A MEC task offloading method based on non-dominated sorting artificial hummingbird algorithm

Through the MEC task unloading method based on the non-dominant sorting artificial hummingbird algorithm, the problem of limited mobile edge computing resources is solved, efficient resource utilization and service quality improvement is achieved, system cost is reduced, and excellent convergence and distribution performance are demonstrated. .

CN118741606BActive Publication Date: 2024-11-19CHENGDU ZHONGQIAN AUTOMATION ENG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411228157.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-19
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

Mobile edge computing resources are limited, making it difficult to give satisfactory answers in service experience, and how to design a reasonable offload strategy to achieve efficient utilization of resources and reduce system costs has become a challenge.

Method used

Using the MEC task unloading method based on the non-dominant sorting artificial hummingbird algorithm, the unloading strategy is mapped to the hummingbird location by building a multi-objective optimization model, and combining the Pareto multi-objective optimization framework and the crowding distance improvement algorithm, the optimal unloading strategy is optimized and solved. To improve the work efficiency of smart mobile devices and reduce costs.

Benefits of technology

Effectively find a task offload solution that weighs latency and energy consumption, improve user service experience, reduce system costs, and obtain the optimal offload strategy solution in a short time, demonstrating high convergence and distribution performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118741606B_ABST
    Figure CN118741606B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for offloading MEC tasks based on a non-dominated sorting artificial hummingbird algorithm. Firstly, a multi-user network scenario architecture under a MEC system is constructed, and the basic data composition of the core elements of the system is expressed. A multi-objective optimization model is constructed according to the indicators to be optimized. Then, the model optimization problem is mapped to the non-dominated sorting artificial hummingbird algorithm to achieve a trade-off solution for multiple indicators, and obtain the optimal Pareto solution set and the optimal offloading strategy. Finally, the MEC system completes the calculation offloading deployment of the task according to the offloading strategy to improve the user's service experience.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent mobile device management, and in particular relates to a MEC task unloading method based on a non-dominated sorting artificial hummingbird algorithm. Background Art

[0002] Now is the golden age when mobile Internet, smart Internet of Things and 5G communication technology are advancing in parallel and developing rapidly. A variety of new generation smart mobile devices (Smart Mobile Device, SMD) are constantly emerging around us, such as smart watches, smart stylus, AR glasses, etc. At the same time, new network application scenarios such as smart medical care, autonomous driving, smart home, etc. are also emerging in an endless stream.

[0003] What follows are countless storage tasks and computing needs. At the same time, users' requirements for mobile device service experience and quality are gradually increasing. Faced with massive new applications with requirements such as high performance, low latency and low energy consumption, SMD's performance is gradually unable to cope. Although the emergence of mobile cloud computing has alleviated its computing pressure to a certain extent, it is still difficult to give a satisfactory answer in terms of service experience.

[0004] Mobile Edge Computing (MEC) is expected to provide users with ultra-low latency, high bandwidth and higher security network services due to its proximity to edge devices. However, MEC resources are relatively limited. How to design a reasonable offloading strategy to achieve efficient use of resources, improve service quality and reduce system costs has become a hot topic widely discussed by scholars and researchers inside and outside the industry. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a MEC task offloading method based on a non-dominated sorting artificial hummingbird algorithm. For the network scenario of multi-users and a single server under mobile edge computing, while considering performance indicators such as latency and energy consumption, the method solves the offloading strategy optimization problem with the goal of minimizing system indicators and balancing user indicator preferences, thereby improving the working efficiency of smart mobile devices and reducing costs.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a method for MEC task offloading based on a non-dominated sorting artificial hummingbird algorithm, comprising the steps of:

[0007] S1. Network scenario construction: Build a multi-user single-server task offloading network architecture under the mobile edge computing system, and represent basic data including mobile devices, MEC servers and computing tasks according to scenario information;

[0008] S2. Construction of multi-objective optimization model: Combined with MEC scenarios and based on the indicators to be optimized and the offloading strategy, a multi-objective optimization model for task offloading calculation is constructed;

[0009] S3. Model-algorithm mapping: The various strategies of the task offloading problem are mapped to the positions of the hummingbirds in the non-dominated sorting artificial hummingbird algorithm, and the multi-objective indicators are mapped to the fitness evaluation function of the hummingbirds searching for food sources. The mapping encoding adopts a multi-strategy dynamic joint encoding method dominated by an offloading strategy;

[0010] S4. Algorithm optimization solution: Combine the single-objective artificial hummingbird algorithm with the Pareto multi-objective optimization framework, introduce the solution update strategy of the crowding distance improvement algorithm, use the dynamic elimination principle to update the algorithm archive, and use the algorithm under the above strategy to solve the optimal Pareto frontier solution set and optimal unloading strategy of the optimization model;

[0011] S5. Optimal strategy scenario application: The optimal offloading strategy obtained by the non-dominated sorting artificial hummingbird algorithm for the system model is used to complete the offloading deployment of mobile edge system tasks.

[0012] Furthermore, the multi-objective optimization model in step S2 is constructed according to the index to be optimized g and the unloading strategy as follows:

[0013] ;

[0014] Among them, N is the number of tasks, K is the number of types of indicators g, They represent the offloading strategy, device transmit power allocation strategy, MEC server resource allocation strategy, and local device resource allocation strategy respectively. is the strategy expression of the Kth indicator of the task N corresponding to the indicator g, Corresponding to the model expression of the Kth indicator of indicator g, the constraint C1 indicates that the value of the strategy expression is 0 or 1.

[0015] Furthermore, in step S3, the multi-strategy dynamic joint encoding dominated by the offloading strategy is performed: each hummingbird individual represents a complete strategy solution, and the offloading strategy and resource allocation strategy are combined through a component The joint representation is used to realize the combination of multiple strategies; each value is a part of the complete strategy solution, X is the offloading strategy, which plays a dominant role in the solution update process, and other resource allocation strategies can participate in the algorithm's solution update process only after completing dynamic updates based on X.

[0016] Furthermore, in step S4, the single-objective artificial hummingbird algorithm is combined with the Pareto multi-objective optimization framework, including the steps of:

[0017] S41: Initialize the hummingbird individual population and obtain the initial solution;

[0018] S42: A new interpretation was obtained based on the behaviors and rules of hummingbirds, such as how they fly and forage;

[0019] S43: updating the solution according to the multi-objective optimization solution updating strategy;

[0020] S44: Construct a non-dominated solution set through non-dominated sorting;

[0021] S45: Adjust the size of the non-dominated solution set of the population based on the archive size;

[0022] S46: Determine whether the iteration is finished according to the iteration cutoff condition. If not, re-enter step S42 until the iteration is finished and the optimal Pareto solution set is obtained.

[0023] Furthermore, the optimal Pareto front solution set and the optimal unloading strategy of the optimization model are solved, including the steps of:

[0024] D1: Algorithm parameter setting: set the population size, maximum number of iterations and number of users;

[0025] D2: Population initialization: Randomly generate the initial population position, and calculate the optimal position, optimal fitness value and optimal non-dominated solution set of the population based on the initial position, and initialize the hummingbird's access table;

[0026] D3: Direction vector update: randomly select one of the three flight skills: omnidirectional flight, axial flight and diagonal flight to update the flight direction vector table;

[0027] D4: Foraging method selection and position update: Randomly select one of the foraging methods of guided foraging or territory foraging to explore a new location, select a better location according to the improved solution update strategy, and complete the update of the visit table, fitness value and elite solution set;

[0028] D5: Update of the worst population position: The individuals in the worst position of the population migrate for food and complete the update of the position, visit table, fitness value and elite solution set;

[0029] D6: Archive update: Update the optimal non-dominated solution set archive based on the crowding distance dynamic elimination method;

[0030] D7: Determine the end of iteration: Determine whether the iteration is over according to the iteration cutoff condition. If not, re-enter step D3 until the iteration is over and the optimal Pareto solution set and optimal unloading strategy corresponding to the optimal position of the hummingbird are obtained.

[0031] Furthermore, the direction vectors under the three flight skills in step D3 are updated as follows:

[0032] D31.Axial flight:

[0033] ;

[0034] Among them, r represents the hummingbird number, Dim represents the maximum dimension index value, and d represents a random dimension index value between dimensions 1 and Dim. A value of 1 means that hummingbird r can fly in the direction of the dimension.

[0035] D32. Diagonal flight:

[0036] ;

[0037] Among them, randperm is a random integer sequence generation function, u is a dimension index value randomly selected from dimension 1 to Dim, e is a dimension index value randomly selected from dimension 2 to Dim-1, and rand is a random number generation function; under this flying skill, the position in at least one dimension direction will be maintained unchanged, which is also the basis for the generation of diagonal vectors;

[0038] D33. Omnidirectional flight:

[0039] ;

[0040] Among them, z is any dimension index value between dimension 1 and Dim;

[0041] During omnidirectional flight, the corresponding positions of all dimensional directions in the solution space are allowed to be updated and changed.

[0042] Furthermore, in step D4, the expressions of the improved solution updating strategy before and after the improvement are respectively:

[0043] Solution update strategy before improvement:

[0044] 1) If the dominance level of the candidate solution is lower than that of the original solution, the candidate solution will replace the original solution;

[0045] 2) If the candidate solution has a higher dominance level than the original solution, the original solution is retained;

[0046] 3) If the two solutions have the same dominance level, randomly select one of them to update;

[0047] Improved solution updating strategy:

[0048] 1) If the dominance level of the candidate solution is lower than that of the original solution, the candidate solution will replace the original solution;

[0049] 2) If the candidate solution has a higher dominance level than the original solution, the original solution is retained;

[0050] 3) If the dominance levels of the two solutions are the same, the crowding distances dis of the two individual solutions are further compared, and the solution with the larger crowding distance is selected to participate in the new iteration.

[0051] Furthermore, the mathematical representation of the improved solution updating strategy in step D4 is:

[0052] ;

[0053] Where i is the individual solution index number, t represents the time, is a candidate solution, For the original solution, is a new solution. If r represents the dominance level, then are non-dominated sets that belong to the same domination level, and are the crowding distance between the candidate solution and the original solution, and the crowding distance between the individual solution It is defined as the sum of all sub-objective crowding distances of the solution. The sub-objective function crowding distance is the ratio of the difference between the distances of the two adjacent body solutions before and after the solution on the objective function to the difference between the maximum distance and the minimum distance on the objective function. The larger the crowding distance, the larger the exploration space near the solution.

[0054] Furthermore, the crowding distance calculation formula of the individual solution is:

[0055] ;

[0056] Where i is the individual solution index number, is the optimization model of index g, max{} is the maximum value function, and min{} is the minimum value function.

[0057] Furthermore, in step D6, the principle of the crowding distance dynamic elimination method is: archive dynamic elimination is performed according to the minimum crowding distance, that is, whenever an individual is eliminated, the crowding distance of its adjacent individuals is immediately adjusted, and the next elimination operation is performed based on the new crowding distance.

[0058] The beneficial effects of adopting this technical solution are:

[0059] The present invention proposes a task offloading method that simultaneously optimizes the offloading strategy and the resource allocation strategy based on the non-dominated sorting artificial hummingbird algorithm. The present invention expands the multi-objective optimization function of the Artificial Hummingbird Algorithm (AHA) and first applies it to the scenario of mobile edge computing task offloading. The obtained non-dominated sorting artificial hummingbird algorithm (NSAHA) can be well matched with the task offloading model. Compared with some classic multi-objective optimization algorithms and multi-objective optimization algorithms of new mainstream applications, its convergence, distribution and comprehensive performance are more outstanding, and it can effectively find a balanced task offloading solution that meets user needs.

[0060] In order to solve the problem of offloading tasks in mobile edge computing, the mapping strategy in the present invention has strong adaptability to the model. For complex models with multiple strategies, there may be a certain order relationship between the strategies. The multi-strategy joint coding method dominated by the offloading strategy in the present invention can avoid coding confusion in the solution update process by setting priorities, and at the same time can improve the efficiency of encoding and decoding, so that the algorithm can obtain the optimal offloading strategy solution in a short time;

[0061] Compared with the general multi-objective implementation method, the improved solution update strategy of the present invention can guide the solution to converge more reasonably and efficiently; for the archive elimination using the elite solution retention strategy algorithm, the general elimination method is to directly remove in batches based on the dominance level and crowding distance, without considering the impact of each elimination on the crowding degree of the adjacent solutions, while the dynamic elimination will recalculate the crowding distance of the adjacent individual solutions in each elimination, so as to maximize the diversity of solutions; BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flow chart of a method for MEC task offloading based on a non-dominated sorting artificial hummingbird algorithm of the present invention;

[0063] Figure 2 A network scene graph constructed for an embodiment of the present invention;

[0064] Figure 3 It is a graph of the convergence effect of the test set function according to an embodiment of the present invention, wherein a is a graph of the result of ZDT test set 1, b is a graph of the result of ZDT test set 2, c is a graph of the result of ZDT test set 3, and d is a graph of the result of ZDT test set 6;

[0065] Figure 4This is a comparison chart of indicator optimization of different unloading methods when the embodiment of the present invention is used for the unloading model, wherein a is the system average delay result chart, and b is the system average energy consumption result chart. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described below with reference to the accompanying drawings.

[0067] In this embodiment, see Figure 1 As shown, the present invention proposes a MEC task offloading method based on a non-dominated sorting artificial hummingbird algorithm.

[0068] The general implementation process is to first build a multi-user single-server network scenario, then build a system model for task offloading based on the scenario, express the objective function of the strategy problem to be optimized and provide the system parameters required for calculation, then map the model to the algorithm and complete the algorithm parameter setting, and finally iterate the algorithm to find the optimal strategy after optimization.

[0069] The embodiment specifically includes the steps of:

[0070] S1. Network scenario construction: Build a multi-user single-server task offloading network architecture under the mobile edge computing system, and represent basic data including mobile devices, MEC servers, and computing tasks according to the scenario information.

[0071] Instantiate a mobile edge computing scenario. This embodiment takes the mobile edge system in a multi-user active small-scale cell as an example. The constructed network scenario diagram is as follows: Figure 2 As shown, the details of the scene are described as follows:

[0072] Users who own mobile devices such as VR glasses, smart watches, and laptops in the cell are considered valid users. Assuming that the number of users is n, there is a task to be processed on the user's device during each task unloading time slot, and each task is an independent task and cannot be divided twice. The task set composed of all users can be recorded as N={1.2,...,n}. A network base station (BS) is deployed within the coverage area of ​​the cell. The BS is connected to the MEC server by wire. Mobile users can transmit the computing tasks on their mobile devices to the server corresponding to the BS through a wireless link for execution, and then return the calculation results; the resources of a single MEC server are limited and insufficient to provide offloading support for a large number of tasks. Therefore, in this embodiment, a cloud server is introduced to expand the computing resources of the MEC system. The task can complete the transmission of task data to the cloud through the backhaul link between the MEC server base station and the cloud core network. In order to maintain generality, it is assumed that all devices in the system can only transmit one task at the same time; in this embodiment, considering the two indicators of latency and energy consumption, the structure of any task can be expressed as ,in Indicates the amount of data for task n, represents the amount of computation (CPU cycles) required for computing task n. The data structure of any server can be expressed as , Represents the computing resources allocated by the server to the task, is the maximum computing resource that can be allocated to the server. is the maximum number of tasks that the server can support offloading, and the data structure of each mobile device can be expressed as , The device computing resources that can be allocated to task n, The device transmit power that can be allocated to task n is is the maximum transmission power of the device where task n is located, is the power consumption factor of the device chip architecture.

[0073] S2. Construction of multi-objective optimization model: Combined with MEC scenarios and based on the indicators to be optimized and the offloading strategy, a multi-objective optimization model for task offloading calculation is constructed.

[0074] The multi-objective optimization model in step S2 is constructed according to the index to be optimized g and the unloading strategy as follows:

[0075] ;

[0076] Among them, N is the number of tasks, K is the number of types of indicators g, They represent the offloading strategy, device transmit power allocation strategy, MEC server resource allocation strategy, and local device resource allocation strategy respectively. is the strategy expression of the k-th indicator corresponding to indicator i of task n, Corresponding to the model expression of the kth indicator of indicator g, the constraint C1 indicates that the value of the strategy expression is 0 or 1.

[0077] S3. Model-algorithm mapping: The various strategies of the task offloading problem are mapped to the positions of hummingbirds in the non-dominated sorting artificial hummingbird algorithm, and the multi-objective indicators are mapped to the fitness evaluation function of the hummingbird in searching for food sources. The mapping encoding adopts a multi-strategy dynamic joint encoding method dominated by an offloading strategy.

[0078] In step S3, the multi-strategy dynamic joint encoding dominated by the offloading strategy is as follows: each hummingbird individual represents a complete strategy solution, and the offloading strategy and resource allocation strategy are combined into a component The joint representation is used to realize the combination of multiple strategies; each value is a part of the complete strategy solution, X is the offloading strategy, which plays a dominant role in the solution update process, and other resource allocation strategies can participate in the algorithm's solution update process only after completing dynamic updates based on X.

[0079] If the model dimension is d, the multi-strategy joint dynamic encoding matrix is ​​expressed as:

[0080] .

[0081] S4. Algorithm optimization solution: The single-objective artificial hummingbird algorithm is combined with the Pareto multi-objective optimization framework, the solution update strategy of the crowding distance improvement algorithm is introduced, the dynamic elimination principle is used to update the algorithm archive, and the algorithm under the above strategy is used to solve the optimal Pareto frontier solution set and optimal unloading strategy of the optimization model.

[0082] In step S4, the single-objective artificial hummingbird algorithm is combined with the Pareto multi-objective optimization framework, including the steps of:

[0083] S41: Initialize the hummingbird individual population and obtain the initial solution;

[0084] S42: A new interpretation was obtained based on the behaviors and rules of hummingbirds, such as how they fly and forage;

[0085] S43: updating the solution according to the multi-objective optimization solution updating strategy;

[0086] S44: Construct a non-dominated solution set through non-dominated sorting;

[0087] S45: Adjust the size of the non-dominated solution set of the population based on the archive size;

[0088] S46: Determine whether the iteration is finished according to the iteration cutoff condition. If not, re-enter step S42 until the iteration is finished and the optimal Pareto solution set is obtained.

[0089] S5. Optimal strategy scenario application: According to the optimal offloading strategy obtained by the non-dominated sorting artificial hummingbird algorithm for the system model, the offloading deployment of mobile edge system tasks is completed, which improves the user service experience and saves system cost expenditure.

[0090] Solving the optimal Pareto front solution set and optimal unloading strategy of the optimization model includes the following steps:

[0091] D1: Algorithm parameter setting: Set the population size npop, the maximum number of iterations MaxIt and the user number parameter N.

[0092] D2: Population initialization: Randomly generate the initial population position PopPos, and calculate the optimal position BestPos, optimal fitness value PopFit, and optimal non-dominated solution set tPop of the population based on the initial position, and access table T.

[0093] Population position PopPos initialization method:

[0094] ;

[0095] Where i is the individual solution index number, rand is the random number generation function, Up and Low are the upper and lower boundaries of the search space respectively.

[0096] The access table T is used to record the maximum non-visit time of other hummingbird locations, reflecting the specific memory ability of hummingbirds during foraging. The access table T initialization method is:

[0097] ;

[0098] Where i is the individual solution index number, Indicates other location hummingbird numbers.

[0099] D3: Direction vector update: randomly select one of the three flight skills: omnidirectional flight, axial flight and diagonal flight to update the flight direction vector table;

[0100] The direction vectors under the three flight skills in step D3 are updated as follows:

[0101] D31.Axial flight:

[0102] ;

[0103] Among them, r represents the hummingbird number, Dim represents the maximum dimension index value, and d represents a random dimension index value between dimensions 1 and Dim. A value of 1 means that hummingbird r can fly in the direction of dimension d.

[0104] D32. Diagonal flight:

[0105] ;

[0106] Among them, randperm is a random integer sequence generation function, u is a dimension index value randomly selected from dimension 1 to Dim, e is a dimension index value randomly selected from dimension 2 to Dim-1, and rand is a random number generation function; under this flying skill, the position in at least one dimension direction will be maintained unchanged, which is also the basis for the generation of diagonal vectors;

[0107] D33. Omnidirectional flight:

[0108] ;

[0109] Among them, z is any dimension index value between dimension 1 and Dim;

[0110] During omnidirectional flight, the corresponding positions of all dimensional directions in the solution space are allowed to be updated and changed.

[0111] D4: Foraging method selection and position update: Randomly select one of the foraging methods of guided foraging or territory foraging to explore a new location, select a better location according to the improved solution update strategy, and complete the update of the visit table, fitness value and elite solution set;

[0112] 1) Guided foraging:

[0113] ;

[0114] Among them, t represents the time, a is the step factor, The location of the target food source for guidance, is the current location of the hummingbird.

[0115] 2) Territory foraging:

[0116] ;

[0117] Where b is the step size factor.

[0118] The formula for updating the position of a hummingbird foraging is as follows:

[0119] ;

[0120] After the location is updated, Hummingbird updates the access table as follows:

[0121] ;

[0122] Under guided foraging, the hummingbird will remember the visit to the target location, so the visit table will also add the following updates:

[0123] ;

[0124] in is the target position index number, i is the individual solution index number, j represents the hummingbird number at other positions, and Fitness() is the fitness value calculation function.

[0125] In step D4, the expressions of the improved solution update strategy before and after improvement are:

[0126] Solution update strategy before improvement:

[0127] 1) If the dominance level of the candidate solution is lower than that of the original solution, the candidate solution will replace the original solution;

[0128] 2) If the candidate solution has a higher dominance level than the original solution, the original solution is retained;

[0129] 3) If the two solutions have the same dominance level, randomly select one of them to update;

[0130] Improved solution updating strategy:

[0131] 1) If the dominance level of the candidate solution is lower than that of the original solution, the candidate solution will replace the original solution;

[0132] 2) If the candidate solution has a higher dominance level than the original solution, the original solution is retained;

[0133] 3) If the dominance levels of the two solutions are the same, the crowding distances dis of the two individual solutions are further compared, and the solution with the larger crowding distance is selected to participate in the new iteration.

[0134] Mathematical representation of the improved solution updating strategy in step D4:

[0135] ;

[0136] Where i is the individual solution index number, t represents the time, is a candidate solution, For the original solution, is a new solution. If r represents the dominance level, then are non-dominated sets that belong to the same domination level, and are the crowding distances between the candidate solution and the original solution The crowding distance of an individual solution is defined as the sum of the crowding distances of all sub-objectives of the solution. The crowding distance of the sub-objective function is the ratio of the difference between the distances of the two adjacent individual solutions before and after the solution on the objective function to the difference between the maximum distance and the minimum distance on the objective function. The larger the crowding distance, the larger the exploration space near the solution.

[0137] The crowding distance calculation formula of the individual solution is:

[0138] ;

[0139] Where i is the individual solution index number, and They are the optimization models for delay and energy consumption respectively, max{} is the maximum value function, and min{} is the minimum value function.

[0140] D5: Update of the worst population position: The individuals in the worst position of the population migrate for food and complete the update of the position, visit table, fitness value and elite solution set.

[0141] D6: Archive update: Update the optimal non-dominated solution set archive based on the crowding distance dynamic elimination method;

[0142] In step D6, the principle of the crowding distance dynamic elimination method is: archive dynamic elimination is performed according to the minimum crowding distance, that is, whenever an individual is eliminated, the crowding distance of its adjacent individuals is immediately adjusted, and the next elimination operation is performed based on the new crowding distance.

[0143] D7: Iteration end judgment: Determine whether the iteration is over according to the iteration cutoff condition. If not, re-enter step D3 until the iteration is over and the optimal Pareto solution set and optimal unloading strategy corresponding to the hummingbird's optimal position are obtained.

[0144] In order to verify the performance of the mobile edge computing task offloading method based on the non-dominated sorting artificial hummingbird algorithm (NSAHA) proposed in the present invention, this embodiment designs and completes relevant control experiments. The algorithms compared with the NSAHA proposed in the present invention mainly include: three classic multi-objective optimization algorithms: multi-objective evolutionary algorithm based on decomposition (MOEA / D), non-dominated sorting genetic algorithm II (NSGA-II) and multi-objective particle swarm optimization algorithm (MOPSO), two new and widely used multi-objective optimization algorithms: multi-objective grey wolf optimization algorithm (MOGWO) and non-dominated sorting whale optimization algorithm (NSWOA); The control experiment uses the most widely used ZDT test set, and selects four test functions ZDT1, ZDT2, ZDT3, and ZDT6 to evaluate the algorithm performance. The experimental parameters set the population size to 100, the archive size to 100, the maximum number of evaluations to 10,000, and each test algorithm instance is repeated 50 times to take the average value.

[0145] Table 1 shows the comparison of five evaluation indicators of NSAHA and the other five multi-objective algorithms for the ZDT test set function. Among the five indicators, IGD, GD and HV are convergence indicators. IGD and GD measure the closeness between the generated solution and the true frontier solution from different angles. The lower the value of the two, the better the convergence of the algorithm. IGD can also reflect the diversity of solutions to a certain extent. HV measures the volume of the generated solution set that can cover the true frontier solution set. The higher its value, the better the overall performance of the algorithm. Spread and Spacing are distribution indicators. Spread measures the discreteness of the Pareto frontier solution. The smaller its value, the better the balance between multiple objectives can be maintained. Spacing measures the standard deviation of the minimum distance between each generated solution and other solutions. The lower its value, the better the uniformity of the solution. From the experimental data in Table 1, we can see that NSAHA has better results in 50% of the indicators compared with the other five algorithms in the ZDT test set function, and its performance in solution set distribution and algorithm comprehensive performance is particularly outstanding, indicating the effectiveness and superiority of the non-dominated sorting artificial hummingbird algorithm NSAHA.

[0146] Table 1 shows the comparison of evaluation indicators of each algorithm

[0147]

[0148] Figure 3The optimization results of NSAHA for the above four test functions are given. The dots are the real Pareto front solutions of the test functions, and the diamonds are the actual Pareto front solutions obtained after 100 iterations of the algorithm proposed in the present invention. It can be seen that the two types of curves of the four functions basically overlap and the surfaces are relatively smooth. It can be obtained that the non-dominated sorting artificial hummingbird algorithm NSAHA has a good solution effect for the four test functions and can effectively obtain the front solution set near the optimal Pareto solution.

[0149] Figure 4 The indicators of different offloading strategies as the number of tasks changes are given. The number of tasks is adjusted from 50 to 250, with every 50 task segments as an incremental level. Six offloading strategies are set for comparative experiments, among which the Local+MEC+Cloud strategy is the end-edge-cloud joint offloading strategy based on the non-dominated sorting artificial hummingbird algorithm NSAHA proposed in this embodiment, All-Local is the all local computing strategy, All-Cloud is the all upload cloud computing strategy, Local+MEC is the joint offloading strategy based on local and edge servers, Local+Cloud is the joint offloading strategy based on local and cloud servers, and Rand is the random offloading strategy. As the number of tasks increases from a small number, the Local+MEC+Cloud strategy based on NSAHA proposed in this embodiment can reduce the system latency by 47.91% to 83.80% compared with the All-Local strategy, and reduce the system energy consumption by 81.95% to 98.20%. When the energy consumption differs from the All-Cloud strategy by only 61.28% at most, the latency can be further reduced by 56.05% to 82.51%. Compared with the Local+MEC strategy, the latency can be reduced by an average of 58.52% when the task volume is low, and by 64.49% when the tasks are intensive. In general, the Local+MEC+Cloud strategy optimization method based on the non-dominated sorting artificial hummingbird algorithm NSAHA proposed in this embodiment can effectively reduce the system latency and energy consumption levels under the condition of balancing the two indicators.

[0150] Experiments show that compared with some classic multi-objective optimization algorithms, such as the multi-objective evolutionary algorithm based on decomposition, the non-dominated sorting genetic algorithm II, the multi-objective particle swarm optimization algorithm, and some new and widely used multi-objective optimization algorithms, such as the multi-objective grey wolf optimization algorithm and the non-dominated sorting whale optimization algorithm, the algorithm proposed in the present invention has great advantages in terms of good convergence, distribution and comprehensive performance when solving the multi-objective unloading model problem.

[0151] The experiments in the examples show that for the network architecture of end-edge-cloud three-layer task offloading, compared with the local computing strategy, all cloud server computing strategy, local-edge server computing joint offloading strategy, local-cloud server computing joint offloading strategy, random offloading strategy and other unloading methods, the computing offloading method of end-edge-cloud joint offloading of the unloading model proposed in the present invention can significantly optimize the target indicators and effectively reduce the indicator cost.

[0152] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A MEC task offloading method based on non-dominated sorting artificial hummingbird algorithm, characterized in that: Includes steps: S1. Network scenario construction: Build a multi-user single-server task offloading network architecture under the mobile edge computing system, and represent basic data including mobile devices, MEC servers and computing tasks according to scenario information; S2. Construction of multi-objective optimization model: Combined with MEC scenarios and based on the indicators to be optimized and the offloading strategy, a multi-objective optimization model for task offloading calculation is constructed; S3. Model-algorithm mapping: The various strategies of the task offloading problem are mapped to the positions of the hummingbirds in the non-dominated sorting artificial hummingbird algorithm, and the multi-objective indicators are mapped to the fitness evaluation function of the hummingbirds searching for food sources. The mapping encoding adopts a multi-strategy dynamic joint encoding method dominated by an offloading strategy; S4. Algorithm optimization solution: Combine the single-objective artificial hummingbird algorithm with the Pareto multi-objective optimization framework, introduce the solution update strategy of the crowding distance improvement algorithm, use the dynamic elimination principle to update the algorithm archive, and use the algorithm under the above strategy to solve the optimal Pareto frontier solution set and optimal unloading strategy of the optimization model; In step S4, the single-objective artificial hummingbird algorithm is combined with the Pareto multi-objective optimization framework, including the steps of: S41: Initialize the hummingbird individual population and obtain the initial solution; S42: A new interpretation of the flying and foraging behaviors and rules of hummingbirds was obtained; S43: updating the solution according to the multi-objective optimization solution updating strategy; S44: Construct a non-dominated solution set through non-dominated sorting; S45: Adjust the size of the non-dominated solution set of the population based on the archive size; S46: Determine whether the iteration is finished according to the iteration cutoff condition. If it is not finished, re-enter step S42 until the iteration is finished and the optimal Pareto solution set is obtained; The crowding distance calculation formula of the individual solution is: ; Where i is the individual solution index number, and They are the optimization models for the two indicators of delay and energy consumption, max{} is the maximum value function, and min{} is the minimum value function; S5. Optimal strategy scenario application: According to the optimal offloading strategy obtained by the non-dominated sorting artificial hummingbird algorithm for the system model, the offloading deployment of the mobile edge system tasks is completed; Solving the optimal Pareto front solution set and optimal unloading strategy of the optimization model includes the following steps: D1: Algorithm parameter setting: set the population size npop, the maximum number of iterations MaxIt and the user number parameter N; D2: Population initialization: randomly generate the initial population position PopPos, and calculate the optimal position BestPos, optimal fitness value PopFit, optimal non-dominated solution set tPop of the population based on the initial position, and access table T; Population position PopPos initialization method: ; Where i is the individual solution index number, rand is the random number generation function, Up and Low are the upper and lower boundaries of the search space respectively; The access table T is used to record the maximum non-visit time of other hummingbird locations, reflecting the specific memory ability of hummingbirds during foraging. The access table T initialization method is: ; Among them, i is the individual solution index number, and j represents the hummingbird number at other locations; D3: Direction vector update: randomly select one of the three flight skills: omnidirectional flight, axial flight and diagonal flight to update the flight direction vector table; D4: Foraging method selection and position update: Randomly select one of the foraging methods of guided foraging or territory foraging to explore a new location, select a better location according to the improved solution update strategy, and complete the update of the visit table, fitness value and elite solution set; Mathematical representation of the improved solution updating strategy in step D4: ; Where i is the individual solution index number, t represents the time, is a candidate solution, For the original solution, is a new solution. If r represents the dominance level, then are non-dominated sets that belong to the same domination level, and are the crowding distances between the candidate solution and the original solution The crowding distance of individual solution is defined as the sum of crowding distances of all sub-objectives of the solution. The crowding distance of sub-objective function is the ratio of the difference between the distances of two adjacent individual solutions before and after the solution on the objective function to the difference between the maximum distance and the minimum distance on the objective function. The larger the crowding distance, the larger the exploration space near the solution. D5: Update of the worst population position: The individuals in the worst position of the population migrate for food and complete the update of the position, visit table, fitness value and elite solution set; D6: Archive update: Update the optimal non-dominated solution set archive based on the crowding distance dynamic elimination method; D7: Iteration end judgment: Determine whether the iteration is over according to the iteration cutoff condition. If not, re-enter step D3 until the iteration is over and the optimal Pareto solution set and optimal unloading strategy corresponding to the hummingbird's optimal position are obtained.

2. According to claim 1, a MEC task offloading method based on non-dominated sorting artificial hummingbird algorithm is characterized in that: The multi-objective optimization model in step S2 is constructed according to the index to be optimized g and the unloading strategy as follows: ; Among them, N is the number of tasks, K is the number of types of indicators g, They represent the offloading strategy, device transmit power allocation strategy, MEC server resource allocation strategy, and local device resource allocation strategy respectively. is the strategy expression of the kth indicator of the indicator g corresponding to task n, Corresponding to the model expression of the k-th indicator of indicator g, the restriction condition indicates that the value of the strategy expression is 0 or 1.

3. According to claim 1, a MEC task offloading method based on non-dominated sorting artificial hummingbird algorithm is characterized in that: In step S3, the multi-strategy dynamic joint encoding dominated by the offloading strategy is as follows: each hummingbird individual represents a complete strategy solution, and the offloading strategy and resource allocation strategy are combined into a component The joint representation is used to realize the combination of multiple strategies; each value is a part of the complete strategy solution, X is the offloading strategy, which plays a dominant role in the solution update process, and other resource allocation strategies can participate in the algorithm's solution update process only after completing dynamic updates based on X.

4. According to claim 1, a MEC task offloading method based on non-dominated sorting artificial hummingbird algorithm is characterized in that: The direction vectors under the three flight skills in step D3 are updated as follows: D31.Axial flight: ; Among them, r represents the hummingbird number, Dim represents the maximum dimension index value, and d represents a random dimension index value between dimensions 1 and Dim. A value of 1 means that hummingbird r flies in the direction of dimension d. D32. Diagonal flight: ; Among them, randperm is a random integer sequence generation function, u is a dimension index value randomly selected from dimension 1 to Dim, e is a dimension index value randomly selected from dimension 2 to Dim-1, and rand is a random number generation function; under this flying skill, the position of at least one dimension direction will be maintained unchanged, which is also the basis for the generation of diagonal vectors; D33. Omnidirectional flight: ; Among them, z is any dimension index value between dimension 1 and Dim; During omnidirectional flight, the corresponding positions of all dimensional directions in the solution space are allowed to be updated and changed.

5. According to claim 1, a MEC task offloading method based on non-dominated sorting artificial hummingbird algorithm is characterized in that: In step D4, the expressions of the improved solution update strategy before and after improvement are: Solution update strategy before improvement: 1) If the dominance level of the candidate solution is lower than that of the original solution, the candidate solution will replace the original solution; 2) If the candidate solution has a higher dominance level than the original solution, the original solution is retained; 3) If the two solutions have the same dominance level, randomly select one of them to update; Improved solution updating strategy: 1) If the dominance level of the candidate solution is lower than that of the original solution, the candidate solution will replace the original solution; 2) If the candidate solution has a higher dominance level than the original solution, the original solution is retained; 3) If the dominance levels of the two solutions are the same, the crowding distances dis of the two individual solutions are further compared, and the solution with the larger crowding distance is selected to participate in the new iteration.

6. According to claim 1, a method for MEC task offloading based on non-dominated sorting artificial hummingbird algorithm is characterized in that: In step D6, the principle of the crowding distance dynamic elimination method is: archive dynamic elimination is performed according to the minimum crowding distance, that is, whenever an individual is eliminated, the crowding distance of its adjacent individuals is immediately adjusted, and the next elimination operation is performed based on the new crowding distance.