terminal device after the task is completed; The task unloading latency includes the task upload latency, edge server task reception latency, edge server computation latency, result transmission latency, and user result reception latency, expressed as:

[0054] The total latency consists of two parts. One part is the local execution latency, and the other part is the latency of offloading to the edge service:

[0055] For example, the process of constructing an energy consumption model is as follows: Terminal computing power consumption generated by user equipment executing locally:

[0056] in, This indicates the energy consumption range generated by local calculations on the user equipment. It is the capability coefficient of the user equipment chip architecture, which depends on the user equipment chip architecture.

[0057] During the execution of the task locally, the edge server will incur basic standby power consumption:

[0058] in, Indicates the user during the local execution of the task. i The j The standby power consumption range generated by each task on the edge server; This indicates the latency of the task executing locally. This represents the standby power of the edge server. This portion of energy consumption is unrelated to task scheduling, but it is not negligible in the total system energy consumption.

[0059] The total energy consumption for local execution is:

[0060] in, Indicates user i The j The total local execution energy consumption range corresponding to each task when it is selected for local execution includes the terminal local computing energy consumption and the edge server standby energy consumption; When a terminal device transmits a task to the MEC server, the energy consumption generated during the unloading process includes transmission energy consumption, MEC server computing energy consumption, and receiving energy consumption.

[0061] The energy consumption of the transmission task is:

[0062] in, Indicates user i Uploaded j The transmission energy consumption range generated by each task; The energy consumption of the edge server receiving tasks is:

[0063] in, This indicates that the edge server receives user data. i The uploaded number j The receiving power consumption range generated by each task; The energy consumption expression for a MEC server is:

[0064] in, Indicates user i The j The computational energy consumption range generated when a task to be processed is executed on the edge server side; It is the capability coefficient of the edge server chip architecture, which depends on the edge server chip architecture.

[0065] The energy consumption for the edge server to send the result is:

[0066] in, This indicates that the edge server is sending data to the user. i Return j The results generated during the calculation of each task are returned to the energy consumption range; The energy consumption for the user to receive the results is:

[0067] in, Indicates user i Receive the j The receiving power consumption range generated when calculating the results of each task; While the task is being executed on the edge server, the local terminal will incur basic standby power consumption:

[0068] in, Indicates the user during the task unloading process. i The j The standby power consumption range of each task generated on the local terminal; This indicates the standby power of the local terminal; This indicates that the MEC server is sending messages to the user. Downlink transmit power when transmitting results; Indicates user i The downlink receiving power when the corresponding terminal device receives the calculation results.

[0069] When unloading a task, the energy consumption of the unloading execution is as follows:

[0070] in, Indicates user i The j The total unloading execution energy consumption range generated when a pending task is unloaded to an edge server for execution; Total energy consumption consists of two parts: local energy consumption and energy consumption offloaded to edge servers.

[0071] in, This represents the range of total system energy consumption corresponding to all pending tasks; For example, the process of constructing a cost model is as follows: In this embodiment, the resource fees charged to users are linearly related to the resource usage allocated to their applications, following the principle of "pay more for more use". For tasks offloaded to the edge for execution, their costs are calculated based on the equivalent benchmark duration of computing resources actually consumed at the edge, and billed according to the corresponding unit price.

[0072]

[0073] in, This indicates the range of resource usage costs corresponding to the unloading task; Indicates user The The equivalent computation time for each pending task when executed on the edge server side. This indicates that the edge server is assigned to the user. The CPU frequency range for each pending task Indicates user The The range of CPU cycles required for each pending task. Indicates the base CPU frequency or minimum CPU frequency. This indicates the base price.

[0074] For example, the user-side cost can be determined based on the edge computing resources consumed by the task to be processed and the preset unit price. The user-side cost can vary with the CPU frequency allocated to the task by the edge server, the number of CPU cycles required by the task, and the base price, and can be expressed as a range. Therefore, the range of variation of the user-side cost can be described under a range of uncertain parameters.

[0075] In this embodiment, the optimization objective is to minimize the total energy consumption of the task. At the same time, multiple constraints such as latency, cost, frequency, data size, local energy consumption, edge energy consumption and bandwidth are considered to establish a single-objective multi-constraint optimization model.

[0076] Optimization goal:

[0077] Indicates the first The user The unloading decision variables for each task. This indicates that the task is offloaded to the edge server. This indicates that the task is executed locally.

[0078] For example, the eight constraints are as follows: (C1) This constraint restricts execution to either a terminal or an edge server.

[0079] (C2) This constraint limits the task latency to a maximum permissible threshold for total latency.

[0080] (C3) This constraint limits the cost charged to users to no more than the maximum allowable threshold.

[0081] (C4) This constraint limits the total offloading frequency resources used by all offloading tasks on the edge server to no more than the maximum allowed threshold of total CPU frequency.

[0082] (C5) This constraint limits the total amount of data that all unloading tasks need to upload to the edge server to not exceed the maximum allowed threshold for the total task data size.

[0083] (C6) This constraint limits the total CPU cycle requirement of all locally executed tasks to a maximum allowed threshold that does not exceed the total number of CPU cycles for locally executed tasks.

[0084] (C7) This constraint limits the total CPU cycle requirement of all offloading tasks on the edge server to no more than the maximum allowed threshold for the total number of CPU cycles of tasks executed on the edge.

[0085] (C8) This constraint limits the total bandwidth consumption required by all unloading tasks during uploading to a maximum allowed threshold for total bandwidth.

[0086] in, N Represents a set of users; M Represents a set of tasks; Indicates the upper limit of end-to-end latency; This indicates the upper limit of user-side costs; This indicates the maximum allowed threshold of CPU frequency resources that can be used for task offloading on the edge server; This indicates the maximum amount of data that the unloading task can upload to the edge server. This indicates the maximum total number of CPU cycles required for a locally executed task. This indicates the maximum total number of CPU cycles required for the offloading task to be executed on the edge server side; Indicates the maximum allocatable total transmission bandwidth; This represents the function that constrains edge CPU frequency resources. This function represents the constraint on the amount of data uploaded by the unloading task. This function represents the CPU cycle count constraint for locally executed tasks. This represents the CPU cycle count constraint function for tasks executed on the edge server side; This represents the function that constrains transmission bandwidth resources.

[0087] S103, based on the interval order relationship, the interval energy consumption target in the interval uncertain optimization model is converted into the interval center value target and the interval radius value target, and based on the interval possibility degree, the interval constraints in the interval uncertain optimization model are converted into deterministic constraints, thus obtaining a deterministic multi-objective optimization model.

[0088] In this embodiment, since at least one of the following parameters—task data volume, number of CPU cycles required for the task, transmission bandwidth, and CPU frequency allocated to the task by the edge server—is represented as an interval parameter, the energy consumption objective function, latency constraint, user-side cost constraint, and MEC resource constraint constructed based on these interval parameters may all have interval uncertainties. To facilitate subsequent optimization, the interval uncertain optimization model can be determinized to obtain a deterministic multi-objective optimization model.

[0089] like Figure 3 As shown in the embodiments of this application, the interval energy consumption target and interval constraints in the interval uncertain optimization model can be determinized. Specifically, for the interval energy consumption target, it can be compared based on the interval order relationship, and the interval energy consumption target can be converted into an interval center value target and an interval radius value target. The interval center value target is used to characterize the central level of the interval energy consumption target, and the interval radius value target is used to characterize the fluctuation range of the interval energy consumption target under the influence of uncertain parameters. Thus, the interval energy consumption target, which is difficult to directly compare and optimize, can be converted into a deterministic target that can be used for multi-objective optimization solutions.

[0090] For interval constraints, deterministic processing can be performed based on the interval possibility level. Specifically, the deterministic constraint corresponding to the interval constraint satisfying the preset possibility level can be determined according to the constraint function, constraint upper limit, and preset possibility level corresponding to each interval constraint. Thus, uncertain constraints caused by uncertain parameters of the interval can be transformed into deterministic constraints.

[0091] Through the above processing, a deterministic multi-objective optimization model can be obtained by converting the interval energy consumption target into the interval center value target and the interval radius value target, and the interval constraint into deterministic constraint. This facilitates subsequent solution using a two-layer optimization unloading algorithm.

[0092] For example, the uncertain objective function is transformed as follows: Interval ordering can be used to rank uncertain objective functions. For minimization problems, a smaller interval center value indicates a lower energy consumption center level, and a smaller interval radius value indicates a smaller energy consumption fluctuation range. Therefore, the interval energy consumption objective can be transformed into an interval center value objective and an interval radius value objective, so as to simultaneously reduce the energy consumption center level and the energy consumption fluctuation range through multi-objective optimization.

[0093] Based on the interval order relationship, the uncertain objective function can be transformed into the following multi-objective optimization problem:

[0094] Hoping to find the optimal To minimize the midpoint and radius of the uncertain objective function, where:

[0095] in, This represents the center value of energy consumption; The radius value representing energy consumption; Indicates the upper bound of energy consumption; This represents the lower bound of energy consumption.

[0096] In the above formula, for any unloading decision variable The upper and lower bounds of the objective function, which are uncertainties caused by interval parameters, can be obtained through two optimization processes:

[0097] In one possible implementation, the interval constraints in the interval uncertain optimization model can be determinized based on the interval probability. Specifically, for interval constraints such as delay constraints, user-side cost constraints, and MEC resource constraints, the constraint function and its upper limit corresponding to each interval constraint can be determined first, and then the degree to which the constraint function satisfies the corresponding upper limit can be determined based on the interval probability.

[0098] For example, for latency constraints, user-side cost constraints, and MEC resource constraints, the degree to which they satisfy the corresponding upper limit of constraints can be determined according to the interval possibility formula, and they can be converted into deterministic constraints under a preset possibility level.

[0099] The formula for the probability of an interval is:

[0100] in, To constrain the upper limit, For interval numbers, Let be the upper bound of the interval number. It is the lower bound of the interval number.

[0101] It should be noted that for uncertain constraints caused by interval parameters, the upper and lower bounds of the constraint functions can be determined within the interval domain corresponding to the uncertain parameters, and then converted into deterministic constraints using the interval possibility formula. The upper and lower bounds of each constraint function can be calculated during the inner optimization process of the subsequent two-layer optimization unloading algorithm.

[0102] S104, the deterministic multi-objective optimization model is solved using a two-layer optimization unloading algorithm.

[0103] In this embodiment, after the determinization process in S103, the original interval uncertain optimization model can be transformed into a deterministic multi-objective optimization model. The deterministic multi-objective optimization model aims to minimize the interval center value objective and the interval radius value objective, and includes deterministic constraints derived from the interval constraints. Since the deterministic multi-objective optimization model involves unloading decision variables and calculating the upper and lower bounds of the objective function and constraint functions caused by the interval parameters, direct solution is difficult. Therefore, this embodiment employs a two-layer optimization unloading algorithm to solve the deterministic multi-objective optimization model.

[0104] In one possible implementation, the two-layer optimization unloading algorithm includes an outer layer optimization and an inner layer optimization. The outer layer optimization performs a multi-objective search on candidate unloading decisions to obtain candidate solutions that satisfy deterministic constraints and have relatively good corresponding interval center value and interval radius value objectives. The inner layer optimization evaluates the candidate solutions generated by the outer layer optimization within the interval domain corresponding to the uncertain parameters and calculates the upper and lower bounds of the objective function and constraint functions. The upper and lower bounds obtained by the inner layer optimization can be fed back to the outer layer optimization, enabling the outer layer optimization to update candidate unloading decisions based on the evaluation results under interval uncertainty.

[0105] like Figure 4 As shown, in the solution process of the two-layer optimization unloading algorithm, the outer layer optimization can first initialize the candidate unloading decisions; then, the outer layer optimization performs a multi-objective search on the candidate unloading decisions based on the interval center value objective and the interval radius value objective, and passes the current candidate solution to the inner layer optimization; the inner layer optimization calculates the upper and lower bounds of the objective function and the constraint function in the interval domain corresponding to the uncertain parameters, and feeds the upper and lower bounds back to the outer layer optimization; the outer layer optimization updates the candidate unloading decisions based on the upper and lower bounds until the preset termination condition is met.

[0106] In one possible implementation, the outer optimization can employ an improved non-dominated sorting genetic algorithm II (INNSGA-II) to perform a multi-objective search for candidate unloading decisions. Since the deterministic multi-objective optimization model includes two optimization objectives: the interval center value objective and the interval radius value objective, the outer optimization can obtain a solution set composed of multiple candidate solutions based on mechanisms such as non-dominated sorting and crowding, and determine the target unloading strategy from among them.

[0107] In one possible implementation, the inner-layer optimization can employ an improved adaptive genetic algorithm to calculate the upper and lower bounds of the objective function and constraint functions within the interval domain corresponding to the uncertain parameters. For each candidate solution generated by the outer-layer optimization, the inner-layer optimization can perform extremum searches for the objective function and constraint functions separately to obtain the corresponding upper and lower bounds. This approach provides the outer-layer optimization with performance evaluation results of candidate solutions under interval uncertainty.

[0108] like Figure 5 As shown, during the inner-layer upper and lower bound calculation process, candidate unloading decisions passed from the outer-layer optimization can be received first, and the interval domain corresponding to the uncertain parameters can be determined. Subsequently, enhanced optimal Latin hypercube sampling (EOLHS) can be used to generate initial candidate samples or initialize the population, and the sampling capability of the interval boundary region can be enhanced by cosine mapping, so that the generated samples can cover the extreme values ​​of uncertain parameters. Further, a residual neural network surrogate model can be trained or called to approximate the function values ​​of the objective function and constraint function under different unloading decisions and interval parameters. Finally, an improved adaptive genetic algorithm can be used to search for the extreme values ​​of the objective function and constraint function based on the surrogate model, thereby determining the corresponding upper and lower bounds, and feeding the upper and lower bounds back to the outer-layer optimization.

[0109] In one possible implementation, the EOLHS strategy can enhance its sampling capability for interval boundaries through a cosine mapping mechanism. For example, standardized samples can be used... Map to the interval boundary region corresponding to the uncertain parameter in the following way:

[0110] in, Represents a standardized sample; This represents the mapped sample; This represents element-wise multiplication. Through the aforementioned cosine mapping, the samples can have higher coverage in the boundary regions of the uncertain parameter interval, thus benefiting the upper and lower bounds of the inner optimization search objective function and constraint function.

[0111] In one possible implementation, the residual neural network surrogate model can be used to approximate the objective function and constraint function. The residual neural network surrogate model can be used for candidate unloading decisions and interval... The parameters are taken as input, and the estimated values ​​of the objective function or constraint function are output. By introducing a residual connection structure, the surrogate model's ability to express nonlinear mapping relationships can be enhanced, thereby reducing the computational overhead caused by frequent calls to the real model in inner optimization.

[0112] In one possible implementation, the improved non-dominated sorting genetic algorithm used in the outer layer may include at least one of the following: a phased hybrid crossover mechanism, an adaptive mutation probability mechanism, a decoding mechanism, a soft constraint penalty mechanism, and an elite re-evolution mechanism.

[0113] The phased hybrid crossover mechanism can be used to generate offspring individuals using a normal distribution crossover operator in the early stages of iteration to enhance population diversity and global search capability; and to generate offspring individuals using a distributed crossover operator in the later stages of iteration to improve local search capability and convergence efficiency. Furthermore, an adaptive perturbation of the covariance matrix can be introduced based on the distributed crossover operator to adjust the perturbation direction and intensity according to the current population distribution.

[0114] To achieve dynamic switching, this application uses the current iteration number as the basis. Gen With maximum number of iterations maxGen Relationship design cross strategy discriminant:

[0115] in, It is an empirical coefficient, usually taking a value in the range [0,1], used to adjust the switching time point.

[0116] In the early stages of the algorithm, to enhance global exploration capabilities and avoid premature convergence, this application employs a normal distribution crossover operator to generate offspring individuals.

[0117] Let the early first The mating pool is denoted as the population matrix:

[0118] in, Represents the set of real numbers; Indicates the population size; Indicates chromosome length. The mating pool is randomly shuffled and divided into rows. Individual and Post Individuals, and paired one by one in rows.

[0119] For any pair of parent vectors Perform normal distribution crossover to obtain two candidate offspring vectors:

[0120]

[0121] in, This controls the perturbation amplitude when generating new individuals, and is usually set to 1.481; It is a random variable that follows a normal distribution and is used to introduce random perturbations to increase the diversity of newly generated individuals.

[0122] In the later stages of the algorithm, to accelerate convergence and enhance local fine-grained search capabilities, this application employs a distributed crossover operator.

[0123] Let the later period be the first The mating pool is denoted as the population matrix:

[0124] in, Indicates the population size; Indicates chromosome length. The mating pool is randomly shuffled and divided into rows. Individual and Post Individuals, and paired one by one in rows.

[0125] For any two parent individuals Distributed crossover in the Generate offspring candidate values ​​at each gene locus The rules are:

[0126] in, Indicates the offspring generated after crossover in the th order. Continuous candidate values ​​for each gene locus; and Two parent individuals were in the first generation. The values ​​of each gene locus; For cross control parameters, It is a uniformly distributed random variable. It is generated randomly. and It can perturb the weighted combinations of parental genes, thereby increasing the diversity of offspring individuals and avoiding getting trapped in local optima.

[0127] To further improve the performance of local search in the later stages, this application incorporates distributed crossover results. Based on this, an adaptive perturbation of the covariance matrix is ​​introduced, and the generator is:

[0128] in, This represents the candidate population matrix generated after introducing adaptive perturbation of the covariance matrix; This represents the candidate population matrix obtained after distributed crossover. The dynamic disturbance intensity factor is updated as follows:

[0129] in, This is the maximum disturbance intensity coefficient, typically set to [0.5, 1.0]. This is the minimum disturbance intensity coefficient, typically set to [0.01, 0.1]. Let represent a random vector that follows a multivariate normal distribution with a mean of 0 and a covariance matrix of C, where C is the covariance matrix of the current parent population, calculated using the following formula:

[0130] in, This represents the number of individuals in the parent population used to calculate the covariance matrix; For the first i Individuals in a population; This is the population mean.

[0131] This mechanism can automatically learn the coupling relationship between various dimensions by using the statistical characteristics of variables in the parent individuals, adaptively guide the direction and magnitude of the perturbation, further accelerate the later local convergence, improve the precision and diversity of the solution, and prevent the algorithm from getting stuck in local optima.

[0132] The adaptive mutation probability mechanism can be used to dynamically adjust the mutation probability based on the current iteration number. For example, in the early stages of iteration, the mutation probability can be increased to enhance global exploration capabilities; in the later stages of iteration, the mutation probability can be decreased to reduce unnecessary perturbations, thereby improving convergence stability.

[0133] Let the first generation population

[0134] in, It indicates the population size, or the number of individuals in the population; Indicates chromosome length. For an individual The For each gene locus, if a mutation is triggered, the polynomial mutation generates a series of candidate gene values:

[0135] in, and These are the upper and lower bounds of the allowed values ​​for this gene locus, respectively.

[0136] Disturbance factor The calculation is as follows:

[0137] in, Uniformly distributed random numbers, The distribution index parameter is usually taken as... , The value is usually 0.5.

[0138] To ensure that the mutation probability changes dynamically with iteration, this application introduces an adaptive mutation probability. Defined as:

[0139] in, and These represent the minimum and maximum mutation probabilities, respectively, which are set to 0.1 and 1 in this application. Gen Indicates the current iteration number. max Gen This represents the maximum number of iterations.

[0140] The decoding mechanism can be used to truncate the boundaries of continuous candidate gene values ​​generated after crossover and mutation, and convert the truncated continuous candidate gene values ​​into binary genes that satisfy the binary unloading decision constraints. This allows candidate individuals obtained from crossover and mutation to meet the value requirements of the unloading decision variable.

[0141] Decoding involves two steps: The first step is boundary truncation: First, consecutive candidate values ​​are truncated gene by gene. The interval step is used to eliminate out-of-bounds values ​​caused by continuous operators, ensuring that each gene position can be interpreted as "the probability of taking 1".

[0142] The second step is binary decoding: Binary genes were obtained using Bernoulli sampling.

[0143] The soft constraint penalty mechanism can be used to add penalty terms to candidate individuals that do not meet deterministic constraints, thereby reducing the priority of infeasible individuals during the ranking process while preserving their potential information for maintaining population diversity. For example, for candidate individuals that do not meet the constraints, penalty terms can be added to their interval center value and interval radius value:

[0144] Among them, here This represents the center value of the interval, distinct from the aforementioned user-side cost; Indicates the radius value of the interval. and These represent the center value and radius value of the interval after adding the penalty term, respectively. and These represent the center value penalty intensity coefficient and the radius value penalty intensity coefficient, respectively. This represents the constraint penalty function.

[0145] The aforementioned elite re-evolution mechanism can be used to re-evaluate and locally fine-tune the elite individuals in the current Pareto front, thereby improving the stability and reliability of the final solution set. For example, the current elite individuals can be... Add local perturbation vector In order to obtain further evolved elite individuals :

[0146] in, This represents a small-amplitude local perturbation vector applied to elite individuals.

[0147] In one possible implementation, the improved adaptive genetic algorithm employed in the inner layer may include at least one of an adaptive crossover mechanism, an adaptive mutation mechanism, and an elite boundary preservation strategy. The adaptive crossover mechanism adjusts the crossover probability based on the fitness of the parent individuals; the adaptive mutation mechanism adjusts the mutation probability based on the individual's fitness; and the elite boundary preservation strategy preserves the optimal individual representing the extreme value of the objective function or constraint function at the end of each generation iteration.

[0148] Adaptive crossover operation:

[0149] in, This represents the larger fitness value among the two parent individuals; and These are the average fitness and maximum fitness of the current population, respectively. and These are the preset maximum and minimum crossover probabilities.

[0150] Adaptive mutation operation:

[0151] in, This represents the fitness value of the current individual; and These are the average fitness and maximum fitness of the current population, respectively. and These represent the maximum and minimum mutation probabilities, respectively.

[0152] For example, the elite boundary preservation strategy can be expressed as:

[0153] in, Indicates the first t The individual with the best fitness in the population; Indicates the first t The individuals to be replaced in the +1 generation population. By employing the elite boundary preservation strategy described above, we can avoid the loss of high-quality individuals used to characterize the extreme values ​​of the objective or constraint functions during crossover or mutation, thereby improving the stability of the upper and lower bound estimates.

[0154] Through the aforementioned two-layer optimized offloading algorithm, the embodiments of this application can combine multi-objective search for candidate offloading decisions with evaluation of the upper and lower bounds of objective and constraint functions within the uncertain parameter interval, thereby determining the target offloading strategy while satisfying deterministic constraints. This reduces system energy consumption and improves the robustness of the offloading strategy even when task parameters and MEC resource parameters exhibit interval uncertainties.

[0155] S105, Based on the search results of the outer layer, determine the target unloading strategy that satisfies the deterministic constraint.

[0156] In this embodiment, after the two-layer optimization offloading algorithm completes its iterations, the outer layer optimization can obtain one or more candidate solutions. These candidate solutions may include candidate offloading decisions. These candidate offloading decisions indicate whether each task to be processed is executed locally on the terminal device or offloaded to the MEC server for execution.

[0157] In one possible implementation, a target solution can be determined from candidate solutions that satisfy deterministic constraints, and a target offloading strategy can be determined based on the target solution. The deterministic constraints may include at least one of end-to-end latency constraints, user-side cost constraints, and MEC resource constraints. This avoids problems such as latency exceeding limits, user-side cost exceeding limits, or MEC resource occupancy exceeding limits during the actual execution of the determined target offloading strategy.

[0158] In this embodiment, the target unloading strategy may include the values ​​of unloading decision variables corresponding to each task to be processed. When the unloading decision variable corresponding to a task to be processed takes a first value, it can be determined that the task to be processed is executed locally on the terminal device; when the unloading decision variable corresponding to the task to be processed takes a second value, it can be determined that the task to be processed is unloaded to the MEC server for execution. For example, when the unloading decision variable... When, it indicates the user The Each pending task is executed locally on the terminal device; when the decision variable is unloaded... When, it indicates the user The The pending tasks are offloaded to the MEC server for execution.

[0159] In the above manner, the embodiments of this application can determine a target offloading strategy that meets deterministic constraints when there is range uncertainty in task parameters and MEC resource parameters. This allows the task to be processed to make a reasonable choice between executing locally on the terminal device and offloading to the MEC server for execution, and reduces system energy consumption while meeting end-to-end latency, user-side cost and MEC resource limitations.

[0160] Based on the same inventive concept, embodiments of this application also provide an MEC task offloading apparatus oriented towards interval uncertainty. This apparatus can be used to execute the MEC task offloading method oriented towards interval uncertainty described in any of the above embodiments.

[0161] Figure 6 This is a schematic diagram of a MEC task offloading device for addressing interval uncertainty, provided as an embodiment of this application. Figure 6 As shown, the MEC task offloading device 600 for interval uncertainty may include a parameter acquisition module 601, a model building module 602, a model transformation module 603, a solution module 604, and a strategy determination module 605.

[0162] The parameter acquisition module is used to acquire the task parameters, terminal resource parameters, and MEC resource parameters of the task to be processed, and to represent the uncertain parameters in the task parameters, terminal resource parameters, and MEC resource parameters as interval parameters; wherein, the uncertain parameters may include at least the task data volume, the number of CPU cycles required by the task, the transmission bandwidth, and the CPU frequency allocated to the task by the edge server; the terminal resource parameters include the terminal local CPU frequency, which is a deterministic parameter.

[0163] The model building module is used to construct an interval uncertainty optimization model based on the offloading decision variables and the interval parameters. The interval uncertainty optimization model may include the energy consumption and latency corresponding to local execution, and the energy consumption, latency, and user-side cost corresponding to offloading to the edge server for execution; the interval uncertainty optimization model may aim at minimizing system energy consumption and include end-to-end latency constraints, user-side cost constraints, and MEC resource constraints.

[0164] The model transformation module is used to perform deterministic processing on the interval uncertain optimization model to obtain a deterministic multi-objective optimization model. The deterministic processing includes: converting the interval energy consumption objective in the interval uncertain optimization model into an interval center value objective and an interval radius value objective based on interval order relations; and converting the interval constraints in the interval uncertain optimization model into deterministic constraints based on interval probability.

[0165] The solution module is used to solve the deterministic multi-objective optimization model using a two-layer optimization unloading algorithm. Specifically, the outer layer of the two-layer optimization unloading algorithm performs a multi-objective search for candidate unloading decisions based on the interval center value objective and the interval radius value objective. The inner layer of the two-layer optimization unloading algorithm calculates the upper and lower bounds of the objective function and constraint function within the interval domain corresponding to the uncertain parameters, and feeds back the upper and lower bounds to the outer layer.

[0166] The strategy determination module is used to determine a target unloading strategy that satisfies the deterministic constraints based on the search results from the outer layer. The target unloading strategy is used to instruct the task to be processed to be executed locally on the terminal device or unloaded to the MEC server for execution.

[0167] It should be noted that the specific functions and implementation processes of each module in the MEC task offloading device for interval uncertainty provided in this application embodiment can be found in the relevant descriptions in the foregoing method embodiments, and will not be repeated here. The above modules can be implemented by software, hardware, or a combination of software and hardware, and this application embodiment does not specifically limit them.

[0168] In one possible implementation, the above-described MEC task offloading method for interval uncertainty can also be implemented by an electronic device or a computer-readable storage medium. The electronic device may include a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the method described in any of the above embodiments; the computer-readable storage medium may store a computer program that, when executed by the processor, implements the method described in any of the above embodiments.

[0169] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for offloading MEC tasks accommodating interval uncertainty, characterized in that, include: The task parameters, terminal resource parameters, and MEC resource parameters of the task to be processed are obtained, and the uncertain parameters in the task parameters, terminal resource parameters, and MEC resource parameters are represented as interval parameters. The uncertain parameters include at least the task data volume, the number of CPU cycles required by the task, the transmission bandwidth, and the CPU frequency allocated to the task by the edge server. The terminal resource parameters include the local CPU frequency of the terminal, which is a deterministic parameter used to calculate the local execution latency and local execution energy consumption of the task when it is executed locally on the terminal. Based on the offloading decision variables and the interval parameters, an interval uncertain optimization model is constructed. The interval uncertain optimization model includes the energy consumption and latency corresponding to local execution, and the energy consumption, latency and user-side cost corresponding to offloading to the edge server for execution. The interval uncertain optimization model aims to minimize system energy consumption and includes end-to-end latency constraints, user-side cost constraints and MEC resource constraints. Based on the interval order relationship, the interval energy consumption target in the interval uncertain optimization model is converted into the interval center value target and the interval radius value target, and the interval constraints in the interval uncertain optimization model are converted into deterministic constraints based on the interval possibility degree, thus obtaining a deterministic multi-objective optimization model; The deterministic multi-objective optimization model is solved using a two-layer optimization unloading algorithm. The outer layer of the two-layer optimization unloading algorithm performs a multi-objective search for candidate unloading decisions based on the interval center value objective and the interval radius value objective. The inner layer of the two-layer optimization unloading algorithm calculates the upper and lower bounds of the objective function and the constraint function in the interval domain corresponding to the uncertain parameters, and feeds back the upper and lower bounds to the outer layer. Based on the search results of the outer layer, a target unloading strategy that satisfies the deterministic constraint is determined. The target unloading strategy is used to instruct the task to be processed to be executed locally on the terminal or unloaded to an edge server for execution.

2. The MEC task offloading method for interval uncertainty according to claim 1, characterized in that, The step of representing the uncertain parameters in the task parameters, terminal resource parameters, and MEC resource parameters as interval parameters includes: The task data volume is represented as a task data volume range, the number of CPU cycles required by the task is represented as a CPU cycle range, the transmission bandwidth is represented as a bandwidth range, and the CPU frequency allocated by the edge server to the task to be processed is represented as an edge CPU frequency range. The task parameters also include the amount of result data to be sent after the task is executed, and the MEC resource parameters also include at least one of the following: terminal uplink transmit power, edge server downlink transmit power, uplink channel gain, downlink channel gain, and noise power spectral density.

3. The MEC task offloading method for interval uncertainty according to claim 1, characterized in that, The step of constructing an interval uncertainty optimization model based on the unloading decision variables and the interval parameters includes: Based on the unloading decision variables, determine the local execution latency and local execution energy consumption of the task to be processed when it is executed locally on the terminal. Based on the unloading decision variables, determine the upload latency, edge server receiving latency, edge computing latency, result return latency, user result receiving latency, and unloading execution energy consumption when the task to be processed is unloaded to the edge server for execution; The end-to-end latency is determined based on the local execution latency, the upload latency, the edge server receiving latency, the edge computing latency, the result return latency, and the user receiving result latency. The user-side cost is determined based on the computing resources allocated by the edge server for the pending tasks offloaded to the edge server for execution. The energy consumption for unloading execution includes at least one of the following: energy consumption generated by the terminal uploading task, energy consumption generated by the edge server receiving the task, energy consumption generated by the edge server executing the task, energy consumption generated by the edge server transmitting the result back, energy consumption generated by the terminal receiving the result, and terminal standby energy consumption; the energy consumption for local execution includes at least one of the following: energy consumption generated by the terminal executing the task and edge server standby energy consumption.

4. The MEC task offloading method for interval uncertainty according to claim 1, characterized in that, The MEC resource constraints include at least one of the following: The total CPU frequency resources allocated to the unloading task on the edge server shall not exceed the CPU frequency resource limit. The total amount of data uploaded to the edge server by the unloading task shall not exceed the task data volume limit. The total number of CPU cycles required to execute a task locally shall not exceed the local CPU cycle limit. The total number of CPU cycles required for the offloading task on the edge server side shall not exceed the upper limit of edge CPU cycles. The total transmission bandwidth required for the unloading task to upload shall not exceed the bandwidth resource limit.

5. The MEC task offloading method for interval uncertainty according to claim 1, characterized in that, The method involves transforming the interval energy consumption objective in the interval uncertain optimization model into interval center value objectives and interval radius value objectives based on interval order relations, and transforming the interval constraints in the interval uncertain optimization model into deterministic constraints based on interval probability, thereby obtaining a deterministic multi-objective optimization model, including: Within the interval domain corresponding to the uncertain parameter, the upper and lower bounds corresponding to the interval energy consumption target are determined, and based on the upper and lower bounds corresponding to the interval energy consumption target, the interval center value and interval radius value of the interval energy consumption target are determined, so as to convert the interval energy consumption target into the interval center value target and the interval radius value target; Within the interval domain corresponding to the uncertain parameter, the upper and lower bounds of the constraint function corresponding to each interval constraint are determined, and the deterministic constraint corresponding to the interval constraint satisfying the preset possibility level is determined based on the upper bound of the constraint function, the lower bound of the constraint function, and the preset possibility level. The deterministic multi-objective optimization model is constructed by taking the minimization of the interval center value objective and the interval radius value objective as the objectives and combining the deterministic constraints.

6. The MEC task offloading method for interval uncertainty according to claim 5, characterized in that, The method of solving the deterministic multi-objective optimization model using a two-layer optimization unloading algorithm includes: The outer layer of the two-layer optimized unloading algorithm performs a multi-objective search for the candidate unloading decisions, and the outer layer employs an improved non-dominated sorting genetic algorithm. The inner layer of the two-layer optimization unloading algorithm calculates the upper and lower bounds of the objective function and the constraint function, and the inner layer uses an improved adaptive genetic algorithm. The upper and lower bounds calculated by the inner layer are fed back to the outer layer, so that the outer layer updates the candidate unloading decision based on the upper and lower bounds.

7. The MEC task offloading method for interval uncertainty according to claim 6, characterized in that, The inner layer of the dual-layer optimized unloading algorithm uses enhanced optimal Latin hypercube sampling to generate initial candidate samples or initialize the population, and uses cosine mapping to map the standardized samples to the boundary region of the interval domain corresponding to the uncertain parameters, so as to enhance the sampling capability for extreme cases. The inner layer of the two-layer optimized unloading algorithm also uses a residual neural network surrogate model to approximate the objective function and the constraint function. The residual neural network surrogate model takes the candidate unloading decisions generated by the outer layer and the interval parameters as inputs, and outputs the estimated value of the objective function or the constraint function.

8. A method for offloading MEC tasks accommodating interval uncertainty according to claim 6, characterized in that, The improved non-dominated sorting genetic algorithm includes at least one of the following: a phased hybrid crossover mechanism, an adaptive mutation probability mechanism, a decoding mechanism, a soft constraint penalty mechanism, and an elite re-evolution mechanism. The phased hybrid crossover mechanism is used to generate offspring individuals by using a normal distribution crossover operator in the early stage of the iteration and a distributed crossover operator in the later stage of the iteration, and an adaptive perturbation of the covariance matrix is ​​introduced on the basis of the distributed crossover operator. The adaptive mutation probability mechanism is used to adjust the mutation probability according to the current iteration number; The decoding mechanism is used to truncate the boundaries of consecutive candidate gene values ​​generated after crossover and mutation, and The continuous candidate gene values ​​after boundary truncation are converted into binary genes that satisfy the binary unloading decision constraints; The soft constraint penalty mechanism is used to add penalty terms to the interval center value and interval radius value of individuals that do not meet the deterministic constraints; The elite re-evolution mechanism is used to re-evaluate and fine-tune the elite individuals in the current Pareto front.

9. A method for offloading MEC tasks accommodating interval uncertainty according to claim 6, characterized in that, The improved adaptive genetic algorithm includes at least one of the following: adaptive crossover mechanism, adaptive mutation mechanism, and elite boundary preservation strategy; The adaptive crossover mechanism is used to adjust the crossover probability based on the fitness of the parent individuals. The adaptive mutation mechanism is used to adjust the mutation probability based on individual fitness. The elite boundary preservation strategy is used to preserve the boundary values ​​used to characterize the objective function or boundary value after each iteration. The optimal individual for finding the extreme value of the constraint function.

10. A MEC task offloading device for interval uncertainty, characterized in that, include: The parameter acquisition module is used to acquire the task parameters, terminal resource parameters, and MEC resource parameters of the task to be processed, and to represent the uncertain parameters in the task parameters, terminal resource parameters, and MEC resource parameters as interval parameters; The model building module is used to build an interval uncertainty optimization model based on the offloading decision variables and the interval parameters. The interval uncertainty optimization model includes the energy consumption and latency corresponding to local execution, and the energy consumption, latency and user-side cost corresponding to offloading to the edge server for execution. The model transformation module is used to perform deterministic processing on the interval uncertain optimization model to obtain a deterministic multi-objective optimization model; wherein, the deterministic processing includes: based on the interval order relationship, converting the interval energy consumption target in the interval uncertain optimization model into the interval center value target and the interval radius value target, and based on the interval possibility degree, converting the interval constraints in the interval uncertain optimization model into deterministic constraints; The solution module is used to solve the deterministic multi-objective optimization model using a two-layer optimization unloading algorithm. The outer layer of the two-layer optimization unloading algorithm performs a multi-objective search on candidate unloading decisions based on the interval center value objective and the interval radius value objective. The inner layer of the two-layer optimization unloading algorithm calculates the upper and lower bounds of the objective function and the constraint function in the interval domain corresponding to the uncertain parameters, and feeds back the upper and lower bounds to the outer layer. The strategy determination module is used to determine the target unloading strategy that satisfies the deterministic constraints based on the search results of the outer layer.

Images