CN114371915A - Manufacturing enterprise data space system task scheduling method based on decomposition strategy - Google Patents

Abstract

The invention discloses a manufacturing enterprise data space system task scheduling method based on a decomposition strategy, and relates to the technical field of manufacturing enterprise data space; from a multi-target view, the optimized allocation of system computing resources can be realized while the task processing time is minimized, the task completion time is minimized, the system load is balanced, and the stability of the system is maintained on the basis of ensuring the satisfaction degree of users; the multi-target scheduling task is decomposed into a plurality of multi-target subproblems through a decomposition strategy, information among the subproblems is exchanged, and a cooperation mode is adopted to solve simultaneously, so that the solving efficiency of the algorithm is improved; by adopting a double-layer chromosome coding mode, the generated solutions are guaranteed to be feasible solutions, short boards which need to set weights for different targets depending on manual experience in a single-target algorithm are effectively avoided, the performance of a data space system is improved, and the stable operation of the system is guaranteed.

Description

Manufacturing enterprise data space system task scheduling method based on decomposition strategy

Technical Field

The invention relates to the technical field of manufacturing enterprise data space, in particular to a manufacturing enterprise data space system task scheduling method based on a decomposition strategy.

Background

With the rapid development of computer technology, industrial internet and internet of things technology, the construction process of a data space system of a manufacturing enterprise is further promoted; the manufacturing enterprise data space system is a novel system mode developed on the basis of distributed computing, parallel computing and grid computing, and can effectively integrate computing resources, service resources and storage resources through the data space system and carry out uniform allocation and management on the resources according to user requirements; with the increasing number of users and service requests, how to efficiently and reasonably distribute and utilize limited system resources and effectively schedule massive tasks submitted by the users is a key technical problem in the construction process of a data space system of a manufacturing enterprise, and the improvement of the user satisfaction to the greatest extent; the process of completing a normal user service in the data space system can be described as follows: the system distributes tasks of different users to appropriate computing resources (virtual machines) through a task scheduler to be executed, the task scheduler needs to solve two problems of system computing resource distribution and task sequencing in the process, and the problem belongs to a typical NP complete problem, namely the problem can be solved without an optimization algorithm of polynomial time;

currently, aiming at a data space of a manufacturing enterprise, a plurality of task scheduling methods still use traditional heuristic or hyper-heuristic algorithms, such as a genetic algorithm, a Max-Min algorithm, a Min-Min algorithm and the like, and the algorithms generally have the defects of easy trapping of local optimization, weak global search capability and the like in the optimization and solution process; in the aspect of objective function setting, most task scheduling methods only pay attention to how to shorten the processing time of tasks during optimization, but neglect to keep the load of the system balanced on the basis of meeting the user requirements so as to ensure the stable operation of the system; in addition, most of the existing methods adopt a linear weighting mode to simplify the original multi-target problem into a single-target optimization problem when processing the condition of a plurality of target functions, and then use the traditional single-target optimization algorithm to solve the problem instead of adopting the multi-target optimization algorithm to solve the problem based on a Pareto dominant view angle; because the weight setting of each objective function is usually based on manual experience, reasonable configuration of the objective weight is difficult to realize, a globally optimal resource allocation and task sequencing scheme is difficult to obtain, and effective algorithm support cannot be provided for performance improvement of a data space system of a manufacturing enterprise;

the patent "cloud computing task scheduling method based on improved genetic algorithm" (ZL202011086788.4) and the patent "genetic and differential mixed evolution cloud computing task scheduling algorithm based on early catastrophe strategy" (ZL201911082199.6) both propose cloud computing task scheduling methods based on improved evolution algorithm, however, these scheduling methods all aim at optimizing task processing time in the cloud computing task scheduling process by adopting single-target optimization algorithm, a decision maker cannot simultaneously consider the influence of different target functions on scheduling when selecting a scheduling scheme, and it is difficult to provide effective algorithm support for performance improvement of a data space system of a manufacturing enterprise.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides a manufacturing enterprise data space system task scheduling method based on a decomposition strategy, which can realize optimized distribution of system computing resources while minimizing task processing time from a multi-target perspective, effectively avoids short boards needing to set weights for different targets depending on manual experience in a single-target algorithm, improves the performance of a data space system and ensures the stable operation of the system;

the technical scheme adopted by the invention is as follows:

a manufacturing enterprise data space system task scheduling method based on a decomposition strategy comprises the following steps:

s1: preprocessing required data according to task requirements;

s2: setting a maximum iteration time termination criterion MaxGen, and setting the value of a parameter;

s3: randomly initializing a first generation initial population Q ═ x¹,x²,...,x^N)；

Calculating each individual x in the population QⁱA fitness value F of (i ═ 1, 2.., N)ⁱ＝(f₁(x)，f₂(x) Wherein, firstA target function f₁(x) For the total completion time of the task, a second objective function f₂(x) The index is the load balance index of the virtual machine;

s4: setting the current iteration number gen as 1;

s5: setting a temporary set R & ltphi & gt and alpha & lt1 & gt;

s6: distributing each individual in the population Q to a corresponding sub-population P according to the fitness value of each individual in the population Q₁,P₂,...,P_KPerforming the following steps;

s7: pair sub-population P_αEach of the individuals x in^β(β ═ 1, 2.., N) performing a cross mutation operation to generate offspring individuals, and storing the offspring individuals into the temporary set R;

s8: combining the current population Q with the temporary set R, wherein Q ═ R ═ U.Q;

s9: setting alpha to alpha +1, if alpha is less than or equal to K, returning to S7, and if alpha is more than K, executing the next step;

s10: setting gen as gen + 1;

s11: judging whether the algorithm meets a termination condition MaxGen according to the current iteration times of the algorithm, if the algorithm does not meet a stop condition, repeating S5-S10, if all non-dominant solutions in the output population Q meet the condition, wherein each non-dominant solution corresponds to a task scheduling scheme, and a decision maker can select one scheme from the schemes to execute according to the preference or the requirement of the decision maker;

the specific process of S1 includes the following steps:

s1.1: selecting an optimization model of a task scheduling problem according to requirements;

s1.2: selecting a task scheduling target according to requirements;

s1.3: selecting an individual coding mode according to requirements;

the task scheduling problem of S1.1 is specifically described or optimized as follows:

taking a plurality of tasks uploaded by all users as a total task, delivering the total task to an enterprise data space system for task scheduling, and distributing the tasks to n virtual machines for processing according to a scheduling result so as to enable the task distribution result to meet a plurality of scheduling targets of a system task decision system, wherein the total task comprises m subtasks which are independent from each other; neglecting delay and loss in the data transmission process, mapping all computing resources in the system into virtual machines, wherein the number of the subtasks is more than that of the virtual machines;

the task scheduling goals of S1.2 are as follows:

distributing the m tasks to the n virtual machines to enable the total completion time of the m tasks to be shortest and the loads of the virtual machines to be balanced;

the chromosome coding mode of S1.3 is as follows:

the chromosome coding adopts an integer coding mode, each chromosome represents a scheduling scheme and reflects the mapping relation between task scheduling arrangement and resources, and the length of the chromosome is equal to the number m of tasks; the gene position number on the chromosome represents a task number, and the value range is {1, 2.., m }; the gene value on the gene position represents the virtual machine number allocated to the task, and the value range is {1, 2.., n };

in the S2, the parameters: comprises a population scale N, a sub-population number K, a sub-population scale S and a cross probability p_cProbability of variation p_mWherein N ═ K × S;

the specific process of S3 is as follows:

evaluation of Each individual x in the population QⁱA fitness value of (i ═ 1, 2.., N); the evaluation of the fitness value of the individual comprises a first target function f₁(x) For the total completion time of the task and a second objective function f₂(x) For two aspects of the virtual machine load balance index, the adaptation degree value function is specifically as follows:

s3.1: function f of fitness value of total completion time of task₁(x)：

When s tasks are distributed to the jth virtual machine M_jThen, virtual machine M_jTime required to complete the assignment task

Wherein, t_ijFor task T_iIn virtual machine M_jAt the upper partThe required time is managed, the tasks are mutually independent, and the completion time C of the total task after the completion time of the task on each virtual machine is obtained_maxComprises the following steps:

f₁(x)＝C_max＝max{C_j},j∈{1,2,...,n} (2)

wherein, C_jDelegate virtual machine M_jThe time required for completing all tasks, wherein n is the number of virtual machines;

task allocation in S3.1: executing tasks according to a first-in first-out principle when a plurality of tasks are distributed to one virtual machine;

s3.2: load balancing objective function f for virtual machine₂(x)：

The average load capacity of all the virtual machines in the server cluster is the average time AL required by all the virtual machines to complete the task, and the calculation mode is as shown in formula (3):

the calculation mode of the virtual machine balanced load standard deviation LSD is shown in formula (4):

the specific process of S6 includes the following steps:

s6.1: generating K uniformly distributed vectors in a solution space omega, and solving the solution space R^wDivided into K sub-regions omega_ε(e ═ 1,2,. K), w is the number of objective functions;

s6.2: setting a sub-population index epsilon as 1;

s6.3: individuals x in population Qⁱ(i ∈ {1, 2.,. N }) according to its fitness value Fⁱ＝(f₁(x)，f₂(x) In the target space, in such a way that they are assigned to the corresponding sub-populations

Wherein<Fⁱ，v^εRepresents the vector FⁱSum vector v^εThe included angle of the acute angle is smaller than the included angle of the acute angle,<Fⁱ，v^l>representative vector FⁱSum vector v^lAcute included angle of (i.e. individual x)ⁱF of (A)ⁱVector and uniformly distributed vector v^εWhen the included angle of (A) is smallest, the individual xⁱWill be assigned to a sub-population P_εPerforming the following steps;

s6.4: judging the sub-population P_εNumber of individuals | P_εIf | exceeds the sub-population scale set value S, if | P_εIf the value is less than or equal to S, S-P is randomly selected from the population Q_εI Individual addition of P_εIn, if | P_ε| ≧ S, perform non-dominant ordering of individuals in P ε, delete | P_ε-S individuals with the highest non-dominated ranking;

the non-dominant ranking calculation method in S6.4 is described as follows:

1): setting the existence of individuals x, S in the population Q_xFor x assignable sets of individuals, n_xFor any individual y in Q, if x dominates y (denoted y > x), then S_x：＝S_xU { y }, otherwise, n_x＝n_x+1；

2): if n is_xX is 0_rankStore it in the set U as 1₁Performing the following steps;

3): setting idx ═ 1, a ═ Φ, pair

Is/are as follows

n_q＝n_q-1; if n is_qQ is added to a and q is 0_rank＝idx+1；

4): let idx be idx +1, U_idx＝A；

5): if U is_idxIf the result is phi, stopping and outputting a non-dominant grade sequencing result; otherwise, go to 3);

s6.5: if ε is not more than K, ε is ε +1 and S6.4 is returned, if ε is more than K, the output isGroup P₁，P₂，...，P_K；

The specific process of S7 includes the following steps:

s7.1: randomly selecting P from the current sub-population_αSelecting an individual y (y ≠ x)^β)；

S7.2: individual x^βPerforming two-point crossing operation with the individual y according to the crossing probability, and generating two filial generation individuals after crossing;

s7.3: comparing the dominance relation and the crowding distance of the two descendant individuals generated in the S7.2, and keeping the individuals y' with good quality;

the domination relationship of the filial generation individuals in the 7.3 is as follows: for a minimization problem involving w targets, the w target components are f_γ(x) γ ═ 1,2,. ·, w }; for any two feasible solutions x in decision space_aAnd x_bIf it satisfies the following condition in the target space, x is called_aDominating x_b：

Condition 1: for γ ∈ {1,2,..., w }, f ∈ } f_γ(x_a)≤f_γ(x_b) Both are true;

condition 2: γ ∈ {1, 2., w }, such that f_γ(x_a)<f_γ(x_b) If true;

s7.4: performing two-point interchange mutation operation on the child individual y' according to the mutation probability to generate a child individual z;

s7.5: adding an individual z into a temporary set R, wherein R ═ R { Z };

the two-point crossing operation in S7.2 is described as follows:

generating a random number r_cBelongs to (0,1) and is associated with the mutation probability P_cMaking a comparison when r_c<P_cThe following operations are executed:

randomly selecting two gene positions on the chromosome, and if the two gene positions are the same, regenerating until the two gene positions are different; defining the region between the two gene positions as a cross domain, and interchanging the cross domains of the two parent individuals to generate two offspring individuals;

the two-point interchange variation in S7.4 is described as follows:

generating a random number r_mBelongs to (0,1) and is associated with the mutation probability P_mMaking a comparison when r_m<P_mThe following operations are executed:

randomly selecting two gene positions on the chromosome, if the two gene positions are the same, regenerating until the two gene positions are different, and interchanging the genes on the two gene positions to generate a filial generation individual.

Advantageous technical effects

The invention provides a manufacturing enterprise data space system task scheduling method based on a decomposition strategy, which aims at minimizing task completion time and balancing system load, and maintains the stability of a system on the basis of ensuring user satisfaction; the multi-target scheduling task is decomposed into a plurality of multi-target subproblems through a decomposition strategy, information among the subproblems is exchanged, and the solution is carried out simultaneously in a cooperation mode, so that the solution efficiency of the algorithm is improved; a double-layer chromosome coding mode is adopted, so that the generated solutions are all feasible solutions; in addition, proper crossover and mutation operators are selected for the algorithm according to the problem characteristics, and the effectiveness of the algorithm is further improved.

Drawings

FIG. 1 is a flowchart of a manufacturing enterprise data space system task scheduling algorithm based on a decomposition strategy according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a manufacturing enterprise data space system task scheduling system according to an embodiment of the present invention;

FIG. 3 is a flow chart of a sub-population division of a task scheduling algorithm of a manufacturing enterprise data space system based on a decomposition strategy according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of double-layer chromosome encoding and decoding provided by an embodiment of the present invention;

FIG. 5 is a diagram illustrating a cross operation of a task coding layer POX according to an embodiment of the present invention;

FIG. 6 is a schematic diagram illustrating a two-point interleaving operation of a machine code layer according to an embodiment of the present invention;

FIG. 7 is a schematic diagram illustrating task coding layer inter-change mutation operations according to an embodiment of the present invention;

FIG. 8 is a diagram illustrating a multi-point mutation operation of a machine code layer according to an embodiment of the present invention;

fig. 9 is a schematic diagram of task scheduling according to an embodiment of the present invention.

Detailed Description

The following describes in further detail embodiments of the present invention with reference to the accompanying drawings;

as shown in FIG. 1, the method for scheduling tasks of a manufacturing enterprise data space system based on a decomposition strategy comprises the following steps:

s1: preprocessing required data according to task requirements, and comprises the following steps;

s1.1: selecting an optimization model of a task scheduling problem according to requirements;

the task scheduling problem in the data space system of the manufacturing enterprise means that a plurality of tasks uploaded by all users are regarded as a total task and are delivered to the data space system of the enterprise for task scheduling, and the tasks are distributed to n virtual machines for processing according to a scheduling result, so that the task distribution result meets a plurality of scheduling targets of a system task decision system, as shown in fig. 3, wherein the total task comprises m subtasks which are independent from each other; neglecting delay and loss in the data transmission process, mapping all computing resources in the system into virtual machines, wherein the number of the subtasks is more than that of the virtual machines; a schematic diagram of a manufacturing enterprise data space system task scheduling system is shown in fig. 2;

s1.2: selecting a task scheduling target according to requirements;

distributing the m tasks to the n virtual machines, and enabling the total completion time of the m tasks to be shortest and the loads of the virtual machines to be balanced;

s1.3: selecting an individual coding mode according to requirements;

the chromosome coding adopts an integer coding mode, each chromosome represents a scheduling scheme and reflects the mapping relation between task scheduling arrangement and resources, and the length of the chromosome is equal to the number m of tasks; the genetic locus number on the chromosome represents a task number, and the value range is {1, 2.., m }; the gene value on the gene position represents the virtual machine number allocated to the task, and the value range is {1, 2.., n };

in this embodiment, 3 existing users submit 3 tasks (including 7 subtasks m ═ 7) to an enterprise data space system including 3 virtual machines (n ═ 3), and the detailed description is shown in table 1;

TABLE 1

For the data in table 1, if there is a chromosome coded as (1,3,2,2,1,3,1), the corresponding relationship between the task and the virtual machine is:

M₁：T₁₁,T₂₂,T₃₂；

M₂：T₁₃,T₂₁；

M₃：T₁₂,T₃₁；

chromosome coding and decoding schematic diagram and task T₁The decoding process of (2) is shown in fig. 4;

s2: setting a maximum iteration time termination criterion MaxGen, and setting the value of a parameter; the parameters include population size (N), number of sub-populations (K), sub-population size (S), and crossover probability (p)_c) Probability of variation (p)_m) Wherein N ═ K × S;

table 2 algorithm parameter selection;

TABLE 2

S3: randomly initializing a first generation initial population Q ═ x¹,x²,...,x^N) Calculating each individual x in the population QⁱA fitness value F of (i ═ 1, 2.., N)ⁱ＝(f₁(x),f₂(x) In which the first objective function f) is₁(x) For the total completion time of the task, a second objective function f₂(x) The index is the load balance index of the virtual machine;

s3.1: total completion time objective function f for a task₁(x)：

When s tasks are distributed to the jth virtual machine M_jThen, virtual machine M_jThe time required to complete the assignment task is given by equation (1):

wherein, t_ijFor task T_iIn virtual machine M_jThe time required by the upper processing is independent, and when a plurality of tasks are distributed to one virtual machine, the tasks are executed according to the first-in first-out principle;

calculating the completion time C of the total task after the completion time of the task on each virtual machine_maxAs shown in formula (2):

f₁(x)＝C_max＝max{C_j},j∈{1,2,...,n} (2)

wherein C is_jDelegate virtual machine M_jThe time required for completing all tasks, wherein n is the number of virtual machines;

s3.2: load balancing objective function f for virtual machine₂(x) The method can be obtained by the following steps:

the average load capacity of all the virtual machines in the server cluster, that is, the average time AL required for all the virtual machines to complete the task, is calculated as shown in equation (3):

the calculation of the virtual machine balanced load standard deviation LSD is shown as the formula (4):

s4: setting the current iteration number gen as 1;

s5: setting a temporary set R & ltphi & gt and alpha & lt1 & gt;

s6: distributing each individual in the population Q to a corresponding sub-population P according to the fitness value of each individual in the population Q₁，P₂，...，P_KComprises the following steps:

s6.1: generating K uniformly distributed vectors in a solution space omega, and solving the solution space R^wDivided into K sub-regions omega_ε(e ═ 1,2,. K), w is the number of objective functions;

s6.2: setting a sub-population index epsilon as 1;

s6.3: individuals x in population Qⁱ(i ∈ {1, 2.,. N }) according to its fitness value Fⁱ＝(f₁(x)，f₂(x) In the target space, in such a way that they are assigned to the corresponding sub-populations

Wherein<Fⁱ，v^ε>Representative vector FⁱSum vector v^εThe included angle of the acute angle is smaller than the included angle of the acute angle,<Fⁱ，v^l>representative vector FⁱSum vector v^lAcute included angle of (i.e. individual x)ⁱF of (A)ⁱVector and uniformly distributed vector v^εWhen the included angle of (A) is smallest, the individual xⁱWill be assigned to a sub-population P_εPerforming the following steps;

s6.4: judging the sub-population P_εNumber of individuals | P_εIf | exceeds the sub-population scale set value S, if | P_εIf the value is less than or equal to S, S-P is randomly selected from the population Q_εI Individual addition of P_εIn, if | P_ε| is not less than S, for P_εThe individuals in (1) perform non-dominant sorting, deleting | P_ε-S individuals with the highest non-dominated ranking;

the non-dominant ranking calculation process in S6.4 is as follows:

1): setting the existence of individuals x, S in the population Q_xFor x assignable sets of individuals, n_xFor any individual y in Q, if x dominates y (denoted y > x), then S_x：＝S_xU { y }, otherwise, n_x＝n_x+1；

2): if n is_xX is 0_rankStore it in the set U as 1₁Performing the following steps;

3): setting idx ═ 1, a ═ Φ, pair

Is/are as follows

n_q＝n_q-1; if n is_qQ is added to a and q is 0_rank＝idx+1；

4): let idx be idx +1, U_idx＝A；

5): if U is_idxIf the result is phi, stopping and outputting a non-dominant grade sequencing result; otherwise, go to 3);

s6.5: if epsilon is less than K, epsilon +1 and return to S6.4, if epsilon > K, the sub-population P is output₁，P₂，...，P_K；

The specific process of S7 includes the following steps:

s7: pair sub-population P_αEach of the individuals x in^β(β ═ 1, 2., N) performing a cross mutation operation to generate offspring individuals, and storing the offspring individuals in the temporary set R, sequentially performing the following steps:

s7.1: randomly selecting P from the current sub-population_αSelecting an individual y (y ≠ x)^β)；

S7.2: individual x^βPerforming two-point crossing operation with the individual y according to the crossing probability, and generating two filial generation individuals after crossing; the cross probability and the method are as follows: generating a random number r_cBelongs to (0,1) and is associated with the mutation probability P_cMaking a comparison when r_c<P_cThe following operations are executed:

randomly selecting two gene positions on the chromosome, and if the two gene positions are the same, regenerating until the two gene positions are different; defining the region between the two gene positions as a cross domain, and interchanging the cross domains of the two parent individuals to generate two offspring individuals; the two-dot intersection is shown in FIG. 5;

s7.3: comparing the domination relationship and the crowding distance of the two descendant individuals, and keeping the individual y' with better quality;

the dominance relationship of the offspring individuals in S7.3 is described as follows:

for a minimization problem involving w targets, the w target components are f_γ(x) γ ═ 1,2,. ·, w }; for any two feasible solutions x in decision space_aAnd x_bIf it satisfies the following condition in the target space, it is called x_aDominating x_b：

Condition 1: for γ ∈ {1,2,..., w }, f ∈ } f_γ(x_a)≤f_γ(x_b) Both are true;

condition 2: γ ∈ {1, 2., w }, such that f_γ(x_a)<f_γ(x_b) If true;

the congestion distance calculation method in S7.3 is described as follows:

respectively sequencing the individuals on different target functions in an ascending order, and calculating the crowding distance of the individuals on the target according to the sequencing; individual xⁱCrowding distance on the gamma objective function

Where γ ∈ {1,2,.., w },

and

respectively represent an individual xⁱ⁺¹And individual x^i-1A target value on the gamma target;

and

respectively representing the maximum value and the minimum value of the gamma-th objective function; further, the crowd distance of the individuals at the first and last in the target value ranking is positive infinity; synthesizing crowding distances on w targets, and taking the crowding distances as an individual xⁱDistance d of degree of crowding_iThe calculation method is as follows:

used for describing the density of the individual in the target space and other individuals;

s7.4: performing two-point interchange mutation operation on the child individual y' according to the mutation probability to generate a child individual z;

the mutation probability and the method are as follows: generating a random number r_mBelongs to (0,1) and is associated with the mutation probability P_mMaking a comparison when r<P_mThe following operations are executed:

randomly selecting two gene positions on the chromosome, if the two gene positions are the same, regenerating until the two gene positions are different, and interchanging the genes on the two gene positions to generate a filial generation individual, wherein the schematic diagram of the two-point interchange variation is shown in figure 6;

s7.5: adding an individual z into a temporary set R, wherein R ═ R { Z };

s8: combining the current population Q with the temporary set R, wherein Q ═ R ═ U.Q;

s9: setting alpha to alpha +1, if alpha is less than or equal to K, returning to S7, and if alpha is more than K, executing the next step;

s10: setting gen as gen + 1;

s11: judging whether the algorithm meets a termination condition or not according to the current iteration times of the algorithm, if the algorithm does not meet a stop condition, repeating the steps from S5 to S10, if the algorithm meets the condition, outputting all non-dominated solutions in the population Q, wherein each non-dominated solution corresponds to a task scheduling scheme, and a decision maker can select one scheme from the schemes to execute according to the preference or the requirement of the decision maker;

the algorithm termination condition in S11 is: the current iteration number gen of the algorithm is equal to the preset maximum iteration number MaxGen; simulation experiment:

carrying out a simulation experiment by using a CloudSim 4.0 simulation platform, wherein the processing rate of the constructed virtual machine is 1000MIPS, and the MIPS represents the number of millions of instructions processed per second and is a measurement unit of the processing resource capacity of the virtual machine in the CloudSim simulation platform; by setting n (n is 10,20,40) virtual machines with the same configuration and m (m is 100,300,500) tasks with the task length ranging from 1000-; at a given placeUnder the experimental condition, simulation experiments are carried out by utilizing the algorithm (MOEA/D-CC) provided by the invention and two comparative multi-target algorithms (SPEA-II and NSGA-III) under different scale calculation examples in 9 so as to verify the effectiveness of the algorithm provided by the invention; the stopping criterion of each algorithm is that the evaluation times of the objective function reach 50000 times; the population size of the MOEA/D-CC and NSGA-III algorithms is set to be 100, and the population size of the SPEA-II algorithm is set to be 40; cross probability P_c0.8, mutation probability P_m0.1; in order to avoid the contingency of the experimental result, each algorithm independently runs for 10 times under the combination of each virtual machine number and task number, and the experimental result is recorded for result analysis;

TABLE 3

Example number	Number of virtual machines n	Number of tasks m	Length of task
				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

Under each calculation example, respectively counting the total completion time (f) of the MOEA/D-CC, SPEA-II and NSGA-III algorithm solving tasks₁) And virtual machine balance load standard deviation (f)₂) The median of (2) is used for drawing different algorithms on different scalesThe comparison graph of the task completion time and the standard deviation of the balance load of the virtual machine is shown in the attached figures 7 and 8; as can be seen from the attached figure 7, the MOEA/D-CC can obtain the shortest total task completion time under all the calculation examples, and the MOEA/D-CC scheduling advantages are more obvious along with the increase of the number of tasks; as can be seen from the attached figure 8, under different scale calculation examples, the fluctuation of the virtual machine balanced load standard deviation of the scheduling scheme obtained by the MOEA/D-CC algorithm is small, and the MOEA/D-CC scheduling advantage is more obvious along with the increase of the number of virtual machines;

to further show the scheduling result, a task scheduling schematic diagram when the virtual machine is 10 and the number of tasks is 100 is drawn, as shown in fig. 9; as can be seen from fig. 9, the computation completion time of 10 virtual machines is basically the same, which avoids the waste of computation resources and improves the computation efficiency of tasks;

the effectiveness of the manufacturing enterprise data space system task scheduling method based on the decomposition strategy can be proved through the experiment, the total execution time of the tasks can be obviously shortened, the load balance of the virtual machine can be guaranteed, and the method has positive significance for improving the performance and the stability of the manufacturing enterprise data space system.

Claims (10)

1. A manufacturing enterprise data space system task scheduling method based on a decomposition strategy is characterized by comprising the following steps:

s1: preprocessing required data according to task requirements;

s2: setting a maximum iteration number termination criterion MaxGen and setting a value of a number;

s3: randomly initializing a first generation initial population Q ═ x¹,x²,...,x^N)；

Calculating each individual x in the population QⁱA fitness value F of (i ═ 1, 2.., N)ⁱ＝(f₁(x)，f₂(x) In which the first objective function f₁(x) For the total completion time of the task, a second objective function f₂(x) The index is the load balance index of the virtual machine;

s4: setting the current iteration number gen as 1;

s5: setting a temporary set R & ltphi & gt and alpha & lt1 & gt;

s6: distributing each individual in the population Q to a corresponding sub-population P according to the fitness value of each individual in the population Q₁,P₂,...,P_KPerforming the following steps;

s7: pair sub-population P_αEach of the individuals x in^β(β ═ 1, 2.., N) performing a cross mutation operation to generate offspring individuals, and storing the offspring individuals into the temporary set R;

s8: combining the current population Q with the temporary set R, wherein Q ═ R ═ U.Q;

s9: setting alpha to alpha +1, if alpha is less than or equal to K, returning to S7, and if alpha is more than K, executing the next step;

s10: setting gen as gen + 1;

s11: and judging whether the algorithm meets a termination condition MaxGen according to the current iteration times of the algorithm, if the algorithm does not meet a stop condition, repeating S5-S10, if all non-dominant solutions in the population Q are output according to the condition, wherein each non-dominant solution corresponds to a task scheduling scheme, and a decision maker can select one scheme from the schemes to execute according to the preference or the requirement of the decision maker.

2. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 1, wherein: the specific process of S1 includes the following steps:

s1.1: selecting an optimization model of a task scheduling problem according to requirements;

taking a plurality of tasks uploaded by all users as a total task, delivering the total task to an enterprise data space system for task scheduling, and distributing the tasks to n virtual machines for processing according to a scheduling result so as to enable the task distribution result to meet a plurality of scheduling targets of a system task decision system, wherein the total task comprises m subtasks which are independent from each other; neglecting delay and loss in the data transmission process, mapping all computing resources in the system into virtual machines, wherein the number of the subtasks is more than that of the virtual machines;

s1.2: selecting a task scheduling target according to requirements;

distributing the m tasks to the n virtual machines to enable the total completion time of the m tasks to be shortest and the loads of the virtual machines to be balanced;

s1.3: selecting an individual coding mode according to requirements;

the chromosome coding adopts an integer coding mode, each chromosome represents a scheduling scheme and reflects the mapping relation between task scheduling arrangement and resources, and the length of the chromosome is equal to the number m of tasks; the genetic locus number on the chromosome represents a task number, and the value range is {1, 2.., m }; the gene value on the gene position represents the virtual machine number allocated to the task, and the value range is {1, 2.., n }.

3. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 1, wherein: in the S2, the parameters: comprises a population scale N, a sub-population number K, a sub-population scale S and a cross probability p_cProbability of variation p_mWherein N ═ K × S.

4. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 1, wherein: the specific process of S3 is as follows:

evaluation of Each individual x in the population QⁱA fitness value of (i ═ 1, 2.., N); the evaluation of the fitness value of the individual comprises a first objective function f₁(x) For the total completion time of the task and a second objective function f₂(x) For two aspects of the virtual machine load balance index, the fitness value function is specifically as follows:

s3.1: function f of fitness value of total completion time of task₁(x)：

When s tasks are distributed to the jth virtual machine M_jThen, virtual machine M_jTime required to complete the assignment task

Wherein, t_ijFor task T_iIn virtual machine M_jTime required for upper treatmentThe tasks are mutually independent, and the completion time C of the total task after the completion time of the task on each virtual machine is obtained_maxComprises the following steps:

f₁(x)＝C_max＝max{C_j},j∈{1,2,...,n} (2)

wherein, C_jDelegate virtual machine M_jThe time required for completing all tasks, wherein n is the number of virtual machines;

task allocation in S3.1: executing tasks according to a first-in first-out principle when a plurality of tasks are distributed to one virtual machine;

s3.2: load balancing objective function f for virtual machine₂(x)：

The average load capacity of all the virtual machines in the server cluster is the average time AL required by all the virtual machines to complete the task, and the calculation mode is as shown in formula (3):

the calculation mode of the virtual machine balanced load standard deviation LSD is shown in formula (4):

5. the decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 1, wherein: the specific process of S6 includes the following steps:

s6.1: generating K uniformly distributed vectors in a solution space omega, and solving the solution space R^wDivided into K sub-regions omega_ε(e ═ 1,2,. K), w is the number of objective functions;

s6.2: setting a sub-population index epsilon as 1;

s6.3: individuals x in population Qⁱ(i ∈ {1, 2.,. N }) according to its fitness value Fⁱ＝(f₁(x),f₂(x) In the target space, assign it to the pairThe mode of the strain group is

Wherein<Fⁱ,v^ε>Representative vector FⁱSum vector v^εThe included angle of the acute angle is smaller than the included angle of the acute angle,<Fⁱ,v^l>representative vector FⁱSum vector v^lAcute included angle of (i.e. individual x)ⁱF of (A)ⁱVector and uniformly distributed vector v^εWhen the included angle of (A) is smallest, the individual xⁱWill be assigned to a sub-population P_εPerforming the following steps;

s6.4: judging the sub-population P_εNumber of individuals | P_εIf | exceeds the sub-population scale set value S, if | P_εIf | < S, randomly selecting S- | P from the population Q_εI Individual addition of P_εIn, if | P_ε| is not less than S, for P_εThe individuals in (1) perform non-dominant sorting, deleting | P_ε-S individuals with the highest non-dominated ranking;

s6.5: if epsilon is less than K, epsilon +1 and return to S6.4, if epsilon > K, the sub-population P is output₁,P₂,...,P_K。

6. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 5, wherein: the non-dominant ranking calculation method in S6.4 is described as follows:

1): setting the existence of individuals x, S in the population Q_xFor x assignable sets of individuals, n_xFor any individual y in Q, if x dominates y (denoted y > x), then S_x:＝S_xU { y }, otherwise, n_x＝n_x+1；

2): if n is_xX is 0_rankStore it in the set U as 1₁Performing the following steps;

3): setting idx ═ 1, a ═ Φ, pair

Is/are as follows

n_q＝n_q-1; if n is_qQ is added to a and q is 0_rank＝idx+1；

4): let idx be idx +1, U_idx＝A；

5): if U is_idxIf the result is phi, stopping and outputting a non-dominant grade sequencing result; otherwise, go to 3).

7. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 1, wherein: the specific process of S7 includes the following steps:

s7.1: randomly selecting P from the current sub-population_αSelecting an individual y (y ≠ x)^β)；

S7.2: individual x^βPerforming two-point crossing operation with the individual y according to the crossing probability, and generating two filial generation individuals after crossing;

generating a random number r_cBelongs to (0,1) and is associated with the mutation probability P_cMaking a comparison when r_c<P_cThe following operations are executed:

randomly selecting two gene positions on the chromosome, and if the two gene positions are the same, regenerating until the two gene positions are different; defining the region between the two gene positions as a cross domain, and interchanging the cross domains of the two parent individuals to generate two offspring individuals;

s7.3: comparing the dominance relation and the crowding distance of the two descendant individuals generated in the S7.2, and keeping the individuals y' with good quality;

s7.4: performing two-point interchange mutation operation on the child individual y' according to the mutation probability to generate a child individual z;

s7.5: and adding the individual z into the temporary set R, wherein R ═ R { z }.

8. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 7, wherein: the domination relationship of the filial generation individuals in the 7.3 is as follows: for a minimization problem involving w targets, the w target components aref_γ(x) γ ═ 1,2,. ·, w }; for any two feasible solutions x in decision space_aAnd x_bIf it satisfies the following condition in the target space, it is called x_aDominating x_b：

Condition 1: for γ ∈ {1,2,..., w }, f ∈ } f_γ(x_a)≤f_γ(x_b) Both are true;

condition 2: γ ∈ {1, 2., w }, such that f_γ(x_a)<f_γ(x_b) This is true.

9. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 7, wherein: the congestion distance calculation process in S7.3 is as follows:

respectively sequencing the individuals on different target functions in an ascending order, and calculating the crowding distance of the individuals on the target according to the sequencing; individual xⁱCrowding distance on the gamma objective function

Where γ ∈ {1,2,.., w },

and

respectively represent an individual xⁱ⁺¹And individual x^i-1A target value on the gamma target;

and

respectively representing the maximum value and the minimum value of the gamma-th objective function; further, the crowd distance of the individuals at the first and last in the target value ranking is positive infinity; synthesizing crowding distances on w targets, and taking the crowding distances as an individual xⁱDistance d of degree of crowding_iWhich isThe calculation method is as follows:

to describe how dense an individual is in the target space relative to the rest of the individual.

10. The decomposition-policy-based manufacturing enterprise data space system task scheduling method of claim 7, wherein: the two-point interchange variation in S7.4 is described as follows:

generating a random number r_mBelongs to (0,1) and is associated with the mutation probability P_mMaking a comparison when r_m<P_mThe following operations are executed:

randomly selecting two gene positions on the chromosome, if the two gene positions are the same, regenerating until the two gene positions are different, and interchanging the genes on the two gene positions to generate a filial generation individual.

Images