A mobile edge computing task offloading and allocation method

Abstract

The application discloses a mobile edge computing task unloading and distribution method, which is applied to a mobile edge computing task unloading and distribution system. The mobile edge computing resource unloading and distribution system comprises a plurality of mobile user devices, a scheduling center node, a plurality of edge servers and a cloud server which are distributed in sequence. The scheduling center node is used for transmitting the tasks unloaded by the mobile user devices to the edge servers and the cloud server for execution. The method has a unified scheduling center node which is arranged between the mobile user devices and the edge servers, integrates all the unloading tasks of the edge servers, uniformly distributes the calculation, constructs a minimum model of the maximum value of the system time delay, solves the optimization problem by designing an iterative algorithm, obtains the best unloading decision, and distributes the tasks of the mobile user devices to the edge servers or the cloud server, so that the time delay is obviously reduced, the energy consumption is reduced, and the efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of mobile edge computing technology, specifically relating to a method for unloading and allocating mobile edge computing tasks. Background Technology

[0002] In recent years, with the rapid development of 5G and mobile smart devices, mobile applications have become increasingly diversified, including image recognition, online video games, augmented and virtual reality, and more. Simultaneously, the development of the Internet of Things (IoT) has led to a rapid increase in the number of mobile devices. More and more applications demand high computing speeds, which mobile devices often cannot meet. Mobile edge computing technology refers to deploying computing and storage resources at the edge of the mobile network to provide an IT service environment and cloud computing speeds, thereby offering users ultra-low latency and high bandwidth network service solutions. Compute offloading refers to migrating resource-intensive computing tasks to nearby, resource-rich infrastructure for processing, ensuring a high-quality user experience.

[0003] In existing edge computing technologies, the entity that issues offloading decisions is not clearly defined, and most studies assume that offloading decisions are provided by local devices. Local devices have limitations in computing speed, so algorithms that theoretically perform well are difficult to deploy in real-world scenarios due to their high complexity. Secondly, specifying the optimal offloading decision should be system-wide; relying on local devices to obtain information about user nodes and edge servers within the current system is difficult, and sharing information would incur additional communication overhead. Summary of the Invention

[0004] The purpose of this invention is to address the problems raised in the background art by proposing a method for unloading and allocating mobile edge computing tasks.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] This invention proposes a method for offloading and allocating mobile edge computing tasks, applied to a mobile edge computing task offloading and allocation system. The mobile edge computing resource offloading and allocation system includes a plurality of mobile user devices distributed sequentially, a scheduling center node, a plurality of edge servers, and a cloud server. The scheduling center node is used to transfer the tasks offloaded by each mobile user device to the edge servers and the cloud server for execution. The mobile edge computing task offloading and allocation method, applied to the scheduling center node, includes:

[0007] The algorithm obtains the computing speed and CPU cycles required to execute tasks on the edge server, the CPU cycles required for local task execution on the mobile user device, the local computing speed, the task size and transmission rate to the scheduling center node, the CPU cycles required for task execution on the cloud server, the computing speed, the number of tasks arriving at the edge server per unit time, the average queue length of tasks waiting, and the latency of transmission from the scheduling center node to the cloud service. An optimization model is then constructed to minimize the maximum system latency. The optimization model is as follows:

[0008]

[0009] Among them, T i local Indicates the latency of local task execution on a mobile user device. This represents the transmission delay from the mobile user equipment to the dispatch center node. T represents the latency of task execution on the edge server. x,c T represents the total execution latency of the cloud server. q This represents the average queue wait time for the edge server.

[0010] The optimal unloading decision is obtained by solving the optimization model.

[0011] Based on the optimal offloading decision obtained, tasks for mobile user devices are assigned to edge servers or cloud servers.

[0012] Preferably, the local execution task latency of the mobile user equipment is expressed as follows:

[0013]

[0014] in,

[0015]

[0016]

[0017] in, f represents the CPU cycles required for mobile user equipment i to perform a local task. i local Let ε represent the local computing speed of mobile user equipment i, and let ε represent the CPU cycles required to execute one bit of data in the local execution task of mobile user equipment. d represents the size of the task data that mobile user device i needs to process locally. ij α represents the task data size of subtask j of mobile user equipment i. ij This represents the offloading decision of subtask j within the overall task of mobile user equipment i.

[0018] The overall task offloading decision for mobile user equipment i is represented as follows:

[0019]

[0020] Where, α i =0 indicates that all tasks of mobile user equipment i are executed locally, 0 < α i <1 indicates that part of the task of mobile user equipment i is transmitted to the edge server for execution, and another part is executed locally, α i =1 indicates that all tasks of mobile user equipment i are transmitted to the edge server for execution.

[0021] Preferably, the transmission delay from the mobile user equipment to the dispatch center node is expressed as follows:

[0022]

[0023] and,

[0024]

[0025]

[0026] Where, r i,k d represents the transmission rate of mobile user equipment i. i Let B represent the total data size of all tasks for mobile user equipment i, B represent the transmission channel bandwidth between mobile user equipment and the scheduling center node, k represent the number of sub-channels, and p represent the total data size of mobile user equipment i. i h represents the signal transmission power of mobile user equipment i. k σ represents the channel gain of the sub-channel. 2 I represents noise power. x,k p indicates interference caused by other mobile user equipment in the vicinity. x This represents the signal transmission power of other nearby mobile user equipment, where N represents the number of mobile user equipment.

[0027] Preferably, the task execution latency of the edge server is expressed as follows:

[0028]

[0029] in, This represents the CPU cycles required for an edge server or cloud server to execute x tasks. This indicates the computing speed of the edge server.

[0030] Preferably, the total execution latency of the cloud server is expressed as follows:

[0031]

[0032] and,

[0033]

[0034] in, f represents the latency of data transmission from the scheduling center node to the cloud service. C This indicates the computing speed of the cloud server. This indicates the latency of task execution on the cloud server.

[0035] Preferably, the average queue wait time of the edge server is expressed as follows:

[0036]

[0037] in,

[0038]

[0039]

[0040]

[0041]

[0042]

[0043] Among them, L q Let λ represent the average queue length for tasks arriving at the edge server, λ represent the number of tasks arriving at the edge server per unit time, M represent the number of edge servers, and S represent the number of tasks received by the edge server, where S = M, ρ * P0 represents the rate at which edge servers execute tasks, P0 represents the probability that each edge server is idle, and μ represents the number of tasks that an edge server can execute per unit of time. This represents the average computing speed of all edge servers and cloud servers. This represents the latency of all tasks unloaded to the scheduling center node. This indicates the CPU cycles required for the edge server to execute a single task.

[0044] Preferably, the optimization model is solved to obtain the optimal unloading decision, including:

[0045] The optimization model was solved using the simulated annealing algorithm. The simulated annealing algorithm was randomly initialized, with the initial temperature defined as t. init The final temperature is t end The cooling coefficient is ξ, and the proportional parameter is... Simultaneously, the offloading decision vector Q = {P, Ω} is randomly initialized, where the set of task offloading decisions for each mobile user equipment is P = {α1, α2, ..., α...}. N}, the set of edge server numbers selected by the unloaded task N sub This represents the total number of all unloaded subtasks.

[0046] Calculate the local task execution latency, transmission latency, edge server task execution latency, total cloud server execution latency, and average queue waiting latency of the edge server based on the input initialization parameters. Calculate D based on the optimization model. optimal The value is Y old The iteration count is initialized to episode and input into the simulated annealing algorithm for iteration. When t init When t approaches a stable value, the random perturbation generates a new unloading decision vector Q. new and output D optimal The value is Y new If Y new <Y old Then D optimal =Y new Q = Q new Until the number of iterations reaches episode and the temperature reaches t end At this point, the decision vector Q is no longer updated, and the optimal output D is reached. optimal And the optimal decision vector Q.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] This method uses a unified scheduling center node deployed between mobile user equipment and various edge servers. It integrates all offloading tasks from the edge servers and uniformly allocates computation. The computational offloading problem is constructed as a model that minimizes the maximum system latency. An iterative algorithm is designed to solve this optimization problem, obtain the best offloading decision, and allocate the tasks of mobile user equipment to edge servers or cloud servers, thereby significantly reducing latency, reducing energy consumption, and improving efficiency. Attached Figure Description

[0049] Figure 1 This is a block diagram of the mobile edge computing task unloading and allocation method of the present invention;

[0050] Figure 2 This is a schematic diagram of the scheduling center node of the present invention;

[0051] Figure 3 This is a flowchart of the simulated annealing algorithm of the present invention;

[0052] Figure 4 This is an algorithm diagram of the simulated annealing algorithm used in this invention. Detailed Implementation

[0053] 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 some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] It should be noted that when a component is referred to as being "connected" to another component, it can be directly connected to the other component or there may be an intervening component. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to limit the application.

[0055] like Figure 1-4 As shown, a method for offloading and allocating mobile edge computing tasks includes a mobile edge computing task offloading and allocation system. The mobile edge computing resource offloading and allocation system includes a plurality of mobile user devices, a scheduling center node, a plurality of edge servers, and a cloud server distributed sequentially. The scheduling center node is used to transfer the tasks offloaded by each mobile user device to the edge servers and the cloud server for execution. The mobile edge computing task offloading and allocation method, applied to the scheduling center node, includes:

[0056] The algorithm obtains the computing speed and CPU cycles required to execute tasks on the edge server, the CPU cycles required for local task execution on the mobile user device, the local computing speed, the task size and transmission rate to the scheduling center node, the CPU cycles required for task execution on the cloud server, the computing speed, the number of tasks arriving at the edge server per unit time, the average queue length of tasks waiting, and the latency of transmission from the scheduling center node to the cloud service. An optimization model is then constructed to minimize the maximum system latency. The optimization model is as follows:

[0057]

[0058] Among them, T i local Indicates the latency of local task execution on a mobile user device. This represents the transmission delay from the mobile user equipment to the dispatch center node. T represents the latency of task execution on the edge server. x,c T represents the total execution latency of the cloud server. q This represents the average queue wait time for the edge server.

[0059] It should be noted that, as Figure 1As shown, the system includes N mobile user devices (Users), M edge servers (MECs), 1 cloud server, and 1 dispatch center node (DCN). The dispatch center node is located between the users and the edge servers, and the distance between the dispatch center node and each edge server is less than the distance between the dispatch center node and the cloud server. Since the distance between the dispatch center node and the edge servers is short and the dispatch center node has high transmission power, the latency caused by the dispatch center node transmitting tasks to the edge servers is ignored.

[0060] like Figure 2 The diagram shows the module composition and working principle of the Dispatch Center Node (DCN). Figure 2 The user terminal in this context refers to a mobile user equipment (MUE). The scheduling center node includes an information module, a decision logic module, a data acquisition module, a transmission module, and a queue buffer. The information module stores the computing speed of each edge server, the computing speed of the cloud server, and the latency of transmission from the scheduling center node to the cloud service. The data acquisition module collects local computing speed, task size transmitted to the scheduling center node, and transmission rate. The information and data acquisition modules send the obtained information to the decision logic module. The decision logic module obtains the CPU cycles required for edge servers to execute tasks, the CPU cycles required for local task execution, the CPU cycles required for cloud server task execution, the number of tasks arriving at the edge server per unit time, and the average queue length for task waiting. Based on this information, the decision logic module generates decisions and allocates tasks. The queue buffer contains unloaded subtasks waiting to be transmitted and executed. The transmission module is responsible for allocating subtasks to edge servers or cloud servers based on the unloading decisions provided by the decision logic module.

[0061] At the start of the unloading process, each mobile user device submits data to the acquisition module, including task data size and channel information. The scheduling center node then runs a specific algorithm to determine which task should be unloaded by collecting information from all users, allocating computing resources accordingly, and measuring factors such as the total unloaded task data size and remote computing resources. When a portion of the unloaded task reaches the scheduling center node, it is forwarded to a designated edge server. Simultaneously, the scheduling center node monitors the computing speed of the designated edge server and dynamically adjusts the task unloading decision. If the edge server the task is destined for is occupied, the task is stored in a queuing buffer. Once the designated edge server resources are released or a better option becomes available, the scheduling center node immediately unloads the task.

[0062] Specifically, the latency of local task execution on mobile user equipment is represented as follows:

[0063]

[0064] in,

[0065]

[0066]

[0067] in, f represents the CPU cycles required for mobile user equipment i to perform a local task. i local Let ε represent the local computing speed of mobile user equipment i, and let ε represent the CPU cycles required to execute one bit of data in the local execution task of mobile user equipment. d represents the size of the task data that mobile user device i needs to process locally. ij α represents the task data size of subtask j of mobile user equipment i. ij α represents the subtask j in the total task of mobile user equipment i that uses the offloading decision. ij ={0,1} represents the offloading decision of the j-th subtask in the total task of mobile user equipment i.

[0068] The overall task offloading decision for mobile user equipment i is represented as follows:

[0069]

[0070] Where, α i =0 indicates that all tasks of mobile user equipment i are executed locally, 0 < α i <1 indicates that part of the task of mobile user equipment i is transmitted to the edge server for execution, and another part is executed locally, α i =1 indicates that all tasks of mobile user equipment i are transmitted to the edge server for execution. For example, α i =0.6 means that 60% of the tasks of mobile user device i are executed on the edge server and 40% are executed locally.

[0071] The transmission delay from mobile user equipment to the dispatch center node is represented as follows:

[0072]

[0073] and,

[0074]

[0075]

[0076] Where, r i,k d represents the transmission rate of mobile user equipment i. i Let B represent the total data size of all tasks for mobile user equipment i, B represent the transmission channel bandwidth between mobile user equipment and the scheduling center node, k represent the number of sub-channels, and p represent the total data size of mobile user equipment i.i h represents the signal transmission power of mobile user equipment i. k σ represents the channel gain of the sub-channel. 2 I represents noise power. x,k p indicates interference caused by other mobile user equipment in the vicinity. x This represents the signal transmission power of other nearby mobile user equipment, where N represents the number of mobile user equipment.

[0077] The task execution latency of the edge server is represented as follows:

[0078]

[0079] in, This represents the CPU cycles required for an edge server or cloud server to execute x tasks. This indicates the computing speed of the edge server, and is set according to the mainstream parameter magnitude in existing research, generally 2.5*10^7 bits per second.

[0080] The total execution latency of the cloud server is expressed as follows:

[0081]

[0082] and,

[0083]

[0084] in, This represents the latency of data transmission from the scheduling center node to the cloud service (since the scheduling center node and the cloud server are far apart, the transmission latency between the scheduling center node and the cloud server needs to be taken into account), f C The computing speed of the cloud server is set according to the mainstream parameter levels in existing research, typically 5 * 10^7 bits per second. This indicates the latency of task execution on the cloud server.

[0085] Although all tasks are transmitted simultaneously, they will arrive at the edge server at different times due to differences in data size. Therefore, when a task arrives, the edge server may still be processing an earlier-arriving task. Queuing latency must be considered. In this application, it is assumed that the number of tasks processed by each edge server is unlimited, and the queue length is unlimited. The system follows a First Come First Service (FCFS) principle.

[0086] The average queue wait time of the edge server is represented as follows:

[0087]

[0088] in,

[0089]

[0090]

[0091]

[0092]

[0093]

[0094] Among them, L q Let λ represent the average queue length for tasks arriving at the edge server, λ represent the number of tasks arriving at the edge server per unit time, M represent the number of edge servers, and S represent the number of tasks received by the edge server, where S = M, ρ * P0 represents the rate at which edge servers execute tasks, P0 represents the probability that each edge server is idle, and μ represents the number of tasks that an edge server can execute per unit of time. This represents the average computing speed of all edge servers and cloud servers (this application assumes that the time for tasks to arrive at the edge servers follows a Poisson distribution. The computing speed of the edge servers is processed to calculate the average computing speed of the edge and cloud servers). This represents the latency of all tasks unloaded to the scheduling center node. This represents the CPU cycles required for the edge server to execute a single task (the CPU cycles required for a single task are equivalent to the mathematical expectation). Since the cloud server has sufficient computing resources, the latency when the task arrives at the cloud server is not considered.

[0095] The optimal unloading decision is obtained by solving the optimization model.

[0096] Specifically, the optimization model is solved using a simulated annealing algorithm. The process of the simulated annealing algorithm is as follows: Figure 3 As shown, the specific approach is as follows: First, an initial temperature is set, and a solution is initialized at this temperature. The objective function (referring to the optimization model in this application) value is then calculated based on the initialized solution. Subsequently, the number of iterations is set for each temperature, and a new solution is generated in each iteration, with the corresponding objective function value calculated. If the objective function value of the new solution is better than the initial solution, the initial solution and initial objective function value are replaced with the new solution and the new objective function value. Otherwise, the new solution is accepted according to the Metropolis criterion, with the probability of acceptance as follows:

[0097]

[0098] Here, f(ω) and f(ω') represent the initial and latest values ​​of the objective function.

[0099] Specifically, the simulated annealing algorithm is randomly initialized, and the initial temperature is defined as t. init The final temperature is t end The cooling coefficient is ξ, and the proportional parameter is... Simultaneously, the offloading decision vector Q = {P, Ω} is randomly initialized, where the set of task offloading decisions for each mobile user equipment is P = {α1, α2, ..., α...}. N}, the set of edge server numbers selected by the unloaded task N sub This represents the total number of all unloaded subtasks.

[0100] Calculate the local task execution latency, transmission latency, edge server task execution latency, total cloud server execution latency, and average queue waiting latency of the edge server based on the input initialization parameters. Calculate D based on the optimization model. optimal The value is Y old The iteration count is initialized to episode and input into the simulated annealing algorithm for iteration. When t init When t approaches a stable value, the random perturbation generates a new unloading decision vector Q. new and output D optimal The value is Y new If Y new <Y old Then D optimal =Y new Q = Q new And t is also reassigned to tξ, until the iteration count reaches episode and the temperature reaches t. end At this point, the decision vector Q is no longer updated, and the optimal output D is reached. optimal And the optimal decision vector Q.

[0101] Initialization parameters include: the number of mobile user devices and edge servers, the local computing rate of mobile user devices, the computing rate of each edge server, the computing rate of the cloud server, the total data size of all tasks of each mobile user device, the transmission channel bandwidth between mobile user devices and the scheduling center node, the channel gain, the transmit power of each mobile user device, and the initial offload decision vector.

[0102] Based on the optimal offloading decision obtained, tasks for mobile user devices are assigned to edge servers or cloud servers.

[0103] Specifically, the task of unloading is preferentially transferred to the edge server for processing because the edge server is geographically closer to the user and there are more of them than the cloud server. However, when the scheduling center node finds that all the edge servers to which the task is sent are busy and the execution latency of the task sent to the remaining edge server exceeds the execution latency of the task sent to the cloud server, the new task can be handed over to the cloud server for processing.

[0104] This method uses a unified scheduling center node deployed between mobile user equipment and various edge servers. It integrates all offloading tasks from the edge servers and uniformly allocates computation. The computational offloading problem is constructed as a model that minimizes the maximum system latency. An iterative algorithm is designed to solve this optimization problem, obtain the best offloading decision, and allocate the tasks of mobile user equipment to edge servers or cloud servers, thereby significantly reducing latency, reducing energy consumption, and improving efficiency.

[0105] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0106] The embodiments described above are merely specific and detailed examples of the embodiments described in this application, and should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the appended claims.

Claims

1. A method for unloading and allocating mobile edge computing tasks, applied to a mobile edge computing task unloading and allocation system, characterized in that: The mobile edge computing resource offloading and allocation system includes a plurality of mobile user devices, a scheduling center node, a plurality of edge servers, and a cloud server distributed sequentially. The scheduling center node is used to transfer the tasks offloaded by each of the mobile user devices to the edge servers and the cloud server for execution. The mobile edge computing task offloading and allocation method is applied to the scheduling center node and includes: The algorithm obtains the computing speed and CPU cycles required to execute tasks on the edge server, the CPU cycles required for local task execution on the mobile user device, the local computing speed, the task size and transmission rate to the scheduling center node, the CPU cycles and computing speed required for task execution on the cloud server, the number of tasks arriving at the edge server per unit time, the average queue length of tasks waiting, and the latency of transmission from the scheduling center node to the cloud service. An optimization model is then constructed to minimize the maximum system latency. The optimization model is as follows: Among them, T i local Indicates the latency of local task execution on a mobile user device. This represents the transmission delay from the mobile user equipment to the dispatch center node. T represents the latency of task execution on the edge server. x,c T represents the total execution latency of the cloud server. q This represents the average queue wait latency of the edge servers; Solve the optimization model to obtain the optimal unloading decision; Based on the optimal offloading decision obtained, tasks for mobile user devices are assigned to edge servers or cloud servers.

2. The mobile edge computing task offloading and allocation method as described in claim 1, characterized in that: The local task execution latency of the mobile user equipment is expressed as follows: in, in, f represents the CPU cycles required for mobile user equipment i to perform a local task. i local Let ε represent the local computing speed of mobile user equipment i, and let ε represent the CPU cycles required to execute one bit of data in the local execution task of mobile user equipment. d represents the size of the task data that mobile user device i needs to process locally. ij α represents the task data size of subtask j of mobile user equipment i. ij This represents the offloading decision of subtask j within the overall task of mobile user equipment i; The overall task offloading decision for mobile user equipment i is represented as follows: Where, α i =0 indicates that all tasks of mobile user equipment i are executed locally, 0 < α i <1 indicates that part of the task of mobile user equipment i is transmitted to the edge server for execution, and another part is executed locally, α i =1 indicates that all tasks of mobile user equipment i are transmitted to the edge server for execution.

3. The mobile edge computing task offloading and allocation method as described in claim 2, characterized in that: The transmission delay from the mobile user equipment to the scheduling center node is expressed as follows: and, Where, r i,k d represents the transmission rate of mobile user equipment i. i Let B represent the total data size of all tasks for mobile user equipment i, B represent the transmission channel bandwidth between mobile user equipment and the scheduling center node, k represent the number of sub-channels, and p represent the total data size of mobile user equipment i. i h represents the signal transmission power of mobile user equipment i. k σ represents the channel gain of the sub-channel. 2 I represents noise power. x,k p indicates interference caused by other mobile user equipment in the vicinity. x This represents the signal transmission power of other nearby mobile user equipment, where N represents the number of mobile user equipment.

4. The mobile edge computing task offloading and allocation method as described in claim 3, characterized in that: The task execution latency of the edge server is expressed as follows: in, This represents the CPU cycles required for an edge server or cloud server to execute x tasks. This indicates the computing speed of the edge server.

5. The mobile edge computing task offloading and allocation method as described in claim 4, characterized in that: The total execution latency of the cloud server is expressed as follows: and, in, f represents the latency of data transmission from the scheduling center node to the cloud service. C This indicates the computing speed of the cloud server. This indicates the latency of task execution on the cloud server.

6. The mobile edge computing task offloading and allocation method as described in claim 5, characterized in that: The average queue wait time of the edge server is expressed as follows: in, Among them, L q Let λ represent the average queue length for tasks arriving at the edge server, λ represent the number of tasks arriving at the edge server per unit time, M represent the number of edge servers, and S represent the number of tasks received by the edge server, where S = M, ρ * P0 represents the rate at which edge servers execute tasks, P0 represents the probability that each edge server is idle, and μ represents the number of tasks that an edge server can execute per unit of time. This represents the average computing speed of all edge servers and cloud servers. This represents the latency of all tasks unloaded to the scheduling center node. This indicates the CPU cycles required for the edge server to execute a single task.

7. The mobile edge computing task offloading and allocation method as described in claim 6, characterized in that: Solving the optimization model to obtain the optimal unloading decision includes: The optimization model was solved using the simulated annealing algorithm. The simulated annealing algorithm was randomly initialized, with the initial temperature defined as t. init The final temperature is t end The cooling coefficient is ξ, and the proportional parameter is... Simultaneously, the offloading decision vector Q = {P, Ω} is randomly initialized, where the set of task offloading decisions for each mobile user equipment is P = {α1, α2, ..., α...}. N }, the set of edge server numbers selected by the unloaded task N sub This represents the total number of all unloaded subtasks; Calculate the local task execution latency, transmission latency, edge server task execution latency, total cloud server execution latency, and average queue waiting latency of the edge server based on the input initialization parameters. Calculate D based on the optimization model. optimal The value is Y old The iteration count is initialized to episode and input into the simulated annealing algorithm for iteration. When t init When t approaches a stable value, the random perturbation generates a new unloading decision vector Q. new and output D optimal The value is Y new If Y new <Y old Then D optimal =Y new Q = Q new Until the number of iterations reaches episode and the temperature reaches t end At this point, the decision vector Q is no longer updated, and the optimal output D is reached. optimal And the optimal decision vector Q.

Images