Heterogeneous cloud computing multi-target task scheduling method based on hybrid algorithm

Abstract

The invention relates to a heterogeneous cloud computing multi-target task scheduling method based on a hybrid algorithm, and belongs to the technical field of target task scheduling. Comprising the steps of realizing mapping between a virtual machine and a task based on an improved grey wolf optimization algorithm and a local search algorithm, and finally realizing multi-target task scheduling optimization in a heterogeneous cloud environment. According to the hybrid algorithm, global search and local search are effectively balanced, and the performance of a traditional algorithm is remarkably improved. The algorithm is used as a scheduling method to be applied to cloud computing task scheduling, the optimal mapping scheme of the task and the virtual machine can be obtained, and resource allocation and task execution are remarkably optimized.

Description

Technical Field

[0001] The present invention relates to a heterogeneous cloud computing multi-objective task scheduling method based on a hybrid algorithm, and belongs to the technical field of target task scheduling. Background Art

[0002] Cloud computing is a revolutionary technology that provides data storage, processing power, and a variety of applications over the internet. This technology allows users to quickly access resources as needed without having to invest in and maintain their own physical hardware. Cloud computing offers a variety of services, including Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS), bringing unprecedented convenience to individuals and businesses. Cloud systems consist of two main entities: consumers and cloud service providers. Consumers seek low latency and high reliability, while cloud service providers need to maximize resource utilization, reduce operating costs, and maintain service quality. For example, in the financial services industry, cloud computing platforms handle highly concurrent transaction processing and data analysis tasks, requiring low latency for critical transactions while optimizing resource utilization and reducing operating costs. In the streaming media industry, consumers have extremely high demands for low latency, ensuring smooth and uninterrupted viewing while also ensuring that providers' profits are not affected. With the surge in workloads and the diversification of resource requirements, heterogeneous cloud computing has become an increasingly important computing model. Comprising a large number of heterogeneous computing nodes, this environment can effectively handle diverse workloads. However, this heterogeneity brings great challenges to the resource management and scheduling of cloud computing, such as unbalanced resource allocation, low task execution efficiency and complexity of energy consumption management.

[0003] In cloud environments, task scheduling is a critical component of resource management. Effective task scheduling strategies not only ensure service quality but also improve user satisfaction and task execution efficiency. However, cloud task scheduling is an NP-hard problem. Furthermore, the performance differences among computing nodes complicate task scheduling. To optimize task scheduling, researchers have proposed various task scheduling strategies and algorithms. Previous approaches to solving task scheduling problems can be broadly categorized into policy-based scheduling algorithms, heuristic algorithms, metaheuristic algorithms, and hybrid algorithms.

[0004] Regarding policy-based scheduling algorithms, Pol et al. (see: Pol, SS, & Singh, A. (2021, May). Task scheduling algorithms in cloud computing: a survey. In 2021 2nd International Conference on Secure Cyber ​​Computing and Communications (ICSCCC) (pp. 244-249). IEEE.) mentioned in the article that the existing task scheduling algorithm in cloud computing is the First Come First Served (FCFS) algorithm. This algorithm is a non-preemptive algorithm, and its application in task scheduling will result in excessive waiting time. Rassan et al. (see: Rassan, IA, & Alarif, N. (2021). Load Balancing Approach to Enhance the Performance in Cloud Computing. International journal of computer science and network security: IJCSNS, 21 (2), 158-170.) described in the article that the use of the round-robin scheduling algorithm (RR) in task scheduling may result in some nodes being heavily loaded and some nodes being idle or lightly loaded. Murad et al. (see Murad, SA, Azmi, ZRM, Muzahid, AJM, Sarker, MMH, Miah, MSU, Bhuiyan, MKB, & Bairagi, AK (2024). Priority-based job scheduling technique that utilizes gaps to increase the efficiency of job distribution in cloud computing. Sustainable Computing: Informatics and Systems, 41, 100942.) note that Longest Job First (LJF) scheduling can lead to long times, high latency, and low utilization. In addition to the aforementioned shortcomings, FCFS, RR, SJF, and LJF often fail to effectively adapt to dynamically changing resource requirements and task characteristics.

[0005] Regarding heuristic algorithms, Younes et al. (see: Younes, A., Elnahary, MK, Alkinani, MH, & El-Sayed, HH (2022). Task Scheduling Optimization in Cloud Computing by Rao Algorithm. Computers, Materials & Continua, 72(3).) use the RAO algorithm to minimize task execution time. Kong et al. (see: Kong, L., Mapetu, JPB, & Chen, Z. (2020). Heuristic load balancing based zero imbalance mechanism in cloud computing. Journal of Grid Computing, 18(1), 123-148.) use the zero imbalance method to optimize load balancing between virtual machines. Alhaidari et al. (see: Alhaidari, F., & Balharith, TZ (2021). Enhanced round-robin algorithm in the cloud computing environment for optimal task scheduling. Computers, 10 (5), 63.) proposed a dynamic round-robin heuristic algorithm (DRRHA), which has obvious competitive advantages in average waiting time, turnaround time, and response time. Mahmoud et al. (see: Mahmoud, H., Thabet, M., Khafagy, MH, & Omara, FA (2021). An efficient load balancing technique for task scheduling in heterogeneous cloud environment. Cluster Computing, 24 (4), 3405-3419.) pointed out that there is no research on using the heterogeneous earliest finish algorithm (HEFT) to optimize the load balancing problem in a heterogeneous environment. To solve this problem, they improved HEFT and proposed the load-balanced heterogeneous earliest finish algorithm (LB-HEFT) to achieve better load balancing effect.

[0006] Regarding metaheuristic algorithms, Mondal et al. (see: Mondal, B., & Choudhury, A. (2024). Multi-objective cuckoo optimizer for task scheduling to balance workload in cloud computing. Computing, 1-32.) proposed using a cuckoo search algorithm for scheduling, effectively distributing the load among available virtual machines while maintaining a low total response time and average task processing time. Zhang et al. (see: Zhang, H., & Jia, R. (2023). Application of chaotic cat swarm optimization in cloud computing multi-objective task scheduling. IEEE Access, 11, 95443-95454.) introduced a cat swarm optimization model and improved it by adjusting weight coefficients and fitness, shortening task execution time and reducing costs. Hamed et al. (see: Hamed, AY, Elnahary, MK, Alsubaei, FS, & El-Sayed, HH (2023). Optimization Task Scheduling Using Cooperation Search Algorithm for Heterogeneous Cloud Computing Systems. Computers, Materials & Continua, 74 (1).) used a cooperative search algorithm (CSA) in heterogeneous cloud computing systems to shorten the maximum completion time. Pan et al. (see: Pan, JS, Yu, N., Chu, SC, Zhang, AN, Yan, B., & Watada, J. (2025). Innovative Approaches to Task Scheduling in Cloud Computing Environments Using an Advanced Willow Catkin Optimization Algorithm. Computers, Materials & Continua, 82 (2).) proposed an improved willow catkin optimization algorithm (AWCO), which enhances the global search capability of the algorithm through a class position-based learning strategy and speeds up its conversion speed through dotted line mapping, thereby improving the performance of the algorithm.Gong et al. (see: Gong, R., Li, D., Hong, L., & Xie, N. (2024). Task scheduling in cloud computing environment based on enhanced marine predator algorithm. Cluster Computing, 27(1), 1109-1123.) proposed the Enhanced Marine Predator Algorithm (EMPA), which reduces the time span and improves resource utilization.

[0007] Regarding hybrid algorithms, Behera et al. (see: Behera, I., & Sobhanayak, S. (2024). HTSA: A novel hybrid task scheduling algorithm for heterogeneous cloud computing environment. Simulation Modelling Practice and Theory, 137, 103014.) proposed a method called HTSA that combines genetic algorithm (GA) and gravitational search algorithm (GSA) to improve system performance and optimize energy consumption, time span, resource utilization, and throughput. Alsubai et al. (see: Alsubai, S., Garg, H., & Alqahtani, A. (2023). A novel hybrid MSA-CSA algorithm for cloud computing task scheduling problems. Symmetry, 15 (10), 1931.) proposed a method that combines moth swarm algorithm (MSA) and chameleon optimization algorithm (CSA) to apply to the task scheduling process, which has advantages in terms of resource utilization, cost, and minimum completion time. Qiuju et al. (see: Qiuju, DENG, Ning, WANG, & Yang, LU (2023). Cloud task scheduling using the squirrel search algorithm and improved genetic algorithm. International Journal of Advanced Computer Science and Applications, 14 (3).) proposed a new hybrid task scheduling algorithm based on the squirrel search algorithm and improved genetic algorithm, which improved the performance of execution time, time span and energy consumption.Huiying et al. (see: Huiying, SHAO (2025). A Novel Hybrid Algorithm Based on Butterfly and FlowerPollination Algorithms for Scheduling Independent Tasks on CloudComputing. International Journal of Advanced Computer Science & Applications, 16 (1).) proposed combining the flower pollination algorithm and the butterfly optimization algorithm to optimize cloud task scheduling and improve resource utilization.

[0008] Policy-based scheduling methods have difficulty adapting to complex load requirements. Heuristic algorithms are prone to falling into local optimality, have poor adaptability, and low computational efficiency. They lack versatility and have difficulty dealing with different types of problems. Metaheuristic algorithms enhance global search capabilities, improve adaptability and computational efficiency, but still have disadvantages such as slow convergence. Hybrid scheduling algorithms combine the advantages of multiple algorithms to improve the quality of solutions and optimize the search process. Although some hybrid scheduling algorithms have been developed, the most advanced hybrid optimization algorithms still have room for improvement in terms of optimization capabilities and convergence performance, and most of them focus on task scheduling in homogeneous computing environments. Therefore, there are still many new hybrid algorithms worth exploring in dealing with scheduling problems in specific heterogeneous cloud computing and meeting diverse needs. Summary of the Invention

[0009] In view of the shortcomings of the existing technology, the present invention provides a heterogeneous cloud computing multi-objective task scheduling method based on a hybrid algorithm;

[0010] The present invention proposes an improved gray wolf optimization algorithm, and proposes a hybrid algorithm based on the improved gray wolf optimization algorithm and the local search algorithm for multi-objective task scheduling optimization in a heterogeneous cloud environment. The hybrid algorithm effectively balances global search and local search, significantly improving the performance of traditional algorithms. Applying the algorithm as a scheduling method in cloud computing task scheduling can obtain the optimal mapping scheme between tasks and virtual machines, significantly optimizing resource allocation and task execution. The present invention uses HPC2N and NASA iPSC real data sets to test and compare the performance of the hybrid algorithm proposed in the present invention with three algorithms: PGSAO, GA-GWO and Whale Optimization Algorithm (WOA). Compared with the baseline algorithm, the hybrid algorithm proposed in the present invention significantly shortens the maximum completion time, reduces load imbalance and improves resource utilization.

[0011] The technical solution of the present invention is:

[0012] A hybrid algorithm-based multi-objective task scheduling method for heterogeneous cloud computing includes:

[0013] Based on the improved grey wolf optimization algorithm and local search algorithm, the mapping between virtual machines and tasks is realized, and finally multi-objective task scheduling optimization in heterogeneous cloud environments is achieved.

[0014] Preferably, according to the present invention, the heterogeneous cloud computing multi-objective task scheduling method is run on a task scheduling system;

[0015] The task scheduling system includes multiple users, a series of tasks, a task scheduler, a resource manager, and a data center;

[0016] Multiple users submit a series of tasks T = {T1, T2, T3, ..., T n}, assuming that a data center has p physical machines PM={PM1,PM2,PM3,...,PM p} and m virtual machines VM={VM1,VM2,VM3,...,VM m}, virtual machines reflect the heterogeneity of resources through different processing capabilities; the resource manager is responsible for obtaining information about these physical machines and virtual machines, and sending this physical machine information and virtual machine information to the task scheduler; physical machine information includes the number of CPU cores, architecture, memory size, network bandwidth and storage capacity; virtual machine information includes the number of CPUs, processor speed, memory size, network bandwidth and storage capacity; the task scheduler is responsible for monitoring the status of tasks and receiving information about available resources. The task scheduler maps tasks to appropriate resources based on the task scheduling algorithm.

[0017] According to the preferred embodiment of the present invention, the multi-objective task scheduling optimization simultaneously optimizes three objectives: reducing the overall completion time of the task, improving the balance of the system load, and improving resource utilization; the objective function ObjectiveFunction of the multi-objective task scheduling optimization is:

[0018] ObjectiveFunction=Min(α*MS-β*RUR+γ * LB);

[0019] Among them, MS refers to completion time, RUR refers to resource utilization, LB refers to load balancing; α, β, and γ are weight factors of different optimization objectives.

[0020] Further preferably, the calculation formula of the completion time MS is as follows:

[0021]

[0022] Among them, CT ijrepresents the execution time of task i on virtual machine j, S j is the ID set of tasks assigned to virtual machine j; CT ij As shown below:

[0023]

[0024] Among them, TL i Indicates the number of instructions per million for the i-th task, VMP j represents the processing speed of the jth virtual machine;

[0025] The calculation formula for resource utilization RUR is as follows:

[0026]

[0027] Among them, CTL j is the time it takes for virtual machine j to complete task execution, and vmNum represents the total number of virtual machines;

[0028] The calculation formula for load balancing LB is as follows:

[0029]

[0030] Among them, L j represents the load of the jth virtual machine, that is, the execution time of virtual machine j; Indicates the average load of all virtual machines, that is, the average execution time of the virtual machines.

[0031] Preferably, according to the present invention, a hybrid algorithm HIGWOLS includes an improved grey wolf optimization algorithm Improved-GWO and a local search algorithm LS to achieve efficient mapping between virtual machines and tasks; including:

[0032] First, initialize a gray wolf population. Each individual in the gray wolf population represents a mapping scheme between virtual machines and tasks. Assume that GW represents a gray wolf population and is defined as follows:

[0033] GW={GW1,GW2,...,GW i ,...,GW N};

[0034] Where N is the size of the population, GW i represents the i-th individual, which is expressed as:

[0035] GW i ={p1,p2,...,p j ,...,p v};

[0036] p jis the position value of the gray wolf individual in the j-th dimension;

[0037] Initialize the population individual p j Calculate as follows:

[0038] p j = (upper - lower) * random + lower;

[0039] where upper is the upper bound of the search area, lower is the lower bound of the search area; random is a random number between [0, 1];

[0040] At the same time, initialize the maximum number of iterations max_iter, the current number of iterations current_iter, the control parameters a, A, C, and the number of local search times; a is the convergence factor, and A and C are coefficient vectors;

[0041] Then, use the objective function ObjectiveFunction to calculate the fitness value of each gray wolf, sort according to the fitness value, select the gray wolf with the smallest fitness value as the α wolf, the second smallest gray wolf as the β wolf, and the third smallest gray wolf as the δ wolf, so as to select the top three best gray wolves;

[0042] Subsequently, iteratively update each individual in the gray wolf population, including:

[0043] During the iteration process, for each gray wolf:

[0044] (1) Generate a new individual using the following formula:

[0045] GW(t + 1) = (GW1 + GW2 + GW3) / 3;

[0046] (2) Use the greedy selection mechanism to decide whether to update the current gray wolf with the current new individual. If the fitness value of the current new individual is less than the fitness value of the current gray wolf individual, then use the current new individual to update the current gray wolf position; otherwise, retain the current gray wolf position;

[0047] (3) Take the current gray wolf as the initial solution of the local search algorithm, perform local search on the current gray wolf, and assign the optimized solution to the current gray wolf; after all gray wolves are updated in the current number of iterations, update a, A, C, and calculate the fitness value of each individual and update the positions of the α wolf, β wolf, and δ wolf;

[0048] Finally, judge whether the termination condition is reached; if current_iter < max_iter, then continue to iteratively update each individual in the gray wolf population and update the position of each gray wolf; otherwise, output the optimal solution as the scheduling scheme between the task and the virtual machine.

[0049] Preferably, according to the present invention, the current gray wolf is used as the initial solution of the local search algorithm, a local search is carried out on the current gray wolf, and the optimized solution is assigned to the current gray wolf; including:

[0050] First, the current gray wolf is used as the initial solution, and the number of local searches and the current number of iterations are initialized.

[0051] Then, a local search is performed; a random number rand is used to select one of the neighbor solutions; rand is a random number that is either 0 or 1;

[0052] If rand = 0, select the solution generated by the first neighborhood operator as the new solution; the first neighborhood operator includes: randomly generating two position indexes pos1 and pos2, and exchanging the values ​​at the corresponding positions of pos1 and pos2 in the current solution; if rand = 1, select the solution generated by the second neighborhood operator as the new solution; if the fitness value of the new solution is less than the fitness value of the current solution, use the new solution to update the solution;

[0053] Finally, determine whether the current number of iterations is greater than the local search number. If not, continue the local search; otherwise, output the solution.

[0054] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the heterogeneous cloud computing multi-objective task scheduling method based on the hybrid algorithm are implemented.

[0055] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the heterogeneous cloud computing multi-objective task scheduling method based on a hybrid algorithm.

[0056] The beneficial effects of the present invention are:

[0057] This paper proposes an improved Gray Wolf Optimization algorithm. Based on the improved Gray Wolf Optimization algorithm and the local search algorithm, a hybrid optimization algorithm is proposed. This hybrid algorithm effectively balances global and local search. Compared with existing optimization algorithms, this hybrid algorithm significantly improves convergence speed and solution accuracy. Using this hybrid algorithm as a task scheduling method for heterogeneous cloud computing can quickly and efficiently find the optimal mapping between tasks and virtual machines.

[0058] 2. The present invention significantly optimizes the maximum completion time, load balancing and resource utilization of heterogeneous cloud computing task scheduling, thereby effectively improving the quality of cloud computing services and the economic benefits of cloud providers. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1This is a diagram of the task scheduling system architecture of the present invention;

[0060] Figure 2 Schematic diagram of the first neighborhood operator of the present invention;

[0061] Figure 3 Schematic diagram of the second neighborhood operator of the present invention;

[0062] Figure 4 Schematic diagram of the HIGWOLS algorithm flow of the present invention;

[0063] Figure 5 This is a schematic diagram of the average maximum completion time results on the HPC2N dataset of the present invention;

[0064] Figure 6 This is a schematic diagram of the average maximum completion time results on the NASA iPSC dataset of the present invention;

[0065] Figure 7 This is a schematic diagram of the average resource utilization results on the HPC2N dataset of the present invention;

[0066] Figure 8 This is a schematic diagram of the average resource utilization results on the NASAiPSC dataset of the present invention;

[0067] Figure 9 This is a schematic diagram of the average load imbalance results on the HPC2N dataset of the present invention;

[0068] Figure 10 This is a schematic diagram of the average load imbalance results on the NASAiPSC dataset of the present invention;

[0069] Figure 11 This is a schematic diagram of the Rastrigin function results with shift and rotation of the present invention;

[0070] Figure 12 Schematic diagram of the convergence behavior results of the NASA iPSC real dataset under 2000 tasks of the present invention. DETAILED DESCRIPTION

[0071] The present invention will be further defined below with reference to the accompanying drawings and embodiments, but is not limited thereto.

[0072] Example 1

[0073] A hybrid algorithm-based multi-objective task scheduling method for heterogeneous cloud computing includes:

[0074] Based on the improved grey wolf optimization algorithm and local search algorithm, the mapping between virtual machines and tasks is realized, and finally multi-objective task scheduling optimization in heterogeneous cloud environments is achieved.

[0075] The heterogeneous cloud computing multi-objective task scheduling method runs on a task scheduling system; the task scheduling system includes multiple users, a series of tasks, a task scheduler, a resource manager and a data center;

[0076] Task scheduling in a cloud computing environment involves the rational scheduling and allocation of tasks to available computing resources. Figure 1 The proposed task scheduling system architecture in a heterogeneous computing environment is presented. In this architecture, multiple users submit a series of tasks T = {T1, T2, T3, ..., T n}, this series of tasks are independent, and the amount of computation is measured in MI (Million Instructions). Assume that a data center has p physical machines PM = {PM1, PM2, PM3, ..., PM p} and m virtual machines VM={VM1,VM2,VM3,...,VM m Virtual machine processing power is measured in MIPS (Million of Instructions Per Second). Virtual machines reflect resource heterogeneity through their varying processing capabilities. The resource manager is responsible for obtaining information about these physical and virtual machines and sending it to the task scheduler. Physical machine information includes the number of CPU cores, architecture, memory size, network bandwidth, and storage capacity; virtual machine information includes the number of CPUs, processor speed, memory size, network bandwidth, and storage capacity. The task scheduler, as the core component of the scheduling system, monitors task status and receives information about available resources. Based on the task scheduling algorithm, the task scheduler maps tasks to appropriate resources. Through this mechanism, the system can configure resources and execute tasks.

[0077] Multi-objective task scheduling optimization simultaneously optimizes three goals: reducing the overall task completion time, improving the balance of system load, and improving resource utilization. The objective function of multi-objective task scheduling optimization is:

[0078] ObjectiveFunction=Min(α*MS-β*RUR+γ * LB);

[0079] Among them, MS refers to completion time, RUR refers to resource utilization, and LB refers to load balancing; α, β, and γ are weight factors of different optimization objectives. The values ​​of α, β, and γ are

[0080] The goal of task scheduling is to maximize resource utilization, so the resource utilization needs to be multiplied by (-1*β).

[0081] The task scheduling problem in this paper is a multi-objective problem. The core of the multi-objective scheduling problem is to find an optimization strategy that aims to simultaneously minimize completion time, optimize load balancing, and maximize resource utilization. The problem definition includes: The calculation formula for completion time MS is as follows:

[0082]

[0083] Among them, CT ij represents the execution time of task i on virtual machine j, S j is the ID set of tasks assigned to virtual machine j; CT ij As shown below: In summary, the completion time is the maximum execution time among all virtual machines.

[0084]

[0085] Among them, TL i Indicates the number of instructions per million for the i-th task, VMP j represents the processing speed of the jth virtual machine;

[0086] The calculation formula for resource utilization RUR is as follows:

[0087]

[0088] Among them, CTL j is the time it takes for virtual machine j to complete task execution, and vmNum represents the total number of virtual machines. A higher utilization rate indicates better resource utilization, while a lower utilization rate indicates that more resources are wasted.

[0089] The calculation formula for load balancing LB is as follows:

[0090]

[0091] Among them, L j represents the load of the jth virtual machine, that is, the execution time of virtual machine j; Indicates the average load of all virtual machines, that is, the average execution time of the virtual machines.

[0092] Based on the hybrid algorithm HIGWOLS, including the improved grey wolf optimization algorithm Improved-GWO and the local search algorithm LS, efficient mapping between virtual machines and tasks is achieved; Figure 4 Shown, including:

[0093] Gray Wolf Optimization (GWO) mimics the hunting strategy of gray wolves and can search over a large area. However, GWO suffers from slow convergence and limited fine-grained search performance, especially when approaching the optimal solution, resulting in low search accuracy. To enhance the algorithm's convergence performance, the Improved Gray Wolf Optimization (Improved-GWO) algorithm introduces a greedy selection mechanism during updates, ensuring that the algorithm consistently evolves toward better solutions, thereby accelerating convergence. This improves convergence efficiency while retaining GWO's global search capabilities. LS focuses on searching within the neighborhood of a solution, enabling local optimization of the solution to improve solution accuracy and enhance search efficiency through structured neighborhood operations. However, the search performance of LS is significantly affected by the initial solution. If the initial solution is of poor quality, it may be difficult to find the global optimal solution. Furthermore, by combining Improved-GWO and LS, the optimal solution can be found quickly and efficiently. Improved-GWO ensures rapid discovery of the global potential optimal solution, while LS performs detailed optimization within the local region of the potential optimal solution to improve search efficiency and solution accuracy. Improved-GWO and LS are combined as an optimization algorithm for task scheduling to obtain a high-quality mapping solution between virtual machines and tasks. The hybrid optimization algorithm is called HIGWOLS.

[0094] First, initialize a gray wolf population. Each individual in the gray wolf population represents a mapping scheme between virtual machines and tasks. Assume that GW represents a gray wolf population and is defined as follows:

[0095] GW={GW1,GW2,...,GW i ,...,GW N};

[0096] Where N is the size of the population, GW i represents the i-th individual, which is expressed as:

[0097] GW i ={p1,p2,...,p j ,...,p v};

[0098] p j is the position value of the gray wolf individual in the jth dimension;

[0099] Initialize the population individual p j Calculated as follows:

[0100] p j =(upper-lower)*random+lower;

[0101] Among them, upper is the upper bound of the search area, lower is the lower bound of the search area; random is a random number between [0,1];

[0102] Initialize the maximum number of iterations max_iter, the current iteration number current_iter, the control parameters a, A, C, and the number of local searches simultaneously; a is the convergence factor, and A and C are coefficient vectors.

[0103] Then, calculate the fitness value of each gray wolf using the objective function ObjectiveFunction, sort them according to the fitness value, select the gray wolf with the minimum fitness value as the α wolf, the second smallest gray wolf as the β wolf, and the third smallest gray wolf as the δ wolf, so as to select the top three best gray wolves.

[0104] Subsequently, iteratively update each individual in the gray wolf population, including:

[0105] During the iteration process, for each gray wolf:

[0106] (1) Generate a new individual using the following formula:

[0107] GW(t + 1) = (GW1 + GW2 + GW3) / 3;

[0108] (2) Use the greedy selection mechanism to decide whether to update the current gray wolf with the current new individual. If the fitness value of the current new individual is less than the fitness value of the current gray wolf individual, then use the current new individual to update the current gray wolf position; otherwise, keep the current gray wolf position.

[0109] (3) Take the current gray wolf as the initial solution (Solution) of the local search algorithm, perform local search on the current gray wolf, and assign the optimized solution to the current gray wolf; after all gray wolves are updated in the current iteration number, update a, A, C, calculate the fitness value of each individual, and update the positions of the α wolf, β wolf, and δ wolf.

[0110] Finally, judge whether the termination condition is reached; if current_iter < max_iter, then continue to iteratively update each individual in the gray wolf population and update the position of each gray wolf; otherwise, output the best solution as the scheduling scheme between the task and the virtual machine. The pseudocode of the HIGWOLS algorithm is shown in Algorithm 1.

[0111]

[0112] The Gray Wolf Optimizer (see Mirjalili, S., Mirjalili, S.M., & Lewis, A. (2014). Grey wolf optimizer. Advances in engineering software, 69, 46-61.) is a metaheuristic algorithm inspired by the natural and social hunting behavior of gray wolves. Gray wolves mostly live in groups and have a strict social hierarchy. They are divided into four hierarchies: α, β, δ, and ω. The first-ranked wolf, α, is the leader of the pack, responsible for hunting and making decisions. The second-ranked wolf, α, assists α in making decisions and reinforces α's orders. δ obeys α and β and dominates ω. ω, the lowest-ranking wolf, obeys the other three ranks and is the last in the pack to be allowed to feed. Beyond this strict hierarchy, pack hunting is an interesting social behavior of gray wolves, consisting primarily of three phases: stalking, encirclement, and attack.

[0113] Encircling prey:

[0114] Gray wolves surround their prey when hunting. The mathematical model for surrounding prey is as follows:

[0115] D=|C·GW p (t)-GW(t)|;

[0116] GW(t+1)=GW p (t)-A·D;

[0117] t represents the current iteration number, and D represents the distance between a wolf and its prey. Assuming that the search space of the wolf pack is v, the position of the i-th wolf is GW i ={p1,p2,...,p v}. p i The value of is in [0,vmNum-1]. A and C are coefficient vectors, and their calculation formulas are as follows:

[0118] A=2a·r1–a;

[0119] C = 2·r2;

[0120] Where r1 and r2 are random vectors in [0,1], and the component of a decreases linearly from 2 to 0 during the iteration. The calculation formula of a is as follows:

[0121] a=2-2t / Max_Iter;

[0122] Max_Iter is the maximum number of iterations.

[0123] Hunting:

[0124] Hunting is led by the alpha wolf, and the beta and delta wolves also participate. To simulate the hunting behavior of gray wolves, it is assumed that the alpha wolf (the best solution), beta (the second-best solution), and delta (the third-best solution) have strong abilities to identify the positions of potential prey. Therefore, the omega wolves are forced to update their positions based on the positions of the alpha, beta, and delta wolves. The formula for individual update is as follows:

[0125] GW(t + 1) = (GW1 + GW2 + GW3) / 3;

[0126] Where GW1, GW2, and GW3 are calculated by the following formulas:

[0127] GW1 = GW α (t) - A1·D α ;

[0128] GW2 = GW β (t) - A2·D β ;

[0129] GW3 = GW δ (t) - A3·D δ ;

[0130] D α 、D β and D } are calculated by the following formulas:

[0131] D α = |c1·GW α (t) - GW(t)|;

[0132] D β = |C2·GW β (t) - GW(t)|;

[0133] D δ = |C3·GW δ (t) - GW(t)|;

[0134] Attacking prey and Search prey:

[0135] The gray wolves update based on the positions of alpha, beta, and delta. They disperse to explore different areas and gather to attack prey. If A > 1 or A < -1, the gray wolves move away from the prey to find a better target, thus achieving global search. If -1 < A < 1, the gray wolves approach the prey. C is a random vector in [0, 2], which helps to increase randomness during the optimization process, facilitating exploration and avoiding local optima.

[0136] The pseudo-code of the Grey Wolf Optimization algorithm is shown in Algorithm 2:

[0137]

[0138] The Grey Wolf Optimization algorithm has the disadvantages of slow convergence speed and low convergence accuracy. To improve the convergence speed, this study improved the traditional Grey Wolf Optimization algorithm by introducing a greedy selection mechanism in the update step. Specifically, it includes:

[0139] First, initialize the maximum number of iterations (max_iter), the current number of iterations (current_iter), the population size, the control parameters a, A, C, and randomly initialize the population according to the upper and lower limit variables.

[0140] Then, calculate the fitness values and select the top three best grey wolves.

[0141] Next, perform iterative updates for each grey wolf. For each grey wolf, generate a new individual using the following formula.

[0142] GW(t + 1) = (GW1 + GW2 + GW3) / 3;

[0143] Evaluate whether its adaptability is better than that of the current grey wolf. If the new individual is better than the current individual, use it to update the position of the grey wolf; otherwise, retain the current position of the grey wolf. This mechanism reduces the interference of inferior solutions on the search direction during the search process of the algorithm, avoids invalid search and convergence oscillation caused by accepting inferior solutions, and thus speeds up the convergence speed. After all grey wolves are updated in the current number of iterations, update a, A, C, and calculate the fitness value of each individual to update α, β, and δ.

[0144] Finally, determine whether the termination condition is reached. If current_iter < max_iter, go back to execute the above update part. Otherwise, output the best solution. The Grey Wolf Optimization algorithm with the greedy selection mechanism introduced is called Improved - GWO. The pseudo-code of the Improved - GWO algorithm is shown in Algorithm 3:

[0145]

[0146] The local search algorithm is a heuristic method for solving optimization problems. Since many optimization problems are NP - hard problems, the solution time of their optimal solutions grows exponentially with the increase of the problem scale. In many practical scenarios, a solution close to the optimal one is sufficient to meet the requirements. The heuristic algorithm can quickly provide such an approximate solution. As an approximate algorithm, this algorithm starts from a certain initial solution, generates a series of neighbor solutions by performing neighborhood operations, and selects a new current solution from them according to a specific selection strategy. This process is repeated continuously until the preset termination condition is met.

[0147] Local search is implemented using a neighborhood operator. The difference between different local search algorithms lies in the definition of the neighborhood operator and the selection of neighbor solutions, which is also the key to determining the quality of the algorithm. The present invention uses two neighborhood operators to generate two neighbor solutions, and uses a random number rand to select one of the neighbor solutions. rand is a random number that is either 0 or 1. If rand = 0, the solution generated by the first neighborhood operator is selected as the new solution. This operator helps to perform a fine search in the local area of ​​the solution through perturbation-type transformation, thereby improving the quality of the solution. Otherwise, the solution generated by the second neighborhood operator is selected. This operator can conduct a wider exploration in the solution space through a large range of exchange operations, which helps the algorithm to escape the local optimum.

[0148] The current gray wolf is used as the initial solution of the local search algorithm. A local search is performed on the current gray wolf, and the optimized solution is assigned to the current gray wolf. This includes:

[0149] First, the current gray wolf is used as the initial solution, and the number of local searches and the current number of iterations are initialized.

[0150] Then, a local search is performed; a random number rand is used to select one of the neighbor solutions; rand is a random number that is either 0 or 1;

[0151] If rand=0, the solution generated by the first neighborhood operator is selected as the new solution; the first neighborhood operator includes: randomly generating two position indexes pos1 and pos2, and exchanging the values ​​at the corresponding positions of pos1 and pos2 in the current solution; Figure 2 shown.

[0152] If rand=1, the solution generated by the second neighborhood operator is selected as the new solution; if the fitness value of the new solution is less than the fitness value of the current solution (Solution), the new solution is used to update the Solution; if Figure 3 shown.

[0153] Finally, determine whether the current number of iterations is greater than the local search number. If not, continue the local search; otherwise, output the solution.

[0154] The pseudo code of the local search algorithm is shown in Algorithm 4.

[0155]

[0156]

[0157] The following is the experimental part:

[0158] CloudSim (see: Calheiros, RN, Ranjan, R., Beloglazov, A., De Rose, CA, & Buyya, R. (2011). CloudSim: a toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms. Software: Practice and experience, 41(1), 23-50.) provides a general and extensible simulation framework for modeling, simulating, and experimenting with cloud computing infrastructure and application services. Table 1 provides the configuration information of the host and virtual machines. The efficiency of the HIGWOLS optimization algorithm is evaluated using the HPC2N and NASA iPSC (Feitelson, DG, Tsafrir, D., & Krakov, D. (2014). Experience with using the parallel workloads archive. Journal of Parallel and Distributed Computing, 74(10), 2967-2982.) datasets. The dataset details are shown in Table 2.

[0159] Table 1 Cloud entities;

[0160]

[0161]

[0162] Table 2. Detailed information of the dataset;

[0163]

[0164] The performance of HIGWOLS is compared with the existing methods PGSAO (see: Shao, K., Fu, H., & Wang, B. (2023). An efficient combination of genetic algorithm and particle swarm optimization for scheduling data-intensive tasks in heterogeneous cloud computing. Electronics, 12(16), 3450.), GA-GWO (see: Behera, I., & Sobhanayak, S. (2024). Task scheduling optimization in heterogeneous cloud computing environments: A hybrid GA-GWO approach. Journal of Parallel and Distributed Computing, 183, 104766.) and Whale Optimization Algorithm (WOA) (see: Mirjalili, S., & Lewis, A. (2016). The whale optimization algorithm. Advances in engineering software, 95, 51-67.) in terms of completion time, resource utilization and load imbalance indicators. MakeSpan refers to the completion time of the last virtual machine to complete the task execution, and is one of the key indicators for evaluating the quality of scheduling schemes. Resource utilization refers to the extent to which resources are effectively used within a period of time. The higher the resource utilization, the more fully the resources are utilized and the less resource waste. The degree of load imbalance reflects the difference in the amount of tasks assigned to virtual machines. This difference may cause some virtual machines to be overly busy and some virtual machines to be relatively idle, thereby affecting the processing time and resource utilization efficiency of the entire system. In a cloud computing environment, a better scheduling algorithm can effectively reduce load imbalance. The calculation formula for load imbalance is:

[0165]

[0166] In the experimental setup, the number of iterations for each algorithm was uniformly set to 100. The parameter configurations for each algorithm are detailed in Table 3. To ensure the reliability of the results, each experiment was independently repeated 30 times, and the average of these 30 experiments was taken as the final result.

[0167] Table 3 Algorithm parameters;

[0168]

[0169]

[0170] User satisfaction with quality of service (QoS) often depends on completion time. Fast response times and strong task completion capabilities can improve user experience and enhance the competitiveness of cloud providers. In this paper, completion time is evaluated by taking input from HPC2N and NASA iPSC workload archives. Figure 5 and Figure 6The average completion time results for the proposed algorithm and the baseline algorithms on the HPC2N and NASA iPSC datasets are plotted. For each dataset, 500-3000 tasks were considered. For the 500 tasks, on the HPC2N dataset, the average completion times for the proposed hybrid algorithm, PGSAO, GA-GWO, and WOA were 4092.136222, 4322.786653, 5026.528863, and 4961.852850, respectively. HIGWOLS improved PGSAO, GA-GWO, and WOA by 5.3%, 18.6%, and 17.5%, respectively. On the NASA iPSC dataset, the average completion times for the four algorithms were 428.025605, 462.551669, 566.879456, and 613.596065, respectively. The proposed hybrid algorithm improves PGSAO, GA-GWO, and WOA by 7.5%, 24.5%, and 30.2%, respectively. For 1000 tasks, on the HPC2N dataset, HIGWOLS achieves an average completion time of 5488.858385, which is 10.0%, 24.0%, and 17.8% better than PGSAO, GA-GWO, and WOA, respectively. On the NASA iPSC dataset, the proposed hybrid algorithm achieves an average completion time of 759.881320, which is 3.4%, 20.3%, and 25.0% better than PGSAO, GA-GWO, and WOA, respectively. For 3000 tasks on the HPC2N dataset, the proposed hybrid algorithm's average completion time (14541.836436) also outperformed the other three comparison algorithms, with improvements of 19.9%, 37.7%, and 8.3%, respectively. On the NASA iPSC dataset, the proposed hybrid algorithm achieved an average completion time of 2007.187549, representing improvements of 3.6%, 17.8%, and 18.7%, respectively, over PGSAO, GA-GWO, and WOA. Tables 4 and 5 detail the performance simulation results of the proposed algorithm and the baseline algorithms on different datasets and with varying numbers of tasks. Overall, on the HPC2N dataset, HIGWOLS achieved completion time improvements of 5.3%-19.9% ​​over PGSAO, 18.6%-37.7% over GA-GWO, and 8.3%-17.8% over WOA. On the NASA iPSC dataset, the proposed hybrid algorithm achieved a 3.3%-7.5% improvement in completion time compared to PGSAO, a 17.8%-24.5% improvement compared to GA-GWO, and a 16.3%-30.2% improvement compared to WOA. These results demonstrate that the HIGWOLS algorithm can complete tasks in a shorter time span, thereby reducing task response time and enabling the system to process requests more quickly, improving overall processing capabilities and user experience.

[0171] Table 4 Average maximum completion time on the HPC2N dataset;

[0172]

[0173] Table 5 Average maximum completion time on NASA iPSC dataset;

[0174]

[0175] Figure 7 and Figure 8 The average resource utilization results of the proposed algorithm and the baseline algorithms on the HPC2N and NASA iPSC datasets are plotted. For 1500 tasks, on the HPC2N dataset, the average resource utilization of the proposed hybrid algorithm, PGSAO, GA-GWO, and WOA are 0.996815, 0.878655, 0.795237, and 0.810105, respectively. The proposed hybrid algorithm improves on GWO, GA-GWO, and WOA by 13.4%, 25.3%, and 23.0%, respectively. On the NASA iPSC dataset, the average resource utilization of the proposed hybrid algorithm, PGSAO, GA-GWO, and WOA are 0.995734, 0.926752, 0.832983, and 0.740158, respectively. The proposed hybrid algorithm improves on PGSAO, GA-GWO, and WOA by 7.4%, 19.5%, and 34.5%, respectively. When the number of tasks is 2500, on the HPC2N dataset, HIGWOLS achieves an average resource utilization of 0.972515, representing improvements of 19.6%, 29.4%, and 17.1% over PGSAO, GA-GWO, and WOA, respectively. On the NASA iPSC dataset, HIGWOLS achieves an average resource utilization of 0.992019, representing improvements of 4.4%, 16.4%, and 24.4% over PGSAO, GA-GWO, and WOA, respectively. Tables 6 and 7 detail the performance simulation results of the proposed algorithm and the baseline algorithm for different datasets and different numbers of tasks. The results demonstrate that the proposed hybrid algorithm outperforms the comparison algorithms. Therefore, the HIGWOLS algorithm is able to more efficiently allocate resources, reduce resource waste, and thus improve the overall performance and responsiveness of the system.

[0176] Table 6 Average resource utilization on the HPC2N dataset;

[0177]

[0178] Table 7 Average resource utilization on NASA iPSC dataset

[0179]

[0180] Figure 9and Figure 10 The average load imbalance of the proposed algorithm and the baseline algorithm on the HPC2N and NASA iPSC datasets is plotted, respectively. Tables 8 and 9 detail the performance simulation results of the proposed algorithm and the baseline algorithm on different datasets and with varying numbers of tasks. Overall, on the HPC2N dataset, HIGWOLS improves load imbalance by 74.0%-99.4% compared to PGSAO, 73.7%-99.1% compared to GA-GWO, and 43.2%-99.5% compared to WOA. On the NASA iPSC dataset, HIGWOLS improves load imbalance by 83.8%-95.9% compared to PGSAO, 88.3%-98.0% compared to GA-GWO, and 93.2%-98.9% compared to WOA. The reduction in average load imbalance demonstrates that the proposed algorithm can more effectively balance task loads, resulting in better allocation and utilization of system resources. This reduces the overload of individual virtual machines, improves overall computational efficiency, and ensures system stability and reliability.

[0181] Table 8 Average load imbalance on the HPC2N dataset;

[0182]

[0183] Table 9 Average load imbalance on NASA iPSC dataset;

[0184]

[0185] In order to evaluate the optimization performance of the HIGWOLS algorithm, the Rastrigin function with rotation and shift characteristics is used as a benchmark function. Figure 11 As shown in the figure, the four algorithms show significant performance differences: the optimal values ​​obtained by PGSAO, GA-GWO, and WOA algorithms are 152.52, 145.81, and 213.86, respectively, while the proposed algorithm significantly outperforms the other compared algorithms with an optimal value of 124.29. The experimental results show that HIGWOLS has obvious advantages in avoiding local optimality and global optimization capabilities. Figure 12 The convergence characteristics of PGSAO, GA-GWO, WOA, and the proposed hybrid algorithm are demonstrated in a 2000-task scheduling scenario based on the HPC2N workload. The curves show that all four tested algorithms are able to optimize the objective function, but HIGWOLS exhibits high volatility in the initial iterations and low variability in the final iterations. The downward trend of the curves indicates that as the number of iterations increases, individuals in the population are updating their positions to improve the results, demonstrating good convergence. This also indicates that the proposed algorithm has a better ability to balance search and exploitation during the iteration process than the comparison algorithms.

[0186] In summary, the present invention proposes a task scheduling method HIGWOLS. The HIGWOLS algorithm is a combination of the Improved-GWO algorithm and the LS algorithm. It effectively integrates the advantages of the Improved-GWO algorithm and the LS algorithm, improves the optimization performance, and thus can improve scheduling efficiency. To verify the effectiveness of the HIGWOLS algorithm, a series of simulation tests were conducted using CloudSim. Under different numbers of tasks on the real data sets HPC2N and NASA iPSC, it was compared in detail with the PGSAO, GA-GWO, and WOA algorithms in terms of maximum completion time, resource utilization, and load imbalance. The experimental results show that HIGWOLS outperforms these three comparison algorithms, demonstrating its advantages in task scheduling.

[0187] Example 2

[0188] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the heterogeneous cloud computing multi-objective task scheduling method based on a hybrid algorithm described in Example 1 are implemented.

[0189] Example 3

[0190] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the heterogeneous cloud computing multi-objective task scheduling method based on a hybrid algorithm described in Example 1.

Claims

1. A method for scheduling multi-objective tasks in heterogeneous cloud computing based on a hybrid algorithm, characterized in that: Including: Based on the improved Grey Wolf Optimization algorithm and local search algorithm, the mapping between virtual machines and tasks is realized, and finally the multi-objective task scheduling optimization in heterogeneous cloud environments is achieved.

2. The method for scheduling multi-objective tasks in heterogeneous cloud computing based on a hybrid algorithm according to claim 1, characterized in that: The multi-objective task scheduling method for heterogeneous cloud computing runs in the task scheduling system; The task scheduling system includes multiple users, a series of tasks, a task scheduler, a resource manager, and a data center; Multiple users submit a series of tasks T = {T1, T2, T3, ..., T n }, assuming that a data center has p physical machines PM={PM1,PM2,PM3,...,PM p } and m virtual machines VM={VM1,VM2,VM3,...,VM m }, virtual machines reflect the heterogeneity of resources through different processing capabilities; the resource manager is responsible for obtaining information about these physical machines and virtual machines, and sending this physical machine information and virtual machine information to the task scheduler; physical machine information includes the number of CPU cores, architecture, memory size, network bandwidth and storage capacity; virtual machine information includes the number of CPUs, processor speed, memory size, network bandwidth and storage capacity; the task scheduler is responsible for monitoring the status of tasks and receiving information about available resources. The task scheduler maps tasks to appropriate resources based on the task scheduling algorithm.

3. The method for scheduling multi-objective tasks in heterogeneous cloud computing based on a hybrid algorithm according to claim 1, characterized in that: The multi-objective task scheduling optimization synchronously optimizes three objectives: reducing the overall completion time of tasks, improving the balance of system load, and increasing resource utilization; the objective function ObjectiveFunction of the multi-objective task scheduling optimization is: ObjectiveFunction=Min(α*MS-β*RUR+γ * LB); Among them, MS refers to the completion time, RUR refers to the resource utilization rate, and LB refers to the load balance degree; α, β, and γ are the weight factors of different optimization objectives.

4. The method for scheduling multi-objective tasks in heterogeneous cloud computing based on a hybrid algorithm according to claim 3 is characterized in that: The calculation formula for the completion time MS is as follows: Among them, CT 6j represents the execution time of task i on virtual machine j, S j is the ID set of tasks assigned to virtual machine j; CT 6j As shown below: Among them, TL6 represents the number of instructions per million of the i-th task, VMP j represents the processing speed of the jth virtual machine; The calculation formula for the resource utilization rate RUR is as follows: Among them, CTL j is the time it takes for virtual machine j to complete task execution, and vmNum represents the total number of virtual machines; The calculation formula for the load balance degree LB is as follows: Among them, L j represents the load of the jth virtual machine, that is, the execution time of virtual machine j; Indicates the average load of all virtual machines, that is, the average execution time of the virtual machines.

5. The method for scheduling multi-objective tasks in heterogeneous cloud computing based on a hybrid algorithm according to claim 1, characterized in that: The hybrid algorithm HIGWOLS is based on the improved Grey Wolf Optimization algorithm Improved-GWO and local search algorithm LS to achieve an efficient mapping between virtual machines and tasks; including: First, initialize a Grey Wolf population, where each individual in the Grey Wolf population represents a mapping scheme between virtual machines and tasks; assume GW represents a Grey Wolf population, defined as follows: GW={GW1,GW2,...,GW i ,...,GW N }; Where N is the size of the population, GW i represents the i-th individual, which is expressed as: GW i ={p1,p2,...,p j ,..., p v }; p j is the position value of the gray wolf individual in the jth dimension; Initialize the population individual p j Calculated as follows: p j =(upper-lower)*random+lower; Among them, upper is the upper bound of the search area, lower is the lower bound of the search area; random is a random number between [0, 1]; At the same time, initialize the maximum number of iterations max_iter, the current number of iterations current_iter, control parameters a, A, C, and the number of local search times; a is the convergence factor, and A and C are coefficient vectors; Then, use the objective function ObjectiveFunction to calculate the fitness value of each Grey Wolf, sort according to the fitness value, select the Grey Wolf with the smallest fitness value as the α wolf, the second smallest Grey Wolf as the β wolf, and the third smallest Grey Wolf as the δ wolf, so as to select the top three best Grey Wolves; Subsequently, iteratively update each individual in the Grey Wolf population; Finally, judge whether the termination condition is reached; if current_iter < max_iter, then continue to iteratively update each individual in the Grey Wolf population and update the position of each Grey Wolf; otherwise, output the best solution as the scheduling scheme between tasks and virtual machines.

6. The method for scheduling multi-objective tasks in heterogeneous cloud computing based on a hybrid algorithm according to claim 5, characterized in that: Iteratively update each individual in the Grey Wolf population; including: During the iteration process, for each Grey Wolf: (1) Generate a new individual using the following formula: GW(t + 1) = (GW1 + GW2 + GW3) / 3; (2) Use the greedy selection mechanism to decide whether to update the current Grey Wolf with the current new individual. If the fitness value of the current new individual is less than the fitness value of the current Grey Wolf individual, then use the current new individual to update the current Grey Wolf position; otherwise, keep the current Grey Wolf position. (3) Take the current gray wolf as the initial solution of the local search algorithm, conduct a local search on the current gray wolf, and assign the optimized solution to the current gray wolf; after all gray wolves are updated in the current iteration, update a, A, C, calculate the fitness value of each individual, and update the positions of α wolf, β wolf, and δ wolf.

7. A method for scheduling multi-objective tasks in heterogeneous cloud computing based on a hybrid algorithm according to any one of claims 1 to 6, characterized in that: Use the current gray wolf as the initial solution of the local search algorithm, conduct a local search on the current gray wolf, and assign the optimized solution to the current gray wolf; including: First, the current gray wolf is used as the initial solution, and the number of local searches and the current number of iterations are initialized. Then, a local search is performed; a random number rand is used to select one of the neighbor solutions; rand is a random number that is either 0 or 1; If rand = 0, select the solution generated by the first neighborhood operator as the new solution; the first neighborhood operator includes: randomly generating two position indexes pos1 and pos2, and exchanging the values ​​at the corresponding positions of pos1 and pos2 in the current solution; if rand = 1, select the solution generated by the second neighborhood operator as the new solution; if the fitness value of the new solution is less than the fitness value of the current solution, use the new solution to update the solution; Finally, determine whether the current number of iterations is greater than the number of local searches; if it is not, switch to local search and continue execution; otherwise, output Solution.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the heterogeneous cloud computing multi-objective task scheduling method based on the hybrid algorithm according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the heterogeneous cloud computing multi-objective task scheduling method based on a hybrid algorithm according to any one of claims 1 to 7 are implemented.

Images