A cloud computing task scheduling method based on multi-strategy fusion improved Harris Hawk algorithm
By improving the Harris Eagle algorithm through multi-strategy fusion, a cloud computing task scheduling model was constructed, which solved the problem of uneven resource utilization in cloud computing, achieved optimal task scheduling, improved system computing efficiency, and reduced operating costs.
Patent Information
- Application Number
- CN202411258500.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-09
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-09-09
AI Technical Summary
Existing cloud computing services suffer from uneven utilization of computing node resources, leading to rapid virtual machine wear and single points of failure. This results in low efficiency and high costs, and how to achieve optimal task scheduling remains an unsolved problem.
A multi-strategy fusion improvement of the Harris Eagle algorithm is adopted to construct a heterogeneous cloud computing network scenario, build a task scheduling system model, and map the task scheduling problem into the improved Harris Eagle algorithm. The optimal task scheduling strategy is solved by multi-strategy fusion improvement of the Harris Eagle algorithm. The algorithm parameters are optimized by combining chaotic opposition learning, gray wolf optimization and adaptive exponential decrease strategy to achieve precise management of task scheduling.
It improves the accuracy of task scheduling, shortens task completion time, enhances system computing efficiency, and reduces operating costs. In particular, it has the advantages of fast convergence and high optimization accuracy in high-dimensional and high-complexity situations.
Smart Images

Figure CN118779116B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of cloud computing resource management, and in particular relates to a cloud computing task scheduling method integrating a multi-strategy improved Harris Hawk algorithm. Background Art
[0002] As the wave of Internet technology continues to penetrate into all aspects of production and life, we have entered the era of the Internet of Everything. Consumers in various fields have increasingly obvious demands for computing services, presenting a large number of diverse application scenarios, and the amount of data generated by these applications is exploding. The growing demand for computing services has prompted academic researchers and corporate organizations to continuously explore new and efficient computing service models. Based on the mutual development and complementary integration of mature computing models such as virtualization technology, grid computing, distributed computing and parallel computing, the cloud computing service model was born in 2006. Since its development, it has brought innovative changes in many fields of the Internet, promoting the rapid development of business delivery models, data processing methods and online application management. In order to effectively utilize and analyze massive user data, embracing cloud computing has become a "compulsory course" for global Internet service providers. When the cloud computing service model began to emerge in various industries, its powerful advantages promoted the deep integration of information technology and industrial production in all aspects, and gradually produced huge changes in the industrial field starting from the infrastructure. Industrial cloud represents the development direction of "Internet +" manufacturing. Cloud computing resource virtualization technology breaks through the limitations of traditional industries and basic software and hardware resources. Based on the basic facilities of Internet cloud computing technology, it uses its computing, storage and deployment capabilities to integrate industrial production resources and applications, collect, store and analyze industrial data, and provide diversified services for the industry. However, how to achieve optimal task scheduling in the process of industrial operation is still a problem to be solved. In addition, the current cloud service providers' computing node resource utilization imbalance causes excessive virtual machine loss and single point failure problems, and there are problems such as low efficiency and high cost. Summary of the invention
[0003] In order to solve the above problems, the present invention proposes a cloud computing task scheduling method based on a multi-strategy fusion improved Harris Eagle algorithm, which can accurately manage task scheduling, achieve optimal scheduling, improve system computing time and reduce operating costs.
[0004] To achieve the above object, the technical solution adopted by the present invention is: a cloud computing task scheduling method based on multi-strategy fusion improved Harris Hawk algorithm, comprising the steps of:
[0005] S10, constructing a cloud computing heterogeneous network scenario, and representing data centers, virtual machines, and independent tasks according to the scenario information;
[0006] S20, build a task scheduling system model based on the cloud computing heterogeneous network scenario, and express the objective function and constraints of the strategy problem to be optimized;
[0007] S30, mapping the task scheduling system model to the improved Harris Hawk algorithm and completing the algorithm parameter setting; mapping the task scheduling to the position of the individual in the Harris Hawk algorithm, and mapping the joint optimization model of the maximum completion time and the average resource utilization rate to the fitness evaluation function of the hawk group searching for prey;
[0008] S40, uses multi-strategy fusion to improve the Harris Eagle algorithm to solve the optimal task scheduling strategy, iterates to find the optimal solution, and obtains the optimized optimal task scheduling strategy.
[0009] Furthermore, the construction of a cloud computing heterogeneous network scenario includes: the set of physical hosts used in the system data center is represented as H={h 1 ,h 2 ,…,h k}, the set of deployed virtual machines is represented as V={v 1 ,v 2 ,…,v m}, each virtual machine resource configuration is represented by a triplet V j ={Com j ,Ram j ,Bw j},Com j Represents virtual machine v j Computing power, Ram j Represents virtual machine v j Memory capacity, Bw j Represents virtual machine v j The bandwidth capacity of the task to be processed is represented by T={t 1 ,t 2 ,…,t n}, each independent task attribute is represented by a two-tuple t i ={Len i ,Size i}, Len i Represented as an independent task t i Required computing resources, Size i represents an independent task t i The amount of data transmitted; the system ultimately realizes resource mapping between independent tasks T and deployed virtual machines V.
[0010] Furthermore, a task scheduling system model is built, including:
[0011] Aiming at the joint optimization problem of maximum completion time and average resource utilization in independent task scheduling in cloud computing, a corresponding target optimization mathematical model is constructed;
[0012] A complete task schedule is defined as s=<P,MK,AVRU> , P is the task matrix, P contains the reasonable mapping p of all tasks t to the deployed virtual machines v tv , MK and AVRU are the maximum completion time and average resource utilization of task t respectively.
[0013] Furthermore, the task scheduling optimization problem with the maximum completion time and average resource utilization as the target in task scheduling is expressed as an integer programming model:
[0014] ;
[0015] Among them, λ 1 represents the weight constraint of optimizing task completion time; 2 represents the weight constraint for optimizing the average resource utilization; 1 ∈[0,1], λ2∈[0,1], and satisfies λ 1 +λ 2 =1.
[0016] Furthermore, a calculation model is established for the maximum completion time and average resource utilization indicators:
[0017] Set the binary decision variable SV ik (1≤i, ≤N) value represents task t i and t k Are they assigned to the same virtual machine at the same time? If they are assigned to the same virtual machine, SV ik =1, otherwise SV ik =0;
[0018] Task i Transfer to matching virtual machine v j The transmission time is:
[0019] ;
[0020] Task i In the specified virtual machine v j The execution time required is:
[0021] ;
[0022] Com j Represents virtual machine v j The computing power of Len i Represented as an independent task t i The computing resources required;
[0023] Task i Matching virtual machine v jThe time required to complete the assigned workload is:
[0024] ;
[0025] etc kj Represents task t k In the specified virtual machine v j The execution time required for task t is SVki, and k and t i Whether they are assigned to the same virtual machine at the same time, p ij Indicates a task;
[0026] Task i When allocating virtual machines v j The actual time when the above execution ends is:
[0027] ;
[0028] For any task matrix P, the maximum task completion time and average resource utilization model are:
[0029] ;
[0030] Get the system optimization target model:
[0031] ;
[0032] λ1 and λ2 are weights, MK and AVRU are the maximum task completion time and average resource utilization, respectively. j Represents virtual machine v j The memory capacity of ret k Represents task t k Matching virtual machine v j The time required to complete the assigned workload, FinT k Task k When allocating virtual machines v j The actual time when the execution ended;
[0033] In the formula, constraint 1: ensure that there is no priority in task execution, and each task is allowed to be executed only once on the assigned specific virtual machine; constraint 2: ensure that tasks can be assigned to virtual machines only when memory requirements are met; if SV is met ik =1, constraint 3 must be established to ensure that the execution time of the same virtual machine does not overlap. At a certain moment, the virtual machine is fully processing the assigned task t i , can process the next assigned task t k .
[0034] Furthermore, mapping the task scheduling system model to the improved Harris Hawk algorithm includes: scheduling the tasks in the model problem to the scheduling strategy of a virtual machine, where the task scheduling strategy is a vector; mapping the scheduling strategy to the position of the Harris Hawk individual in the Harris Hawk algorithm, and mapping the optimization target model to the fitness evaluation function of the food source sought by the eagle group. The smaller the fitness evaluation value, the better the target optimization effect obtained.
[0035] Furthermore, the multi-strategy fusion improved Harris Eagle algorithm is used to solve the optimal scheduling strategy, including the following steps:
[0036] S31, algorithm parameter setting: set the population size, maximum number of iterations and optimization dimension;
[0037] S32, population initialization: using chaotic adversarial learning strategy to initialize population position;
[0038] S33, prey escape energy update: Adopt adaptive exponential decreasing strategy to update the prey escape energy, and control the transition of the hawk flock from the search stage to the prey chasing stage; if the absolute value of the escape energy is greater than 1, it enters the prey searching stage, and if the absolute value of the escape energy is less than 1, it enters the chasing hunting stage;
[0039] S34, prey search phase: the exploration mechanism of the gray wolf optimization algorithm is introduced to conduct elite representative search on the original search strategy and update the prey location information;
[0040] S35, pursuit method selection: randomly select one of four improved pursuit methods, namely, soft trapping, hard trapping, progressive fast dive soft trapping and progressive fast dive hard trapping, to develop a new position according to different escape probabilities and escape energies, and finally select a position with a better fitness value to update the prey position information;
[0041] S36, population position update: after the new position is determined, the corresponding position and fitness value of the eagle group are updated, and the position and optimal fitness of the prey are updated;
[0042] S37, iteration end judgment: judge whether the iteration is ended according to the iteration cutoff condition, if not ended, re-enter step S33 until the iteration is ended to obtain the optimal scheduling strategy corresponding to the prey position.
[0043] Furthermore, by combining chaotic mapping with adversarial learning strategy to realize the chaotic optimization of the initial position of the eagle group, the initial position x of the population is obtained. i,j , the mapping formula is:
[0044] ;
[0045] Where i represents the Harris Hawk number, j represents the search space position number, and x i,jFor the Harris Hawk location, The opposite position of Harris Hawk I, UB j is the upper limit of position, lb j is the lower limit of position, Z i is the chaotic number obtained by the i-th eagle in the t-th iteration, t is the current iteration number, μ is the chaos coefficient, and F is the integer programming model function; according to the initial position of the population, the optimal fitness value of the population under the initial conditions and the optimal position of the population are set as the position of the prey.
[0046] Furthermore, the exploration mechanism of the gray wolf optimization algorithm is introduced to improve the original search strategy to an elite representative search, including:
[0047] The elite population includes four social levels from high to low: α, β, δ, and ω. α, β, and δ represent the elite individuals with the best, second-best, and second-best fitness values in the population, respectively. In each generation of search and update, β and δ individuals help α locate the eagle group, and the elite individuals guide ω individuals to update their positions and cooperate to explore better solutions. Let q be a random number in the range of [0,1]. When q≥0.5, the elite search strategy is used to obtain x re is the elite individual position obtained in the current iteration, replacing the random individual; when q<0.5, a representative search mechanism is adopted to store the α, β and δ elite individuals in the iteration process. The upper limit of storage is the population size. If it exceeds the upper limit, the stored elite representative is randomly overwritten.
[0048] Furthermore, the prey escape energy is updated by using an exponential decreasing strategy to improve the escape energy E and introducing an adaptive feedback factor H s , the escape energy of the current iteration is adjusted using the individual update feedback received in the previous iteration.
[0049] The beneficial effects of adopting this technical solution are:
[0050] The present invention first constructs a heterogeneous network scenario architecture under a cloud computing system, and represents the basic data composition of the core elements of the system. According to the indicators to be optimized, a joint optimization model of maximum completion time and average resource utilization is constructed. Then, the model optimization problem is mapped to the algorithm, and a fast and accurate solution is achieved through a multi-strategy improved Harris Eagle algorithm to obtain the optimal scheduling strategy. Finally, the cloud computing task scheduling system completes the deployment calculation of the task according to the scheduling strategy, improves the precise management of task scheduling, realizes optimal scheduling, and realizes the improvement of system computing time and the reduction of operating costs.
[0051] In the cloud computing task scheduling system environment of the present invention, cloud service providers need to avoid excessive virtual machine loss and single point failure problems caused by uneven computing node resource utilization to reduce operating costs, while shortening task completion time as much as possible to protect the interests of cloud users. To this end, the present invention takes average resource utilization and maximum task completion time as joint optimization goals to study the cloud computing task scheduling algorithm.
[0052] The present invention proposes a task scheduling strategy based on multi-strategy improved Harris Eagle optimization, which can be used for optimizing and solving complex models in task scheduling. The algorithm has better optimization performance than traditional heuristic algorithms due to its unique optimization method. After improvement, it can effectively reduce the maximum completion time of the task and improve the average resource utilization of the virtual machine. It still has the advantages of fast convergence speed and high optimization accuracy when the model complexity is high. When the improved algorithm provided by the present invention is used to solve the joint optimization model problem of the maximum completion time and average resource utilization of task scheduling, it has the advantages of high optimization accuracy, strong flexibility, fast convergence and stability. The improved strategy in the present invention has strong adaptability to the model, especially for solving high-dimensional and high-complexity target optimization models, it has certain performance advantages, and can achieve the optimal scheduling strategy in a short time, saving time and operating costs for the system. The present invention can be well matched with the task scheduling model, and its convergence accuracy and convergence speed are significantly improved compared with the original algorithm. In addition, for task scheduling scenarios with different distributions and different scales, the strategy proposed in the present invention has certain advantages in reducing the maximum completion time and improving the average resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 A schematic diagram of a cloud computing task scheduling method of the present invention that integrates multiple strategies and improves the Harris Eagle algorithm;
[0054] Figure 2 This is a flow chart of a method for training a user tag model in an embodiment of the present invention;
[0055] Figure 3 It is a schematic diagram of the comparison of the convergence curves in the embodiments of the present invention;
[0056] Figure 4 This is a schematic diagram of the maximum completion time comparison in an embodiment of the present invention;
[0057] Figure 5 Schematic diagram of average resource utilization in an embodiment of the present invention. DETAILED DESCRIPTION
[0058] 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.
[0059] In this embodiment, see Figure 1 As shown, the present invention proposes a cloud computing task scheduling method based on a multi-strategy fusion improved Harris Hawk algorithm, comprising the steps of:
[0060] S10, construct a cloud computing heterogeneous network scenario and represent data centers, virtual machines and independent tasks based on the scenario information.
[0061] Preferably,
[0062] The construction of cloud computing heterogeneous network scenarios includes: the set of physical hosts used by the system data center is represented as H={h 1 ,h 2 ,…,h k}, the set of deployed virtual machines is represented as V={v 1 ,v 2 ,…,v m}, each virtual machine resource configuration is represented by a triplet V j ={Com j ,Ram j ,Bw j},Com j Represents virtual machine v j Computing power, Ram j Represents virtual machine v j Memory capacity, Bw j Represents virtual machine v j The bandwidth capacity of the task to be processed is represented by T={t 1 ,t 2 ,…,t n}, each independent task attribute is represented by a two-tuple t i ={Len i ,Size i}, Len i Represented as an independent task t i Required computing resources, Size i represents an independent task t i The amount of data transmitted; the system finally realizes the resource mapping between independent tasks T and deployed virtual machines V
[0063] The scenario of building a cloud computing heterogeneous network includes: in the current cloud computing task scheduling system, k independent physical hosts and m virtual machine nodes have been deployed, and n independent tasks need to be processed; the set of physical hosts used in the system data center is represented as H={h 1 ,h 2 ,…,h k}, the set of deployed virtual machines is represented as V={v 1 ,v 2 ,…,v m}, each virtual machine resource configuration is represented by a triplet V j ={Com j ,Ram j ,Bw j},Com j Represents virtual machine v j Computing power, Ram j Represents virtual machine v j Memory capacity, Bw j Represents virtual machine v j The bandwidth capacity of the task to be processed is represented by T={t 1 ,t 2 ,…,t n}, each independent task attribute is represented by a two-tuple t i ={Len i ,Size i}, Len i Represented as an independent task t i Required computing resources, Size i represents an independent task t i The amount of data transmitted; the system finally realizes the resource mapping between independent tasks T and deployed virtual machines V, P=(p ij )(1≤i≤n,1≤j≤m) is represented as the task allocation matrix, and the binary variable p ij =1 represents task t i Assign to virtual machine v j , p ij =0 means unallocated.
[0064] S20, based on the cloud computing heterogeneous network scenario, build a task scheduling system model and express the objective function and constraints of the strategy problem to be optimized.
[0065] Build a task scheduling system model, including:
[0066] Aiming at the joint optimization problem of maximum completion time and average resource utilization in independent task scheduling in cloud computing, a corresponding target optimization mathematical model is constructed;
[0067] A complete task schedule is defined as s=<P,MK,AVRU> , P is the task matrix, P contains the reasonable mapping p of all tasks t to the deployed virtual machines v tv , MK and AVRU are the maximum completion time and average resource utilization of task t respectively.
[0068] S30, mapping the task scheduling system model to the improved Harris Hawk algorithm and completing the algorithm parameter setting; mapping the task scheduling to the position of the individual in the Harris Hawk algorithm, and mapping the joint optimization model of the maximum completion time and the average resource utilization rate to the fitness evaluation function of the hawk group searching for prey.
[0069] Preferably, the task scheduling optimization problem with the maximum completion time and average resource utilization as the target in task scheduling is expressed as an integer programming model:
[0070] ;
[0071] Among them, λ 1 represents the weight constraint of optimizing task completion time; 2 represents the weight constraint for optimizing the average resource utilization; 1 ∈[0,1], λ2∈[0,1], and satisfies λ 1 +λ 2 =1.
[0072] Establish a calculation model for the maximum completion time and average resource utilization indicators:
[0073] Set the binary decision variable SV ik (1≤i, ≤N) value represents task t i and t k Are they assigned to the same virtual machine at the same time? If they are assigned to the same virtual machine, SV ik =1, otherwise SV ik =0;
[0074] Task i Transfer to matching virtual machine v j The transmission time is:
[0075] ;
[0076] Size i Indicates the amount of data transmitted by independent task ti, Bw j Represents virtual machine v j bandwidth capacity;
[0077] Task i In the specified virtual machine v j The execution time required is:
[0078] ;
[0079] Com j Represents virtual machine v j The computing power of Len i Represented as an independent task t iThe computing resources required;
[0080] Task i Matching virtual machine v j The time required to complete the assigned workload is:
[0081] ;
[0082] etc kj Represents task t k In the specified virtual machine v j The execution time required for task t is SVki, and k and t i Whether they are assigned to the same virtual machine at the same time, p ij Indicates a task;
[0083] Task i When allocating virtual machines v j The actual time when the above execution ends is:
[0084] ;
[0085] For any task matrix P, the maximum task completion time and average resource utilization model are:
[0086] ;
[0087] Get the system optimization target model:
[0088] ;
[0089] λ1 and λ2 are weights, MK and AVRU are the maximum task completion time and average resource utilization, respectively. j Represents virtual machine v j The memory capacity of ret k Represents task t k Matching virtual machine v j The time required to complete the assigned workload, FinT k Task k When allocating virtual machines v j The actual time when the execution ended;
[0090] In the formula, constraint 1: ensure that there is no priority in task execution, and each task is allowed to be executed only once on the assigned specific virtual machine; constraint 2: ensure that tasks can be assigned to virtual machines only when memory requirements are met; if SV is met ik =1, constraint 3 must be established to ensure that the execution time of the same virtual machine does not overlap. At a certain moment, the virtual machine is fully processing the assigned task t i, can process the next assigned task t k .
[0091] As an optimization solution of the above embodiment:
[0092] Mapping the task scheduling system model to the improved Harris Hawk algorithm includes: scheduling the tasks in the model problem to a scheduling strategy p for a virtual machine ij ∈{0,1},i∈n,j∈m, integrate and express a task scheduling strategy as,S=[s 1 ,s 2 ,…,s n ],s n ∈{0, m}, all feasible task scheduling strategies are S N ; Map the scheduling strategy to the position of the Harris Hawk individual in the Harris Hawk algorithm, and map the optimization target model to the fitness evaluation function Fit(S) of the food source sought by the hawk group. The smaller the fitness evaluation value, the better the target optimization effect. After completing the mapping of the model and the algorithm, the execution of the algorithm also requires setting system parameters. The relevant parameters are shown in Table 1. In order to avoid the impact of dynamic allocation on the analysis of the scheduling strategy and algorithm performance results, it is assumed in this embodiment that the allocation strategy used for scheduling tasks to different virtual machines will remain fixed once it is determined until the end of the algorithm iteration.
[0093] Table 1 is a parameter setting table of the cloud computing system in the embodiment;
[0094]
[0095] Specifically, the multi-strategy fusion improved Harris Eagle algorithm is used to solve the optimal scheduling strategy, such as Figure 2 As shown, the steps include:
[0096] S31, algorithm parameter setting: set the population size N, the maximum number of iterations MaxIt and the optimization dimension Dim.
[0097] S32, population initialization: Use chaotic adversarial learning strategy to initialize the population position.
[0098] Since the original algorithm uses a common random initialization method, although the individual positions can be dispersed, the distribution cannot be guaranteed to be uniform, and the quality of the generated initial solution cannot be predicted, which directly affects the diversity of the algorithm's solution space and the convergence speed. In order to make the algorithm better fit the optimized model and increase the diversity of the initial population, the present invention combines the Tent chaotic mapping with the adversarial learning strategy to achieve chaos optimization of the initial position of the eagle group, and obtains the initial position x of the population. i,j , the mapping formula is:
[0099] ;
[0100] Where i represents the Harris Hawk number, j represents the search space position number, and x i,j For the Harris Hawk location, The opposite position of Harris Hawk I, UB j is the upper limit of position, lb j is the lower limit of position, Z i is the chaotic number obtained by the i-th eagle in the t-th iteration, t is the current iteration number, μ is the chaos coefficient, and F is the integer programming model function; according to the initial position of the population, the optimal fitness value of the population under the initial conditions and the optimal position of the population are set as the position of the prey.
[0101] S33, prey escape energy update: Adopt adaptive exponential decreasing strategy to update the prey escape energy, and control the transition of the hawk flock from the search phase to the prey chasing phase; if the absolute value of the escape energy is greater than 1, it enters the prey searching phase, and if the absolute value of the escape energy is less than 1, it enters the chasing hunting phase.
[0102] Prey escape energy update: The original algorithm uses a linearly decreasing escape energy E to simulate the physical changes of the prey during the escape process, and controls the transition between the balanced global search and local development phases. However, setting E as a simple linear decreasing function is not enough to describe the complex energy decay process of prey in natural hunting, and the final value decreasing to 0 does not conform to the final state of prey energy consumption. The present invention adopts an exponential decreasing strategy to improve the escape energy E in order to better simulate the energy changes of the prey. On this basis, the adaptive feedback factor H is introduced. s , the escape energy of the current iteration is adjusted using the individual update feedback received in the previous iteration, further improving the effective conversion between the global search and local development phases, thereby improving the accuracy of the population search and accelerating the convergence speed of the algorithm.
[0103] The improved adaptive exponential decreasing energy update formula is as follows:
[0104] ;
[0105] Among them, S i Record the individual that updates the best position during the iteration, feedback factor H s The greedy strategy is used to compare the fitness values of individuals before and after the update, and the number of optimizations of the individual best value during the iteration process is recorded. In the early stage of the algorithm iteration, the E value of the improved energy update strategy has a wider range and a faster change speed, which expands the global exploration range; in the later stage of the iteration, the E value is smaller and the change speed is slower, which makes the population concentrate on the prey position in the local development stage, thereby ensuring that the algorithm achieves a faster convergence speed.
[0106] S34, prey search phase: The exploration mechanism of the gray wolf optimization algorithm is introduced to perform elite representative search on the original search strategy and update the prey location information to improve the global exploration and spatial diversity of the Harris Hawk algorithm.
[0107] In the original algorithm, Harris hawks will search for prey according to different roosting strategies. Some hawks update their roosting points according to the location of the prey and the average position of all hawks, while other hawks choose to roost at close range based on the random position of another hawk. This position update mechanism does not save the current individual position, nor does it utilize the valuable elite individual positions. There are limitations in information sharing between populations, which may lead to insufficient search, reduced solution diversity, and easy to fall into local optimality.
[0108] The exploration mechanism of the Grey Wolf Optimization Algorithm is introduced to improve the original search strategy by elite representative search, including:
[0109] The elite population includes four social levels from high to low: α, β, δ, and ω. α, β, and δ represent the elite individuals with the best, second-best, and second-best fitness values in the population, respectively. In each generation of search and update, β and δ individuals help α locate the eagle group, and the elite individuals guide ω individuals to update their positions and cooperate to explore better solutions. Let q be a random number in the range of [0,1]. When q≥0.5, the elite search strategy is used to obtain x re is the elite individual position obtained in the current iteration, replacing the random individual; when q<0.5, a representative search mechanism is adopted to store the α, β and δ elite individuals in the iteration process. The upper limit of storage is the population size. If it exceeds the upper limit, the stored elite representative is randomly overwritten.
[0110] The improved update formula for the search phase is:
[0111] ;
[0112] Among them, x re Indicates the elite individual position obtained in the current iteration, r 1 and r 2 is the search step control parameter, x Rreb represents two individuals randomly selected from the five best representatives in the stored solution, x Rre represents two individuals randomly selected from the entire storage space, γ 1 and γ 2 and represent representative exploration control parameters, x r1 and x r2 is a random individual in the population, r 6 Convergence control parameter, MaxIt is the maximum number of iterations;
[0113] x re The positions x of individuals α, β, and δ are represented by the eliteα (t), x β (t) and x δ (t) is determined, and the search update method adopts the search mechanism of the gray wolf optimization algorithm, and the calculation formula is:
[0114] ;
[0115] Among them, C 1 , C 2 and C 3 Expressed as a swing control vector, the swing control vector is calculated as follows:
[0116] ;
[0117] A 1 , A 2 and A 3 Expressed as the convergence control vector, the convergence control vector is calculated as follows:
[0118] .
[0119] S35, pursuit method selection: randomly select one of the four improved pursuit methods, namely soft encirclement, hard encirclement, progressive rapid dive soft encirclement and progressive rapid dive hard encirclement, to develop a new position according to different escape probabilities and escape energies, and finally select the position with better fitness value to update the prey position information.
[0120] The choice of pursuit mode is determined by the escape energy E and the escape energy p, where p is a random number in [0,1].
[0121] D51. Soft pursuit: When |E|≥0.5 and p≥0.5 are satisfied, the prey still has sufficient energy and a greater chance of escaping. It will try to move its position misleadingly to deceive the pursuer. Therefore, the Harris Hawk will choose to move around the prey to gradually consume its escape energy instead of attacking directly:
[0122] ;
[0123] Among them, j represents the jumping distance of the prey to escape, r 5 is a random number in [0,1].
[0124] D52. Progressive rapid dive soft capture: When |E|≥0.5 and p<0.5 is satisfied, compared with the soft capture strategy, the prey has sufficient energy to escape but the chance of escape is lower. Harris's hawk will adjust according to the prey's deceptive escape behavior, introduce the Levy flight concept to simulate the prey's deceptive escape route and the Harris's hawk's irregular rapid descent. The hawk group will compare the obtained Y and Z with the current position fitness value to decide which strategy to use for the prey raid:
[0125] ;
[0126] Among them, Dim is the dimension of the problem to be solved, and S represents a random vector. At present, the Mantegna algorithm is usually used to achieve Levy stable distribution, and its distribution expression and random step length calculation formula are as follows:
[0127] ;
[0128] in is the factorial calculation function, , μ and v are random numbers that obey the normal distribution;
[0129] D53. Hard Encirclement: When |E|<0.5 and p≥0.5 are satisfied, the prey's energy for escaping is greatly reduced in the face of continuous pursuit and hunting, and it is marked as an exhausted individual. Harris's Hawk chooses the hard encirclement strategy to hunt directly, and updates its position based on the distance between itself and the prey:
[0130] ;
[0131] D54. Progressive fast dive hard capture: When |E|<0.5 and p<0.5 are satisfied, the prey's escape energy is very low and the escape probability is also low. Therefore, Harris's hawks choose a progressive fast dive hard capture strategy to narrow the gap between the average position of the hawk group and the prey. The execution method is as follows:
[0132] ;
[0133] Among them, X m (t) represents the mean of the N Harris hawks in this iteration population, and the calculation formula is as follows:
[0134] ;
[0135] S36, population position update: After the new position is determined, complete the update of the corresponding position and fitness value of the eagle group, and update the position and optimal fitness of the prey.
[0136] The final position x of each Harris hawk individual corresponds to a cloud computing task scheduling strategy S. The feasible scheduling solutions represented by all positions constitute the solution space of the MIHHO optimization algorithm, among which the feasible solution with the best global fitness value is the optimal multi-objective scheduling solution found. Since task scheduling is a discrete optimization problem, the value ranges of the two main variables, tasks and virtual machines, in its mathematical description are both a set of natural numbers, and the solution space obtained is also a discrete solution set. Therefore, it is necessary to add discretization processing in the encoding process of the MIHHO algorithm to make it suitable for cloud computing task scheduling optimization problems:
[0137] ;
[0138] After completing the current iteration, update the optimal fitness of the population BestF, and update the optimal position of the population, that is, the prey position x Rb ;
[0139] ;
[0140] S37, iteration end judgment: judge whether the iteration is ended according to the iteration cutoff condition, if not ended, re-enter step S33 until the iteration is ended to obtain the optimal scheduling strategy corresponding to the prey position.
[0141] S40, uses multi-strategy fusion to improve the Harris Eagle algorithm to solve the optimal task scheduling strategy, iterates to find the optimal solution, and obtains the optimized optimal task scheduling strategy.
[0142] Compared with the existing methods, the comprehensive performance of the present invention is significantly improved after the improvement: first, the population is initialized by introducing the Tent chaotic sequence and the adversarial learning fusion strategy to enhance the population diversity and improve the quality of the initial solution, thereby improving the population optimization efficiency; then, the energy adaptive exponential decreasing factor is introduced as a control parameter for global search and local development to improve the accuracy and convergence speed of the algorithm; finally, the elite representative exploration mechanism of the gray wolf optimization algorithm is applied to the global search stage of the population to enhance the global search capability of the algorithm and avoid falling into the local optimal state.
[0143] Experiments have shown that Figure 3-Figure 5 Compared with some classic meta-heuristic algorithms, such as particle swarm algorithm PSO, bee colony algorithm ABC, Harris Hawk optimization algorithm HHO, and some advanced improved algorithms, such as improved Harris Hawk algorithm MHHO and cuckoo-whale hybrid optimization algorithm CWOA, the improved algorithm (MIHHO) provided by the present invention has the advantages of high optimization accuracy, strong flexibility, fast convergence and stability when solving the joint optimization model problem of maximum completion time and average resource utilization of task scheduling.
[0144] 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 cloud computing task scheduling method based on multi-strategy fusion improved Harris Hawk algorithm, characterized in that: Includes steps: S10, constructing a cloud computing heterogeneous network scenario, and representing data centers, virtual machines, and independent tasks according to the scenario information; S20, build a task scheduling system model based on the cloud computing heterogeneous network scenario, and express the objective function and constraints of the strategy problem to be optimized; Build a task scheduling system model, including: Aiming at the joint optimization problem of maximum completion time and average resource utilization in independent task scheduling in cloud computing, a corresponding target optimization mathematical model is constructed; A complete task schedule is defined as s=<P,MK,AVRU> , P is the task matrix, P contains the reasonable mapping p of all tasks t to the deployed virtual machines v tv , MK and AVRU are the maximum completion time and average resource utilization of task t respectively; S30, mapping the task scheduling system model to the improved Harris Hawk algorithm and completing the algorithm parameter setting; mapping the task scheduling to the position of the individual in the Harris Hawk algorithm, and mapping the joint optimization model of the maximum completion time and the average resource utilization rate to the fitness evaluation function of the hawk group searching for prey; The task scheduling optimization problem with the maximum completion time and average resource utilization as the target in task scheduling is expressed as an integer programming model: ; Among them, λ1 represents the weight constraint of optimizing task completion time; λ2 represents the weight constraint of optimizing average resource utilization; λ1∈[0,1], λ2∈[0,1], and satisfies λ1+λ2=1; Establish a calculation model for the maximum completion time and average resource utilization indicators: Set the binary decision variable SV ik (1≤i, ≤N) value represents task t i and t k Are they assigned to the same virtual machine at the same time? If they are assigned to the same virtual machine, SV ik =1, otherwise SV ik =0; Task i Transfer to matching virtual machine v j The transmission time is: ; Size i Indicates the amount of data transmitted by independent task ti, Bw j Represents virtual machine v j bandwidth capacity; Task i In the specified virtual machine v j The execution time required is: ; Com j Represents virtual machine v j The computing power of Len i Represented as an independent task t i The computing resources required; Task i Matching virtual machine v j The time required to complete the assigned workload is: ; etc kj Represents task t k In the specified virtual machine v j The execution time required for task t is SVki, and k and t i Whether they are assigned to the same virtual machine at the same time, p ij Indicates a task; Task i When allocating virtual machines v j The actual time when the above execution ends is: ; For any task matrix P, the maximum task completion time and average resource utilization model are: ; Get the system optimization target model: ; λ1 and λ2 are weights, MK and AVRU are the maximum task completion time and average resource utilization, respectively. j Represents virtual machine v j The memory capacity of ret k Represents task t k Matching virtual machine v j The time required to complete the assigned workload, FinT k Task k When allocating virtual machines v j The actual time when the execution ended; In the formula, constraint 1: ensure that there is no priority in task execution, and each task is allowed to be executed only once on the assigned specific virtual machine; constraint 2: ensure that tasks can be assigned to virtual machines only when memory requirements are met; if SV is met ik =1, constraint 3 must be established to ensure that the execution time of the same virtual machine does not overlap. At a certain moment, the virtual machine is fully processing the assigned task t i , can process the next assigned task t k ; The multi-strategy fusion improved Harris Eagle algorithm is used to solve the optimal scheduling strategy, including the following steps: S31, algorithm parameter setting: set the population size N, the maximum number of iterations MaxIt and the optimization dimension Dim; S32, population initialization: using chaotic adversarial learning strategy to initialize population position; S33, prey escape energy update: Adopt adaptive exponential decreasing strategy to update the prey escape energy, and control the transition of the hawk flock from the search stage to the prey chasing stage; if the absolute value of the escape energy is greater than 1, it enters the prey searching stage, and if the absolute value of the escape energy is less than 1, it enters the chasing hunting stage; S34, prey search phase: the exploration mechanism of the gray wolf optimization algorithm is introduced to perform elite representative search on the original search strategy and update the prey location information to improve the global exploration and spatial diversity of the Harris Hawk algorithm; S35, pursuit method selection: randomly select one of four improved pursuit methods, namely, soft trapping, hard trapping, progressive fast dive soft trapping and progressive fast dive hard trapping, to develop a new position according to different escape probabilities and escape energies, and finally select a position with a better fitness value to update the prey position information; S36, population position update: after the new position is determined, the corresponding position and fitness value of the eagle group are updated, and the position and optimal fitness of the prey are updated; S37, iteration end judgment: determine whether the iteration is ended according to the iteration end condition, if not, re-enter step S33 until the iteration is ended to obtain the optimal scheduling strategy corresponding to the prey position; S40, uses multi-strategy fusion to improve the Harris Eagle algorithm to solve the optimal task scheduling strategy, iterates to find the optimal solution, and obtains the optimized optimal task scheduling strategy.
2. According to claim 1, a cloud computing task scheduling method based on multi-strategy fusion improved Harris Hawk algorithm is characterized in that: The construction of cloud computing heterogeneous network scenarios includes: the set of physical hosts used in the system data center is represented as H={h1,h2,…,h k }, the set of deployed virtual machines is represented as V={v1,v2,…,v m }, each virtual machine resource configuration is represented by a triplet V j ={Com j ,Ram j ,Bw j },Com j Represents virtual machine v j Computing power, Ram j Represents virtual machine v j Memory capacity, Bw j Represents virtual machine v j The bandwidth capacity of the task to be processed is represented by T={t1,t2,…,t n }, each independent task attribute is represented by a two-tuple t i ={Len i ,Size i }, Len i Represented as an independent task t i Required computing resources, Size i represents an independent task t i The amount of data transmitted; the system ultimately realizes resource mapping between independent tasks T and deployed virtual machines V.
3. According to the multi-strategy fusion improved Harris Hawk algorithm cloud computing task scheduling method of claim 1, it is characterized by: Mapping the task scheduling system model to the improved Harris Hawk algorithm includes: scheduling the tasks in the model problem to the scheduling strategy of a virtual machine, where the task scheduling strategy is a vector; mapping the scheduling strategy to the position of the Harris Hawk individual in the Harris Hawk algorithm, and mapping the optimization target model to the fitness evaluation function of the food source sought by the hawk group. The smaller the fitness evaluation value, the better the target optimization effect.
4. According to claim 1, a cloud computing task scheduling method based on multi-strategy fusion improved Harris Hawk algorithm is characterized in that: By combining chaotic mapping with adversarial learning strategy to realize the chaotic optimization of the initial position of the eagle group, the initial position x of the population is obtained. i,j , the mapping formula is: ; Where i represents the Harris Hawk number, j represents the search space position number, and x i,j For the Harris Hawk location, The opposite position of Harris Hawk I, UB j is the upper limit of position, lb j is the lower limit of position, Z i is the chaotic number obtained by the i-th eagle in the t-th iteration, t is the current iteration number, μ is the chaos coefficient, and F is the integer programming model function; according to the initial position of the population, the optimal fitness value of the population under the initial conditions and the optimal position of the population are set as the position of the prey.
5. According to claim 1, a cloud computing task scheduling method based on multi-strategy fusion improved Harris Hawk algorithm is characterized in that: The exploration mechanism of the Grey Wolf Optimization Algorithm is introduced to improve the original search strategy by elite representative search, including: The elite population includes four social levels from high to low: α, β, δ, and ω. α, β, and δ represent the elite individuals with the best, second-best, and second-best fitness values in the population, respectively. In each generation of search and update, β and δ individuals help α locate the eagle group, and the elite individuals guide ω individuals to update their positions and cooperate to explore better solutions. Let q be a random number in the range of [0,1]. When q≥0.5, the elite search strategy is used to obtain x re is the elite individual position obtained in the current iteration, replacing the random individual; when q<0.5, a representative search mechanism is adopted to store the α, β and δ elite individuals in the iteration process. The upper limit of storage is the population size. If it exceeds the upper limit, the stored elite representative is randomly overwritten.
6. The cloud computing task scheduling method of the multi-strategy fusion improved Harris Hawk algorithm according to claim 1 is characterized in that: Prey escape energy update: An exponential decreasing strategy is used to improve the escape energy E and an adaptive feedback factor H is introduced s , the escape energy of the current iteration is adjusted using the individual update feedback received in the previous iteration.
