Computational task offloading method based on multi-task dynamic membrane structure

Abstract

The present invention discloses a method for offloading computational tasks based on a multi-task dynamic membrane structure, comprising: constructing a membrane structure comprising twelve basic membranes, four non-basic membranes, and a skin membrane, wherein the four non-basic membranes are arranged in the skin membrane, and each non-basic membrane is provided with four basic membranes; initializing the four populations of the non-basic membranes and equally dividing them into the four basic membranes within the non-basic membrane; releasing the subpopulations to the corresponding non-basic membranes for merging to obtain a merged subpopulation; merging the four merged subpopulations in the skin membrane to obtain a merged population; and using an elite strategy to select a solution set from the merged population and output it through the skin membrane. The present invention employs a multi-task decomposition strategy to decompose the original constrained multi-objective optimization problem into multiple subtasks, which are then gradually evolved at different stages. Through the characteristics of independent parallelism within the membrane and information exchange between membranes, the global search capability of the population is effectively improved, ensuring that the optimal solution to the objective is achieved while satisfying complex constraints.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and in particular to a computing task offloading method based on a multi-task dynamic membrane structure. Background Art

[0002] Edge computing offloading is a technology that offloads computationally intensive tasks from users' mobile smart devices to edge servers or cloud servers for execution. This can improve user experience while reducing the cost of executing tasks locally. However, considering indicators such as latency and energy consumption generated during task execution to compare the losses and benefits of local execution and edge execution, computation offloading is not the optimal way to solve data processing problems for smart devices. Therefore, it is necessary to determine a computation offloading strategy that balances losses and benefits within a limited time to decide whether the computation task needs to be offloaded. Membrane computing, as a new research field in natural computing, is a computational model derived from the structure and function of living cells. Due to its distributed parallel computing characteristics, it is often introduced to solve some optimization problems and has demonstrated excellent performance and great potential. Therefore, it has significant application potential in solving computation offloading problems.

[0003] Existing technologies include edge computing offloading methods based on linear programming, deep reinforcement learning, and heuristic algorithms. Linear programming-based edge computing offloading methods suffer from limitations: not all computational offloading problems can be modeled as linear programming problems. Linear programming methods are also slow, making it difficult to find an appropriate offloading strategy within a limited timeframe. Deep reinforcement learning-based edge computing offloading methods require extensive training data to achieve more accurate offloading decisions, resulting in long computation times. Furthermore, the high-dimensionality and large-scale nature of computational offloading problems complicates deep reinforcement learning training. Furthermore, deep reinforcement learning generally suffers from low interpretability, which can make it difficult to ensure the reliability of offloading strategies. Heuristic algorithm-based edge computing offloading methods may exhibit varying optimization performance for different edge computing offloading models. Furthermore, since computational offloading problems are often multi-objective constrained optimization problems, the choice of constraint handling techniques can significantly impact the final results. Therefore, further research is needed to design constraint handling techniques to solve computational offloading problems and achieve more effective offloading solutions. Summary of the Invention

[0004] The embodiment of the present invention provides a computing task offloading method based on a multi-task dynamic membrane structure to solve the problems of the multi-user and multi-server edge computing offloading model in the prior art, the impracticality of the computing offloading strategy in the end-edge-cloud scenario, the complex structure, the slow and inefficient solution speed of the existing solution technology, and the limitations of the multi-task and multi-target processing.

[0005] On the one hand, an embodiment of the present invention provides a method for offloading computing tasks based on a multi-task dynamic membrane structure, comprising:

[0006] Constructing a membrane structure, the membrane structure comprising twelve basic membranes, four non-basic membranes and a skin membrane, wherein the four non-basic membranes are arranged in the skin membrane, and four basic membranes are arranged in each non-basic membrane;

[0007] Initializing the four populations of the non-basic membrane and then equally dividing them into the four basic membranes within the non-basic membrane;

[0008] evolving a subpopulation of the non-essential membrane and dissolving the essential membrane;

[0009] releasing the subpopulations to the corresponding non-basic membranes for merging to obtain a merged subpopulation;

[0010] evolving the merged subpopulation and dissolving the non-essential membrane;

[0011] merging the four merged subpopulations in the skin membrane to obtain a merged population;

[0012] Selecting a solution set from the merged population using an elite strategy and outputting it through the skin membrane;

[0013] The computing tasks of the Internet of Things are offloaded through the solution set.

[0014] In a possible implementation, the initializing and then equally dividing the four populations of the non-basic membrane into the four basic membranes within the non-basic membrane includes:

[0015] performing a non-dominated sorting of the population of the non-elementary membranes based on an optimization task;

[0016] selecting one of the four basic membranes among the non-basic membranes as an elite basic membrane;

[0017] Selecting elite individuals of one quarter of the population from the population through a fitness evaluation strategy calculation method based on SPEA2 and placing them into the elite basic membrane;

[0018] The remaining three quarters of the population are divided into three equal parts and placed into the remaining three basic membranes of the non-basic membrane to complete the division.

[0019] In a possible implementation, evolving the subpopulation of the non-basic membrane and dissolving the basic membrane includes:

[0020] The subpopulation is evolved by using a basic membrane population to explore multi-task effective information;

[0021] Dissolving the basic film into the corresponding non-basic film.

[0022] In one possible implementation, the releasing of the subpopulations into the corresponding non-basic membranes for merging to obtain the merged subpopulation is to dissolve one of the elite basic membranes and three basic membranes in the non-basic membrane in the non-basic membrane, and then releasing the subpopulations in the elite basic membrane and the subpopulations in the three basic membranes into the corresponding same non-basic membrane for merging to obtain the merged subpopulation.

[0023] In a possible implementation, evolving the merged subpopulation and dissolving the non-essential membrane includes:

[0024] evolving the merged subpopulation by optimizing the main task;

[0025] The four non-essential films were dissolved into the skin film.

[0026] In a possible implementation, the selecting a solution set from the merged population using an elite strategy and outputting the solution set through the skin membrane includes:

[0027] Selecting task execution location solutions from the merged population using an elite strategy as the solution set;

[0028] The solution set is output through the skin membrane as the final decision of safe unloading decision.

[0029] The computational task offloading method based on the multi-task dynamic membrane structure of the present invention has the following advantages:

[0030] (1) By adopting a multi-task decomposition strategy, the original constrained multi-objective optimization problem is decomposed into multiple subtasks and gradually evolved in different stages. The global search capability of the population is effectively improved by the characteristics of independent parallelism within the membrane and information exchange between membranes.

[0031] (2) Through multi-task decomposition strategies and multi-swarm optimization techniques, the trade-offs between latency, energy consumption, and security vulnerability costs can be more accurately balanced, ensuring that the optimal solution to the goal is achieved while satisfying complex constraints.

[0032] (3) By adopting a multi-population strategy and a three-stage evolution mechanism, the diversity of the population and the quality of the solution are significantly improved. The resulting solution set is superior to existing algorithms in terms of convergence and distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 A flowchart of a method for offloading computing tasks based on a multi-task dynamic membrane structure provided in an embodiment of the present application;

[0035] Figure 2 A schematic diagram of a membrane structure for a method for offloading computational tasks based on a multi-task dynamic membrane structure according to an embodiment of the present application;

[0036] Figure 3 Schematic diagram of the overall framework of a constrained multi-objective evolutionary algorithm based on a multi-task dynamic membrane structure for a computational task offloading method based on a multi-task dynamic membrane structure provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] Figure 1 A flowchart of a method for offloading computing tasks based on a multi-task dynamic membrane structure provided by an embodiment of the present invention; an embodiment of the present invention provides a method for offloading computing tasks based on a multi-task dynamic membrane structure, comprising:

[0039] Constructing a membrane structure, the membrane structure comprising twelve basic membranes, four non-basic membranes and a skin membrane, wherein the four non-basic membranes are arranged in the skin membrane, and four basic membranes are arranged in each non-basic membrane;

[0040] Initializing the four populations of the non-basic membrane and then equally dividing them into the four basic membranes within the non-basic membrane;

[0041] evolving a subpopulation of the non-essential membrane and dissolving the essential membrane;

[0042] releasing the subpopulations to the corresponding non-basic membranes for merging to obtain a merged subpopulation;

[0043] evolving the merged subpopulation and dissolving the non-essential membrane;

[0044] merging the four merged subpopulations in the skin membrane to obtain a merged population;

[0045] Selecting a solution set from the merged population using an elite strategy and outputting it through the skin membrane;

[0046] The computing tasks of the Internet of Things are offloaded through the solution set.

[0047] The initializing and equally dividing the four populations of the non-basic membrane into the four basic membranes within the non-basic membrane comprises:

[0048] performing a non-dominated sorting of the population of the non-elementary membranes based on an optimization task;

[0049] selecting one of the four basic membranes among the non-basic membranes as an elite basic membrane;

[0050] Selecting elite individuals of one quarter of the population from the population through a fitness evaluation strategy calculation method based on SPEA2 and placing them into the elite basic membrane;

[0051] The remaining three quarters of the population are divided into three equal parts and placed into the remaining three basic membranes of the non-basic membrane to complete the division.

[0052] The evolving the subpopulation of the non-basic membrane and dissolving the basic membrane comprises:

[0053] The subpopulation is evolved by using a basic membrane population to explore multi-task effective information;

[0054] Dissolving the basic film into the corresponding non-basic film.

[0055] The releasing of the subpopulations into the corresponding non-basic membranes for merging to obtain the merged subpopulation is to dissolve one of the elite basic membranes and three basic membranes in the non-basic membrane in the non-basic membrane, and then releasing the subpopulations in the elite basic membrane and the subpopulations in the three basic membranes into the corresponding same non-basic membrane for merging to obtain the merged subpopulation.

[0056] Evolving the merged subpopulation and dissolving the non-essential membrane comprises:

[0057] evolving the merged subpopulation by optimizing the main task;

[0058] The four non-essential films were dissolved into the skin film.

[0059] The method of using the elite strategy to select a solution set from the merged population and outputting it through the skin membrane includes:

[0060] Selecting task execution location solutions from the merged population using an elite strategy as the solution set;

[0061] The solution set is output through the skin membrane as the final decision of safe unloading decision.

[0062] For example, Figure 1 、 2 As shown, the non-basic membranes in the membrane structure handle different tasks for the same optimization problem. For the algorithm designed in this chapter, the four tasks handled are the constrained delay-energy optimization problem, the unconstrained delay-energy optimization problem, the constrained delay optimization problem, and the constrained energy optimization problem. In other words, for a constrained dual-objective optimization problem, the problem is decomposed from two dimensions. The first dimension is the constraint aspect, which is divided into two tasks, the constrained problem and the unconstrained problem, for collaborative optimization. The second dimension is the objective aspect, which separates the two objectives into two single-objective constrained optimization tasks for collaborative optimization. Therefore, for a constrained dual-objective optimization problem, it is ultimately divided into four related optimization tasks and assigned to four different non-basic membranes for optimization. For the same optimization task, such as the original constrained multi-objective optimization problem optimized in the non-basic membrane, its internal population is divided into four sub-populations, among which there is an elite sub-population. The elite subpopulation in the non-basic membrane resides within the basic membrane, containing the elite individuals of the entire population. The remaining three non-elite subpopulations each employ different mutation strategies through differential evolution, utilizing information from the elite population to generate offspring, achieving a balance between convergence and diversity. The optimization tasks in the remaining three non-basic membranes are optimized using the same approach. Thanks to the independent and parallel nature of each membrane, the optimization tasks in the four basic membranes can evolve in parallel, accelerating the extraction of information from different tasks. After evolving to a certain degree within the basic membrane, the basic membrane is dissolved, and the four subpopulations are transferred to their corresponding non-basic membranes and merged. At this point, the populations in the four non-basic membranes already contain valid information for their respective optimization tasks. Further optimization is then performed within each of these four populations using an auxiliary population strategy with dynamically changing constraint thresholds. This aims to leverage this information to solve the original constrained multi-objective optimization problem and ultimately identify the final task execution location solution. After the non-basic membrane populations have evolved, they are transferred to the skin membrane, where the final task execution location solution is selected. After obtaining this solution, the entire model construction problem becomes a three-objective constrained optimization problem that includes the cost of security vulnerabilities. For this constrained optimization problem, non-dominated sorting is used to optimize the security vulnerability cost and the delay and energy consumption caused by its security encryption in the skin membrane, thereby obtaining the final offloading decision of the model.

[0063] The specific method process is as follows Figure 3As shown, the overall framework of the constrained multi-objective evolutionary algorithm (MTDMS-CMOEA) based on multi-task dynamic membrane structure of the present application can be divided into three stages. First, the four non-basic membrane populations are initialized, and each non-basic membrane population is evenly divided so that it can be distributed to the four basic membranes within it. The division strategy is to first perform a non-dominated sorting of the entire population based on the corresponding optimization task, and use the fitness evaluation strategy calculation method based on SPEA2 to select elite individuals of one-quarter of the population and put them into the elite basic membrane, and divide the remaining three-quarters of the population into three equal parts and put them into the other three basic membranes. After the division is completed, the first stage of evolution is carried out through Algorithm 2. For the four subpopulations in each non-basic membrane, the basic membrane population is used to explore the effective information of multiple tasks for evolution. The purpose of the first stage of evolution is to obtain effective information of different optimization tasks without interference from other tasks by taking advantage of the independent and parallel characteristics within the membrane. In addition, the four subpopulations are respectively located in the four basic membranes in the non-basic membrane, and the convergence of the population is guaranteed by exchanging information between the membranes. After the first stage of evolution is completed, it enters the second stage and is carried out through Algorithm 3. At this time, all basic membranes are dissolved, and the internal subpopulations are released into the corresponding non-basic membranes, and merged in the non-basic membranes to generate subpopulations. Group, the populations in the four non-basic membranes are further evolved by merging sub-populations of non-basic membranes to accelerate the optimization of the main task. The main purpose of the second stage is to use the effective information of related tasks to guide the optimization of the original optimization problem of the main task; after the second stage of evolution, it enters the third stage and is evolved by Algorithm 4. At this time, the non-basic membrane is dissolved, and the four populations are merged in the skin membrane. An elite strategy is used to select a certain number of solution sets. The position of the task execution on this solution set has been determined, and the security vulnerability cost is only for the task to be unloaded. Therefore, a security vulnerability cost optimization problem is constructed on this solution set, and the skin membrane reconstruction optimization problem is used to solve the security unloading decision to solve the final unloading decision.

[0064] Among them, Algorithm 1 is the overall framework algorithm, and the pseudo code is as follows:

[0065] enter: (population size), (Problem Dimension), (maximum number of function evaluations)

[0066] Output: (Final population)

[0067]

[0068] while no-reached do

[0069] while the first stage is not over do

[0070] for 1to4

[0071] right Use Algorithm 2 to perform the first stage evolution;

[0072] end for

[0073] end while

[0074] while the second stage is not over do

[0075] for 1 to 4

[0076] right Use Algorithm 3 for the second stage of evolution;

[0077] end for

[0078] end while

[0079] ;

[0080] ;

[0081] Use Algorithm 4 for the third stage of evolution;

[0082] endwhile

[0083] return ;

[0084] The first stage is carried out in the basic membrane. Its main purpose is to explore the effective information in different optimization tasks. Therefore, after dividing the population into four sub-populations, they are assigned to four basic membranes and different mutation methods in the differential evolution algorithm are used to generate sub-population offspring in parallel, thereby enhancing the diversity of the population. The elite population is introduced to improve the convergence of the other three sub-populations through information exchange between membranes. is an individual randomly obtained from elite subpopulation 1, and Then, two non-repeated individuals are randomly selected from subpopulation 2. The mutation strategies of subpopulation 3 and subpopulation 4 are shown in formula 2.2 and formula 2.3 respectively. 2,i 、v 3,i 、v 4,i represents the mutant individuals generated after the mutation operation. The subscript numbers (2, 3, 4) correspond to the subpopulations. i can be understood as the index of the individual in the population, which is used to distinguish different individuals. x represents the individual in the population, like x 1,bestis the best individual in elite subpopulation 1; x 2,n 、x 2,r2 is a randomly selected individual from subpopulation 2, n and r2 are individual indices; x 3,n 、x 3,r3 Corresponding to the subpopulation of 3 individuals, x 4,i 、x 4,r2 、x 4,r3 Corresponding to the four individuals in the subpopulation, it is used to identify the individuals participating in the mutation in different subpopulations. F, F1, and F2 are scaling factors in the differential evolution algorithm. They control the scaling of the mutation vector, affecting the strength of the mutation operation and balancing the algorithm between exploration (diversified search) and exploitation (refinement using known information). Different F values ​​can be set to adapt to the subpopulation mutation requirements.

[0085] The pseudocode for the first phase is shown in Algorithm 2. The number of the non-basic membrane represents the optimization problem it addresses. The individual evaluation strategy and environment selection strategy depend on the optimization problem. Lines 3-6 calculate the fitness value for individual evaluation and environment selection based on the original constrained delay-energy optimization problem; lines 7-10 calculate the fitness value based on the unconstrained delay-energy optimization problem; lines 11-14 calculate the fitness value using delay and constraint violation as two objectives; lines 15-18 determine the fitness value based on energy consumption and constraint violation. After the iteration, line 21 merges the population and finally returns a population containing valid information for different optimization tasks. The formula in Algorithm 2 is as follows:

[0086] (2.1)

[0087] (2.2)

[0088] (2.3)

[0089] The pseudo code of Algorithm 2 is as follows:

[0090] enter: (population size), (Problem Dimension), (non-essential membrane populations), (number of function evaluations in the first stage), (Non-basic membrane number)

[0091] Output: (Final population)

[0092] Bundle Divided into four subpopulations;

[0093] while no-reached do

[0094] if then

[0095] Use (2.1) (2.2) (2.3) to generate offspring;

[0096] ;

[0097] Select the next generation based on the constraints of delay and energy consumption;

[0098] elseif then

[0099] Use (2.1) (2.2) (2.3) to generate offspring;

[0100] ;

[0101] Select the next generation based on latency and energy consumption issues;

[0102] elseif then

[0103] Use (2.1) (2.2) (2.3) to generate offspring;

[0104] ;

[0105] Select the next generation based on the time-constrained delay problem;

[0106] elseif then

[0107] Use (2.1) (2.2) (2.3) to generate offspring;

[0108] ;

[0109] Select the next generation based on the energy consumption constraint;

[0110] end if

[0111] end while

[0112] ;

[0113] return ;

[0114] The purpose of the second stage is to utilize the effective information in different optimization tasks to speed up the search for the best feasible solution for the main task, which is the task for the constrained delay energy consumption optimization problem. In addition, since the populations in multiple tasks are utilized in the optimization process, the diversity of the final population can be guaranteed. The pseudo code of the second stage is shown in Algorithm 3. When using the current mainstream constrained multi-objective optimization algorithm to solve the computing task offloading problem, it is found that when a completely unconstrained auxiliary population is used to assist in the optimization, it is often found that the constraints are difficult to meet. The continuous optimization goals of the auxiliary population will guide the main population to the non-feasible domain area, making it difficult for the population to find a feasible domain. Therefore, the second stage of the algorithm proposed in this chapter uses a dynamic constraint violation degree threshold as the environmental selection indicator of the auxiliary population. Its constraint violation degree threshold The calculation of is as follows (3.1) and changes dynamically, where represents the maximum number of iterations, is the initial constraint violation degree, and its value is the maximum constraint violation degree in the initial main population and auxiliary population. The parameter for controlling the rate of decline is set to 0.5. Lines 1-2 in the pseudocode initialize the auxiliary population and the constraint violation threshold. Line 4 updates the constraint violation threshold. Lines 5-6 generate offspring for the main and auxiliary populations, respectively. Lines 7-8 transfer knowledge between the two populations. Lines 9-10 use different environmental selection strategies to select the next generation. The main population uses a feasibility rule and fitness value, while the auxiliary population uses a constraint threshold to control the selection of offspring individuals.

[0115] Algorithm 3 Formula 3.1 is as follows:

[0116] (3.1)

[0117] The pseudo code of Algorithm 3 is as follows:

[0118] enter: (population size), (Problem Dimension), (population to be evolved), (number of function evaluations)

[0119] Output: (Final population)

[0120] ;

[0121] Initialize the constraint violation threshold to ;

[0122] while no-reached do

[0123] Update the constraint violation threshold according to (3.1);

[0124] Using differential evolution algorithm from Generated N / 2 offspring;

[0125] Using differential evolution algorithm from Generated N / 2 offspring;

[0126] ;

[0127] ;

[0128] Using the feasibility rule and fitness value from Selected N individual;

[0129] use As the constraint threshold from Selected N individual;

[0130] end while

[0131] return ;

[0132] The third stage mainly involves finding the computing task dimensions that require the addition of security encryption algorithms from the population that has obtained the computing task offloading location, thereby constructing a three-objective optimization problem with security vulnerability cost constraints and then performing optimization. Its pseudocode is shown in Algorithm 4, where lines 1-3 are used to determine the dimension of the optimization problem, line 4 initializes the population based on the problem dimension, and lines 5-9 use the feasibility rule and differential evolution algorithm for iterative optimization to find the final offloading strategy.

[0133] The pseudo code of Algorithm 4 is as follows:

[0134] enter: (population size), (unloading location population), (Number of function evaluations in the third stage)

[0135] Output: (Final population)

[0136] calculate The fitness value of ;

[0137] according to The values ​​are sorted to select the best individual;

[0138] according to The number of offloaded tasks determines the dimension of the security vulnerability cost optimization problem;

[0139]

[0140] while no-reached do

[0141] Using differential evolution algorithm from Generated N offspring;

[0142] ;

[0143] Using the feasibility rule and fitness value from Selected N individual;

[0144] end while

[0145] return ;

[0146] In one possible embodiment, Table 1 shows the HV index of the computational offloading problem under different numbers of IoT devices near each edge node. As can be seen from the table, MTDMS-CMOEA did not achieve the optimal solution for the offloading problem with K=10, but according to the rank sum test results, the algorithm achieved similar performance to IMTCMO, which achieved the optimal solution. As the number of IoT devices increases, the dimension of the decision variables also increases. Thanks to the ability of the membrane computing framework to extract effective information from multiple tasks in parallel and accelerate the main task to find the feasible domain through information exchange between membranes, the algorithm proposed in this application achieved optimal results for the remaining four scales of problems. The symbols "+", "−", and "≈" indicate that the results of another algorithm are significantly better, significantly worse, or similar to the performance of the proposed algorithm.

[0147] Table 1 HV index results of computing task offloading problem

[0148]

[0149] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including all changes and modifications that fall within the scope of the present invention and the preferred embodiments.

[0150] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A computational task offloading method based on a multi-task dynamic membrane structure, characterized in that: include: Constructing a membrane structure, the membrane structure comprising twelve basic membranes, four non-basic membranes and a skin membrane, wherein the four non-basic membranes are arranged in the skin membrane, and four basic membranes are arranged in each non-basic membrane; Initializing the four populations of the non-basic membrane and then equally dividing them into the four basic membranes within the non-basic membrane; evolving a subpopulation of the non-essential membrane and dissolving the essential membrane; releasing the subpopulations to the corresponding non-basic membranes for merging to obtain a merged subpopulation; evolving the merged subpopulation and dissolving the non-essential membrane; merging the four merged subpopulations in the skin membrane to obtain a merged population; Selecting a solution set from the merged population using an elite strategy and outputting it through the skin membrane; offloading computing tasks of the Internet of Things through the solution set; The initializing and equally dividing the four populations of the non-basic membrane into the four basic membranes within the non-basic membrane comprises: performing a non-dominated sorting of the population of the non-elementary membranes based on an optimization task; selecting one of the four basic membranes among the non-basic membranes as an elite basic membrane; Selecting elite individuals of one quarter of the population from the population through a fitness evaluation strategy calculation method based on SPEA2 and placing them into the elite basic membrane; The remaining three quarters of the population are equally divided into three parts and placed into the remaining three basic membranes of the non-basic membrane to complete the division; The releasing of the subpopulations into the corresponding non-basic membranes for merging to obtain the merged subpopulation is to dissolve one of the elite basic membranes and three basic membranes in the non-basic membrane in the non-basic membrane, and then releasing the subpopulations in the elite basic membrane and the subpopulations in the three basic membranes into the corresponding same non-basic membrane for merging to obtain the merged subpopulation.

2. The computing task offloading method based on a multi-task dynamic membrane structure according to claim 1 is characterized in that: The evolving the subpopulation of the non-basic membrane and dissolving the basic membrane comprises: The subpopulation is evolved by using a basic membrane population to explore multi-task effective information; Dissolving the basic film into the corresponding non-basic film.

3. The computing task offloading method based on a multi-task dynamic membrane structure according to claim 1 is characterized in that: Evolving the merged subpopulation and dissolving the non-essential membrane comprises: evolving the merged subpopulation by optimizing the main task; The four non-essential films were dissolved into the skin film.

4. The computing task offloading method based on a multi-task dynamic membrane structure according to claim 1 is characterized in that: The method of using the elite strategy to select a solution set from the merged population and outputting it through the skin membrane includes: Selecting task execution location solutions from the merged population using an elite strategy as the solution set; The solution set is output through the skin membrane as the final decision of safe unloading decision.

Images