A task scheduling method for manufacturing enterprise data space system based on decomposition strategy

Through a multi-objective optimization method based on decomposition strategy, the local optimization and multi-objective consideration of task scheduling in the data space system of the manufacturing enterprise are solved, minimizing task completion time and load balancing are achieved, and system performance and stability are improved.

CN114371915BActive Publication Date: 2024-11-22NORTHEASTERN UNIV CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210029767.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-12
Publication Date
2024-11-22
Estimated Expiration
2042-01-12

AI Technical Summary

Technical Problem

The existing manufacturing enterprise data space system task scheduling methods have local optimal traps, weak global search capabilities, and it is difficult to take into account the weight configuration of multi-objective functions, resulting in poor resource allocation and task sorting, and unable to effectively improve system performance.

Method used

A multi-objective optimization method based on decomposition strategy is adopted, through bilayer chromosome encoding and cross-mutation operators, the decomposition task is scheduled to multiple subproblems, and information among subpopulations is used for cooperative solution to ensure the feasibility and effectiveness of the solution.

Benefits of technology

It minimizes task completion time and system load balancing, improves algorithm solution efficiency and user satisfaction, and ensures stable operation and performance improvement of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114371915B_ABST
    Figure CN114371915B_ABST
Patent Text Reader

Abstract

This invention discloses a task scheduling method for a manufacturing enterprise data space system based on a decomposition strategy, belonging to the field of manufacturing enterprise data space technology. From a multi-objective perspective, it can optimize the allocation of system computing resources while minimizing task processing time, aiming to minimize task completion time and balance system load, thus maintaining system stability while ensuring user satisfaction. By decomposing the multi-objective scheduling task into several multi-objective sub-problems through a decomposition strategy, and using information exchange among individuals in the sub-population to solve them simultaneously in a cooperative manner, the algorithm's solution efficiency is improved. A two-layer chromosome encoding method is used to ensure that all generated solutions are feasible, effectively avoiding the shortcomings of single-objective algorithms that rely on human experience to set weights for different objectives, thus improving the performance of the data space system and ensuring its stable operation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of manufacturing enterprise data space, and in particular to a method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy. Background Art

[0002] With the rapid development of computer technology, industrial Internet and Internet of Things network technology, the construction process of manufacturing enterprise data space system has been further promoted; the manufacturing enterprise data space system is a new system model developed on the basis of distributed computing, parallel computing and grid computing. Manufacturing enterprises can effectively integrate computing resources, service resources and storage resources through the data space system, and uniformly allocate and manage resources according to user needs; with the increasing number of users and service requests, how to efficiently and reasonably allocate and utilize limited system resources, effectively schedule the massive tasks submitted by users, and maximize user satisfaction are the key technical issues faced in the construction of manufacturing enterprise data space system; the process of the data space system completing a normal user service can be described as: multiple users submit tasks to the cloud data space system, and the system allocates different user tasks to appropriate computing resources (virtual machines) for execution through the task scheduler. In this process, the task scheduler needs to solve the two problems of system computing resource allocation and task sorting at the same time, which is a typical NP-complete problem, that is, there is no polynomial time optimization algorithm that can solve this problem;

[0003] Currently, for the data space of manufacturing enterprises, many task scheduling methods still use traditional heuristic or hyper-heuristic algorithms, such as genetic algorithms, Max-Min algorithms and Min-Min algorithms. These algorithms generally have the disadvantages of being easily trapped in local optimality and having weak global search capabilities during the optimization process. In terms of objective function setting, most task scheduling methods only focus on how to shorten the processing time of tasks during optimization, but ignore the fact that the system load should be balanced on the basis of meeting user requirements to ensure the stable operation of the system. In addition, when dealing with multiple objective functions, most existing methods use a linear weighted method to simplify the original multi-objective problem into a single-objective optimization problem, and then use the traditional single-objective optimization algorithm to solve it, instead of using a multi-objective optimization algorithm based on the Pareto dominance perspective to solve the problem. Since the weight setting of each objective function is usually based on manual experience, it is difficult to achieve a reasonable configuration of the objective weights, and it is also difficult to obtain the global optimal resource allocation and task sorting scheme, which cannot provide effective algorithm support for the performance improvement of the data space system of manufacturing enterprises.

[0004] The patent "Cloud computing task scheduling method based on improved genetic algorithm" (ZL202011086788.4) and the patent "A genetic and differential hybrid evolutionary cloud computing task scheduling algorithm based on a previous disaster strategy" (ZL201911082199.6) both proposed cloud computing task scheduling methods based on improved evolutionary algorithms. However, these scheduling methods are aimed at optimizing the task processing time of the cloud computing task scheduling process using a single-objective optimization algorithm. Decision makers cannot simultaneously take into account the impact of different objective functions on scheduling when selecting a scheduling plan, and it is difficult to provide effective algorithmic support for the performance improvement of the data space system of manufacturing enterprises. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention proposes a method for scheduling tasks in the data space system of a manufacturing enterprise based on a decomposition strategy. From a multi-objective perspective, it can achieve optimal allocation of system computing resources while minimizing task processing time, effectively avoiding the shortcoming of relying on manual experience to set weights for different objectives in the single-objective algorithm, improving the performance of the data space system and ensuring the stable operation of the system.

[0006] The technical solution adopted by the present invention is:

[0007] A method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy comprises the following steps:

[0008] S1: Preprocess the required data according to task requirements;

[0009] S2: Set the maximum number of iterations termination criterion MaxGen and set the parameter value;

[0010] S3: Randomly initialize the first generation of initial population Q = (x 1 ,x 2 ,...,x N );

[0011] Calculate the value of each individual x in population Q i The fitness value F of (i=1,2,...,N) i =(f 1 (x), f 2 (x)), where the first objective function f 1 (x) is the total completion time of the task, and the second objective function f 2 (x) is the virtual machine load balancing index;

[0012] S4: Set the current number of iterations gen = 1;

[0013] S5: Set the temporary set R = Φ, α = 1;

[0014] S6: Assign each individual in population Q to the corresponding subpopulation P according to its fitness value 1 ,P 2 ,...,P K middle;

[0015] S7: Pair population P α For each individual x β (β=1,2,...,N) Perform crossover mutation operation to generate offspring individuals, and store the offspring individuals in a temporary set R;

[0016] S8: Merge the current population Q with the temporary set R, Q:=R∪Q;

[0017] S9: Set α=α+1. If α≤K, return to S7. If α>K, execute the next step.

[0018] S10: set gen=gen+1;

[0019] S11: According to the current number of iterations of the algorithm, determine whether the algorithm meets the termination condition MaxGen. If the termination condition is not met, repeat S5 to S10. If the condition is met, output all non-dominated solutions in the population Q. At this time, each non-dominated solution corresponds to a task scheduling scheme, and the decision maker can select a scheme to execute according to his preference or demand;

[0020] The specific process of S1 includes the following steps:

[0021] S1.1: Select the optimization model for task scheduling problem according to the requirements;

[0022] S1.2: Select the task scheduling target according to the demand;

[0023] S1.3: Select individual coding method according to needs;

[0024] The specific description or optimization model of the task scheduling problem in S1.1 is as follows:

[0025] All tasks uploaded by users are considered as the total task and handed over to the enterprise data space system for task scheduling. According to the scheduling results, the tasks are assigned to n virtual machines for processing, so that the task allocation results meet the multiple scheduling goals of the system task decision system. The overall task contains m subtasks, and the subtasks are independent of each other. The delay and loss in the data transmission process are ignored, all computing resources in the system are mapped to virtual machines, and the number of subtasks is greater than the number of virtual machines.

[0026] The task scheduling objectives of S1.2 are as follows:

[0027] Assign m tasks to n virtual machines so that the total completion time of m tasks is the shortest and the load of virtual machines is balanced;

[0028] The chromosome encoding method of S1.3 is as follows:

[0029] Chromosome encoding adopts integer encoding. Each chromosome represents a scheduling scheme, reflecting the mapping relationship between task scheduling and resources. The length of the chromosome is equal to the number of tasks m. The gene bit number on the chromosome represents the task number, and the value range is {1,2,...,m}. The gene value on the gene bit represents the virtual machine number assigned to the task, and the value range is {1,2,...,n}.

[0030] The parameters in S2 include: population size N, number of subpopulations K, subpopulation size S, crossover probability p c , mutation probability p m , where N = K × S;

[0031] The specific process of S3 is as follows:

[0032] Evaluate each individual x in population Q i (i=1,2,...,N)’s fitness value; the individual fitness value evaluation includes the first objective function f 1 (x) is the total completion time of the task and the second objective function f 2 (x) is the load balancing index of the virtual machine. The fitness value function is as follows:

[0033] S3.1: Fitness value function f about the total completion time of the task 1 (x):

[0034] When there are s tasks assigned to the jth virtual machine M j When the virtual machine M j The time required to complete the assigned tasks

[0035]

[0036] Among them, t ij For task T i In the virtual machine M j The time required for processing is independent of each other. After obtaining the task completion time on each virtual machine, the total task completion time C max for:

[0037] f 1 (x) = C max =max{C j},j∈{1,2,...,n} (2)

[0038] Among them, C j Represents virtual machine M j The time required to complete all tasks, n is the number of virtual machines;

[0039] Task allocation in S3.1: when multiple tasks are allocated to a virtual machine, the tasks are executed according to the first-in-first-out principle;

[0040] S3.2: About the virtual machine load balancing objective function f 2 (x):

[0041] The average load of all virtual machines in the server cluster is the average time AL required for all virtual machines to complete tasks, which is calculated as shown in formula (3):

[0042]

[0043] The calculation method of the virtual machine load balancing standard deviation LSD is shown in formula (4):

[0044]

[0045] The specific process of S6 includes the following steps:

[0046] S6.1: Generate K uniformly distributed vectors in the solution space Ω and convert the solution space R w Divide into K sub-regions Ω ε (ε=1, 2, ..., K), w is the number of objective functions;

[0047] S6.2: Set subpopulation index ε=1;

[0048] S6.3: Individual x in population Q i (i∈{1,2,...,N}) according to its fitness value F i =(f 1 (x), f 2 (x)) belongs to the region in the target space, and the way to assign it to the corresponding subpopulation is as follows: in <F i , v ε > represents vector F i and vector v ε The acute angle of <F i , v l > represents the vector F i and vector v l The acute angle of individual x i F i Vector and uniformly distributed vector v ε When the angle between individual x and i will be assigned to subpopulation Pε middle;

[0049] S6.4: Determine the subpopulation P ε The number of individuals in |P ε |Whether the subpopulation size exceeds the set value S, if |P ε |≤S, then randomly select S-|P from the population Q ε | Individuals join P ε In, if |P ε |≥S, perform non-dominated sorting on the individuals in Pε, delete |P ε |- S individuals with the highest non-dominated ranking;

[0050] The non-dominated sorting calculation method in S6.4 is described as follows:

[0051] 1): Assume that there is an individual x in the population Q, S x is the set of individuals that x can control, n x is the number of individuals that can dominate individual x. For any individual y in Q, if x dominates y (denoted as y>x), then S x :=S x ∪{y}, otherwise, n x =n x +1;

[0052] 2): If n x =0, record x rank =1, store it in the set U 1 middle;

[0053] 3): Set idx = 1, A = Φ, of n q =n q -1; if n q = 0, add q to A and q rank =idx+1;

[0054] 4): Let idx = idx + 1, U idx =A;

[0055] 5): If U idx =Φ, then stop and output the non-dominated rank sorting result; otherwise, go to 3);

[0056] S6.5: If ε≤K, ε=ε+1 and return to S6.4. If ε>K, output subpopulation P 1 , P 2 , ..., P K ;

[0057] The specific process of S7 includes the following steps:

[0058] S7.1: Randomly select P from the current subpopulation α Select an individual y(y≠x β );

[0059] S7.2: Individual x β Perform a two-point crossover operation with individual y according to the crossover probability, and generate two offspring individuals after crossover;

[0060] S7.3: Compare the dominance relationship and crowding distance of the two offspring individuals generated by S7.2, and retain the individual y′ with good quality;

[0061] The dominance relationship of the offspring individuals in 7.3 is: for a minimization problem involving w objectives, the w objective components are f γ (x), γ = {1, 2, ..., w}; for any two feasible solutions x in the decision space a and x b , if it satisfies the following conditions in the target space, then x is called a Dominate x b :

[0062] Condition 1: For γ∈{1,2,...,w}, f γ (x a )≤f γ (x b ) are established;

[0063] Condition 2: γ∈{1,2,...,w}, such that f γ (x a ) <f γ (x b ) is established;

[0064] S7.4: Perform a two-point swap mutation operation on the offspring individual y' according to the mutation probability to generate the offspring individual z;

[0065] S7.5: Add individual z to the temporary set R, R:=R∪{z};

[0066] The two-point crossover operation in S7.2 is described as follows:

[0067] Generate a random number r c ∈(0,1) and the mutation probability P c For comparison, when r c <P c Do the following:

[0068] Randomly select two gene positions on the chromosome. If the two gene positions are the same, regenerate until the two gene positions are different. Define the area between the two gene positions as the crossover domain, and swap the crossover domains of the two parent individuals to generate two offspring individuals.

[0069] The two-point swap mutation in S7.4 is described as follows:

[0070] Generate a random number r m ∈(0,1) and the mutation probability P m For comparison, when r m <P m Execute the following operations:

[0071] Randomly select two gene positions on the chromosome. If the two gene positions are the same, regenerate until the two gene positions are different, and swap the genes at the two gene positions to generate offspring individuals.

[0072] Beneficial technical effects

[0073] The present invention proposes a method for scheduling tasks in a data space system of a manufacturing enterprise based on a decomposition strategy, with the goal of minimizing task completion time and balancing system load, maintaining the stability of the system on the basis of ensuring user satisfaction; decomposing a multi-objective scheduling task into a number of multi-objective sub-problems through a decomposition strategy, communicating information between individuals in the sub-populations, and solving them simultaneously in a cooperative manner, thereby improving the algorithm's solution efficiency; a double-layer chromosome encoding method is used to ensure that the generated solutions are all feasible solutions; in addition, suitable crossover and mutation operators are selected for the algorithm according to the characteristics of the problem, further improving the effectiveness of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A flowchart of a task scheduling algorithm for a manufacturing enterprise data space system based on a decomposition strategy provided by an embodiment of the present invention;

[0075] Figure 2 A schematic diagram of a task scheduling system for a manufacturing enterprise data space system provided by an embodiment of the present invention;

[0076] Figure 3 A flowchart of subpopulation division of a task scheduling algorithm for a manufacturing enterprise data space system based on a decomposition strategy provided by an embodiment of the present invention;

[0077] Figure 4 A schematic diagram of double-layer chromosome encoding and decoding provided by an embodiment of the present invention;

[0078] Figure 5 A schematic diagram of the cross-operation of the task coding layer POX provided in an embodiment of the present invention;

[0079] Figure 6 A schematic diagram of a two-point crossover operation of a machine coding layer provided by an embodiment of the present invention;

[0080] Figure 7A schematic diagram of a task coding layer swap variation operation provided by an embodiment of the present invention;

[0081] Figure 8 A schematic diagram of a multi-point mutation operation of a machine coding layer provided by an embodiment of the present invention;

[0082] Fig. 9 A schematic diagram of task scheduling provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0083] The specific embodiments of the present invention are described in further detail below in conjunction with the accompanying drawings;

[0084] As attached Figure 1 As shown, the task scheduling method of the manufacturing enterprise data space system based on the decomposition strategy includes the following steps:

[0085] S1: Preprocess the required data according to task requirements, including the following steps;

[0086] S1.1: Select the optimization model for task scheduling problem according to the requirements;

[0087] The task scheduling problem in the data space system of a manufacturing enterprise refers to treating multiple tasks uploaded by all users as a total task and assigning them to the enterprise data space system for task scheduling. According to the scheduling results, the tasks are assigned to n virtual machines for processing, so that the task allocation results meet the multiple scheduling goals of the system task decision system, such as Figure 3 As shown in the figure, the overall task contains m subtasks, and the subtasks are independent of each other; the delay and loss in the data transmission process are ignored, all computing resources in the system are mapped to virtual machines, and the number of subtasks is greater than the number of virtual machines; the schematic diagram of the task scheduling system of the manufacturing enterprise data space system is shown in the attached figure. Figure 2 As shown;

[0088] S1.2: Select the task scheduling target according to the demand;

[0089] Assign m tasks to n virtual machines, and make the total completion time of m tasks as short as possible and balance the load of virtual machines;

[0090] S1.3: Select individual coding method according to needs;

[0091] Chromosome encoding adopts integer encoding. Each chromosome represents a scheduling scheme, reflecting the mapping relationship between task scheduling and resources. The length of the chromosome is equal to the number of tasks m. The gene bit number on the chromosome represents the task number, and the value range is {1,2,...,m}. The gene value on the gene bit represents the virtual machine number assigned to the task, and the value range is {1,2,...,n}.

[0092] In this embodiment, there are 3 users who submit 3 tasks (including 7 subtasks m=7) to the enterprise data space system including 3 virtual machines (n=3), and the detailed description is shown in Table 1;

[0093] Table 1

[0094]

[0095] For the data in Table 1, if there is a chromosome coded as (1,3,2,2,1,3,1), the corresponding relationship between tasks and virtual machines is:

[0096] M 1 :T 11 ,T 22 ,T 32 ;

[0097] M 2 :T 13 ,T 21 ;

[0098] M 3 :T 12 ,T 31 ;

[0099] Chromosome encoding and decoding diagram and task T 1 The decoding process is shown in the attached Figure 4 As shown;

[0100] S2: Set the maximum number of iterations termination criterion MaxGen, set the parameter values; the parameters include population size (N), number of sub-populations (K), sub-population size (S), crossover probability (p c ), mutation probability (p m ), where N = K × S;

[0101] Table 2 shows the algorithm parameter selection;

[0102] Table 2

[0103]

[0104] S3: Randomly initialize the first generation of initial population Q = (x 1 ,x 2 ,...,x N ), calculate each individual x in population Q i The fitness value F of (i=1,2,..., N) i =(f 1 (x),f 2 (x)), where the first objective function f 1 (x) is the total completion time of the task, and the second objective function f 2(x) is the virtual machine load balancing index;

[0105] S3.1: The total completion time objective function f of the task 1 (x):

[0106] When there are s tasks assigned to the jth virtual machine M j When the virtual machine M j The time required to complete the assigned task is shown in formula (1):

[0107]

[0108] Among them, t ij For task T i In the virtual machine M j The time required for processing is small and the tasks are independent of each other. When multiple tasks are assigned to a virtual machine, they are executed according to the first-in-first-out principle.

[0109] After calculating the task completion time on each virtual machine, the total task completion time C max As shown in formula (2):

[0110] f 1 (x) = C max =max{C j},j∈{1,2,...,n} (2)

[0111] Among them C j Represents the virtual machine M j The time required to complete all tasks, n is the number of virtual machines;

[0112] S3.2: About virtual machine load balancing objective function f 2 (x) can be obtained by the following process:

[0113] The average load of all virtual machines in the server cluster, that is, the average time AL required for all virtual machines to complete tasks, is calculated as shown in formula (3):

[0114]

[0115] The calculation of the virtual machine load balancing standard deviation LSD is shown in formula (4):

[0116]

[0117] S4: Set the current number of iterations gen = 1;

[0118] S5: Set the temporary set R = Φ, α = 1;

[0119] S6: Assign each individual in population Q to the corresponding subpopulation P according to its fitness value1 , P 2 , ..., P K The steps include:

[0120] S6.1: Generate K uniformly distributed vectors in the solution space Ω and convert the solution space R w Divide into K sub-regions Ω ε (ε=1, 2, ..., K), w is the number of objective functions;

[0121] S6.2: Set subpopulation index ε=1;

[0122] S6.3: Individual x in population Q i (i∈{1,2,...,N}) according to its fitness value F i =(f 1 (x), f 2 (x)) belongs to the region in the target space, and the way to assign it to the corresponding subpopulation is as follows: in <F i , v ε > represents the vector F i and vector v ε The acute angle of <F i , v l > represents the vector F i and vector v l The acute angle of individual x i F i Vector and uniformly distributed vector v ε When the angle between individual x and i will be assigned to subpopulation P ε middle;

[0123] S6.4: Determine the subpopulation P ε The number of individuals in |P ε |Whether the subpopulation size exceeds the set value S, if |P ε |≤S, then randomly select S-|P from the population Q ε | Individuals join P ε In, if |P ε |≥S, for P ε Perform non-dominated sorting on the individuals in the table, and delete |P ε |- S individuals with the highest non-dominated ranking;

[0124] The non-dominated sorting calculation process in S6.4 is as follows:

[0125] 1): Assume that there is an individual x in the population Q, S x is the set of individuals that x can control, n xis the number of individuals that can dominate individual x. For any individual y in Q, if x dominates y (denoted as y>x), then S x :=S x ∪{y}, otherwise, n x =n x +1;

[0126] 2): If n x =0, record x rank =1, store it in the set U 1 middle;

[0127] 3): Set idx = 1, A = Φ, of n q =n q -1; if n q = 0, add q to A and q rank =idx+1;

[0128] 4): Let idx = idx + 1, U idx =A;

[0129] 5): If U idx =Φ, then stop and output the non-dominated rank sorting result; otherwise, go to 3);

[0130] S6.5: If ε≤K, ε=ε+1 and return to S6.4. If ε>K, output subpopulation P 1 , P 2 , ..., P K ;

[0131] The specific process of S7 includes the following steps:

[0132] S7: Pair population P α For each individual x β (β=1, 2, ..., N) performs crossover mutation operation to generate offspring individuals, and stores the offspring individuals in a temporary set R, and performs the following steps in sequence:

[0133] S7.1: Randomly select P from the current subpopulation α Select an individual y(y≠x β );

[0134] S7.2: Individual x β Perform a two-point crossover operation with individual y according to the crossover probability, and generate two offspring individuals after crossover; The crossover probability and method are: Generate a random number r c ∈(0,1) and the mutation probability P c For comparison, when r c <P c Perform the following operations:

[0135] Randomly select two gene positions on the chromosome. If the two gene positions are the same, regenerate until the two gene positions are different. Define the area between the two gene positions as the crossover domain, and swap the crossover domains of the two parent individuals to generate two offspring individuals. The schematic diagram of the two-point crossover is shown in the attached figure. Figure 5 As shown;

[0136] S7.3: Compare the dominance relationship and crowding distance of two offspring individuals and retain the individual y' with better quality;

[0137] The dominance relationship of the offspring individuals in S7.3 is described as follows:

[0138] For a minimization problem with w objectives, the w objective components are f γ (x), γ={1,2,...,w}; for any two feasible solutions x in the decision space a and x b , if it satisfies the following conditions in the target space, then x is called a Dominate x b :

[0139] Condition 1: For γ∈{1,2,...,w}, f γ (x a )≤f γ (x b ) are established;

[0140] Condition 2: γ∈{1,2,...,w}, such that f γ (x a ) <f γ (x b ) is established;

[0141] The crowding distance calculation method in S7.3 is described as follows:

[0142] Sort the individuals in ascending order on different objective functions, and use this to calculate the crowding distance of the individuals on the objective; Individual x i Crowding distance on the γth objective function where γ∈{1, 2,...,w}, and Represents individual x i+1 and individual x i-1 The target value at the γth target; and represent the maximum and minimum values ​​of the γth objective function respectively; in addition, the crowding distances of the first and last individuals in the objective value ranking are positive infinity; the crowding distances on the w objectives are combined and used as the individual x i The crowding distance di , which is calculated as: Used to describe the density of individuals in the target space compared to other individuals;

[0143] S7.4: Perform a two-point swap mutation operation on the offspring individual y' according to the mutation probability to generate the offspring individual z;

[0144] The mutation probability and method are: Generate a random number r m ∈(0,1) and the mutation probability P m For comparison, when r <P m Do the following:

[0145] Randomly select two gene positions on the chromosome. If the two gene positions are the same, regenerate until the two gene positions are different, and swap the genes at the two gene positions to generate offspring individuals. The schematic diagram of the two-point swap mutation is shown in the attached figure. Figure 6 As shown;

[0146] S7.5: Add individual z to the temporary set R, R:=R∪{z};

[0147] S8: Merge the current population Q with the temporary set R, Q:=R∪Q;

[0148] S9: Set α=α+1. If α≤K, return to S7. If α>K, execute the next step.

[0149] S10: set gen=gen+1;

[0150] S11: According to the current number of iterations of the algorithm, determine whether the algorithm meets the termination condition. If the termination condition is not met, repeat S5 to S10. If the condition is met, output all non-dominated solutions in the population Q. At this time, each non-dominated solution corresponds to a task scheduling scheme, and the decision maker can select a scheme to execute according to his preference or demand;

[0151] The termination condition of the algorithm in S11 is: the current iteration number gen of the algorithm is equal to the preset maximum iteration number MaxGen; Simulation experiment:

[0152] The simulation experiment was carried out using the CloudSim 4.0 simulation platform. The processing rate of the constructed virtual machine was 1000 MIPS, where MIPS means the number of million instructions processed per second, which is the unit of measurement for the processing resource capacity of the virtual machine in the CloudSim simulation platform. By setting n (n = 10, 20, 40) virtual machines with the same configuration and m (m = 100, 300, 500) tasks with a task length range of 1000-10000 MI (million instructions), 9 simulation examples of different scales were constructed as shown in Table 3. Under the given experimental conditions, the algorithm (MOEA / D-CC) proposed in the present invention and two comparative multi-objective algorithms (SPEA-II and NSGA-III) were used to carry out simulation experiments in 9 examples of different scales to verify the effectiveness of the algorithm proposed in the present invention. The stopping criterion of each algorithm is that the number of objective function evaluations reaches 50,000 times. The population size of the MOEA / D-CC and NSGA-III algorithms is set to 100, and the population size of the SPEA-II algorithm is set to 40. The crossover probability P c =0.8, mutation probability P m =0.1; To avoid the contingency of the experimental results, each algorithm was independently run 10 times under each combination of the number of virtual machines and the number of tasks, and the experimental results were recorded for result analysis;

[0153] Table 3

[0154] Case No. Number of virtual machines n Number of tasks m Mission Length 1 10 100 1000-10000 2 10 300 1000-10000 3 10 500 1000-10000 4 20 100 1000-10000 5 20 300 1000-10000 6 20 500 1000-10000 7 40 100 1000-10000 8 40 300 1000-10000 9 40 500 1000-10000

[0155] In each case, the total task completion time (f 1 ) and the standard deviation of virtual machine load balancing (f 2 ), and then draw a comparison chart of the task completion time and the standard deviation of the virtual machine load balance of different algorithms under different scale examples, as shown in the attached figure. Figure 7 and attached Figure 8 As shown; from the attached Figure 7 It can be seen that in all cases, MOEA / D-CC can obtain the shortest total task completion time, and as the number of tasks increases, the scheduling advantage of MOEA / D-CC becomes more obvious; Figure 8 It can be seen that under different scale examples, the standard deviation of the virtual machine load balance of the scheduling scheme obtained by the MOEA / D-CC algorithm fluctuates less, and as the number of virtual machines increases, the scheduling advantage of MOEA / D-CC becomes more obvious;

[0156] To further demonstrate the scheduling results, a task scheduling diagram is drawn when there are 10 virtual machines and 100 tasks, as shown in the attached figure. Fig. 9 As shown; from the attached Fig. 9 It can be seen that the calculation completion time of the 10 virtual machines is basically the same, which avoids the waste of computing resources and improves the computing efficiency of the task;

[0157] The above experiments prove the effectiveness of the task scheduling method for the manufacturing enterprise data space system based on the decomposition strategy proposed in the present invention, which can significantly shorten the total execution time of the task and ensure the load balancing of the virtual machine, and has positive significance for improving the performance and stability of the manufacturing enterprise data space system.

Claims

1. A method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy, characterized in that: The following steps are involved: S1: Preprocess the required data according to task requirements; The specific process of S1 includes the following steps: S1.1: Select the optimization model for task scheduling problem according to the requirements; All tasks uploaded by all users are considered as the total task and handed over to the enterprise data space system for task scheduling. According to the scheduling results, the tasks are assigned to n virtual machines for processing, so that the task allocation results meet the multiple scheduling goals of the system task decision system. The overall task contains m subtasks, and the subtasks are independent of each other. Ignoring the delay and loss in the data transmission process, all computing resources in the system are mapped to virtual machines, and the number of subtasks is greater than the number of virtual machines. S1.2: Select the task scheduling target according to the demand; Assign m tasks to n virtual machines so that the total completion time of m tasks is the shortest and the load of virtual machines is balanced; S1.3: Select individual coding method according to needs; Chromosome encoding adopts integer encoding. Each chromosome represents a scheduling scheme, reflecting the mapping relationship between task scheduling and resources. The length of the chromosome is equal to the number of tasks m. The gene bit number on the chromosome represents the task number, and the value range is {1,2,...,m}. The gene value on the gene bit represents the virtual machine number assigned to the task, and the value range is {1,2,...,n}. S2: Set the maximum number of iterations termination criterion MaxGen and set the parameter value; S3: Randomly initialize the first generation of initial population Q = (x 1 ,x 2 ,...,x N ); Calculate the value of each individual x in population Q i The fitness value F of (i=1,2,...,N) i =(f1(x), f2(x)), where the first objective function f1(x) is the total completion time of the task, and the second objective function f2(x) is the virtual machine load balancing index; S4: Set the current number of iterations gen = 1; S5: Set the temporary set R = Φ, α = 1; S6: According to the fitness value of each individual in population Q, it is assigned to the corresponding sub-population P1, P2, ..., P K middle; The specific process of S6 includes the following steps: S6.1: Generate K uniformly distributed vectors in the solution space Ω and convert the solution space R w Divide into K sub-regions Ω ε (ε=1,2,...,K), w is the number of objective functions; S6.2: Set subpopulation index ε=1; S6.3: Individual x in population Q i (i∈{1,2,...,N}) according to its fitness value F i =(f1(x),f2(x)) belongs to the region in the target space, and the way to assign it to the corresponding subpopulation is as follows in <F i ,v ε > represents the vector F i and vector v ε The acute angle of <F i ,v l > represents the vector F i and vector v l The acute angle of individual x i F i Vector and uniformly distributed vector v ε When the angle between individual x and i will be assigned to subpopulation P ε middle; S6.4: Determine the subpopulation P ε The number of individuals in |P ε |Whether the subpopulation size exceeds the set value S, if |P ε |≤S, then randomly select S-|P from the population Q ε | Individuals join P ε In, if |P ε |≥S, for P ε Perform non-dominated sorting on the individuals in the table, and delete |P ε |-S individuals with the highest non-dominated ranking; S6.5: If ε≤K, ε=ε+1 and return to S6.

4. If ε>K, output subpopulations P1, P2, ..., P K ; S7: Pair population P α For each individual x β (β=1,2,...,N) Perform crossover mutation operation to generate offspring individuals and store the offspring individuals in a temporary set R; S8: Merge the current population Q with the temporary set R, Q:=R∪Q; S9: Set α=α+1. If α≤K, return to S7. If α>K, execute the next step. S10: set gen=gen+1; S11: According to the current number of iterations of the algorithm, determine whether the algorithm meets the termination condition MaxGen. If the termination condition is not met, repeat S5 to S10. If the condition is met, output all non-dominated solutions in the population Q. At this time, each non-dominated solution corresponds to a task scheduling plan, and the decision maker can choose a plan to execute according to his preference or demand.

2. The method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy as claimed in claim 1, characterized in that: The parameters in S2 include: population size N, number of subpopulations K, subpopulation size S, crossover probability p c , mutation probability p m , where N = K × S.

3. The method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy as claimed in claim 1, characterized in that: The specific process of S3 is as follows: Evaluate each individual x in population Q i (i=1,2,...,N); the individual fitness value evaluation includes the first objective function f1(x) which is the total completion time of the task and the second objective function f2(x) which is the load balancing index of the virtual machine. The fitness value function is as follows: S3.1: Fitness value function f1(x) about the total completion time of the task: When there are s tasks assigned to the jth virtual machine M j When the virtual machine M j The time required to complete the assigned tasks Among them, t ij For task T i In the virtual machine M j The time required for processing is independent of each other. After obtaining the task completion time on each virtual machine, the total task completion time C max for: f1(x)=C max =max{C j },j∈{1,2,...,n} (2) Among them, C j Represents the virtual machine M j The time required to complete all tasks, n is the number of virtual machines; Task allocation in S3.1: when multiple tasks are allocated to a virtual machine, the tasks are executed according to the first-in-first-out principle; S3.2: About the virtual machine load balancing objective function f2(x): The average load of all virtual machines in the server cluster is the average time AL required for all virtual machines to complete tasks, which is calculated as shown in formula (3): The calculation method of the virtual machine load balancing standard deviation LSD is shown in formula (4):

4. The method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy as claimed in claim 1, characterized in that: The non-dominated sorting calculation method in S6.4 is described as follows: 1): Assume that there is an individual x in the population Q, S x is the set of individuals that x can control, n x is the number of individuals that can dominate individual x. For any individual y in Q, if x dominates y, it is recorded as Then S x :=S x ∪{y}, otherwise, n x =n x +1; 2): If n x =0, record x rank =1, store it in set U1; 3): Set idx = 1, A = Φ, of n q =n q -1; if n q = 0, add q to A and q rank =idx+1; 4):Let idx=idx+1,U idx =A; 5): If U idx =Φ, then stop and output the non-dominated rank sorting result; otherwise, go to 3).

5. The method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy as claimed in claim 1, characterized in that: The specific process of S7 includes the following steps: S7.1: Randomly select P from the current subpopulation α Select an individual y(y≠x β ); S7.2: Individual x β Perform a two-point crossover operation with individual y according to the crossover probability, and generate two offspring individuals after crossover; Generate a random number r c ∈(0,1) and the mutation probability P c For comparison, when r c <P c Do the following: Randomly select two gene positions on the chromosome. If the two gene positions are the same, regenerate until the two gene positions are different. Define the area between the two gene positions as the crossover domain, and swap the crossover domains of the two parent individuals to generate two offspring individuals. S7.3: Compare the dominance relationship and crowding distance of the two offspring individuals generated by S7.2, and retain the individual y' with good quality; S7.4: Perform a two-point swap mutation operation on the offspring individual y' according to the mutation probability to generate the offspring individual z; S7.5: Add individual z to the temporary set R, R:=R∪{z}.

6. A method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy as claimed in claim 5, characterized in that: The dominance relationship of the offspring individuals in 7.3 is: for a minimization problem involving w objectives, the w objective components are f γ (x), γ={1,2,...,w}; for any two feasible solutions x in the decision space a and x b , if it satisfies the following conditions in the target space, then x is called a dominatex b : Condition 1: For γ∈{1,2,...,w}, f γ (x a )≤f γ (x b ) are established; Condition 2: γ∈{1,2,...,w}, such that f γ (x a ) <f γ (x b ) was established.

7. The method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy as claimed in claim 5, characterized in that: The crowding distance calculation process in S7.3 is as follows: Sort the individuals in ascending order on different objective functions, and use this to calculate the crowding distance of the individuals on the objective; individual x i Crowding distance on the γth objective function where γ∈{1,2,...,w}, and Represents individual x i+1 and individual x i-1 The target value at the γth target; and represent the maximum and minimum values ​​of the γth objective function respectively; in addition, the crowding distances of the first and last individuals in the objective value ranking are positive infinity; the crowding distances on the w objectives are combined and used as the individual x i The crowding distance d i , which is calculated as: Used to describe the density of an individual in the target space compared to other individuals.

8. The method for scheduling tasks in a manufacturing enterprise data space system based on a decomposition strategy as claimed in claim 5, characterized in that: The two-point swap mutation in S7.4 is described as follows: Generate a random number r m ∈(0,1) and the mutation probability P m For comparison, when r m <P m Execute the following operations: Randomly select two gene positions on the chromosome. If the two gene positions are the same, regenerate until the two gene positions are different, and swap the genes at the two gene positions to generate offspring individuals.