CN106845642A - A kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule - Google Patents

A kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule Download PDF

Info

Publication number
CN106845642A
CN106845642A CN201710046208.0A CN201710046208A CN106845642A CN 106845642 A CN106845642 A CN 106845642A CN 201710046208 A CN201710046208 A CN 201710046208A CN 106845642 A CN106845642 A CN 106845642A
Authority
CN
China
Prior art keywords
population
pareto
evolution
individuality
entropy
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710046208.0A
Other languages
Chinese (zh)
Other versions
CN106845642B (en
Inventor
刘丽
张淼
李慧琦
范琦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Mingyida Technology Co ltd
Original Assignee
University of Science and Technology Beijing USTB
Beijing Institute of Technology BIT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Science and Technology Beijing USTB, Beijing Institute of Technology BIT filed Critical University of Science and Technology Beijing USTB
Priority to CN201710046208.0A priority Critical patent/CN106845642B/en
Publication of CN106845642A publication Critical patent/CN106845642A/en
Application granted granted Critical
Publication of CN106845642B publication Critical patent/CN106845642B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/10—Office automation; Time management
    • G06Q10/103—Workflow collaboration or project management

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Operations Research (AREA)
  • Biomedical Technology (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Artificial Intelligence (AREA)
  • General Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The present invention provides a kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule, it is possible to increase the global detection of multi-target evolution method and local producing capacity.Methods described includes：S1, the number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution, according to the population for detecting Evolving State residing during evolution, adaptively using corresponding individuality assessment strategy treatment constraints, and the individuality in population is ranked up, wherein, in individual assessment strategy, constraints is processed using constraint violation processing method；S2, according to individual ranking results, from population select individuality carry out genetic manipulation, obtain sub- population, wherein, when genetic manipulation is carried out, evolution parameter is adaptively adjusted according to population Evolving State residing during evolution.The present invention is applied to the multi-objective optimization question for solving belt restraining, and can be applied to workflow schedule technical field in cloud computing environment.

Description

A kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule
Technical field
The present invention relates to solve the problems, such as the multiple-objection optimization of belt restraining, and cloud workflow schedule technical field is applied to, Particularly relate to a kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule.
Background technology
Workflow schedule under cloud environment is (referred to as：Cloud workflow schedule) it is to find suitable cloud resource to perform workflow Task, and meet the QoS requirement of user.Cloud workflow schedule problem is a multi-objective optimization question for belt restraining, many Target evolution algorithm can effectively process problems.But in the prior art, come simply by static penalty Treatment constraints, is so easily caused Premature Convergence, even into infeasible search space, for example：
Prior art one, multiple target grain is assessed by using Pareto (Pareto) entropy information and Pareto difference entropy informations The diversity and Evolving State of population in swarm optimization, and carry out Design evolution strategy as feedback information so that algorithm With more preferable convergence and diversity.
Prior art two, is improved on the basis of NSGA-II algorithms, by the whole Pareto of discretization it is optimal before Edge, finds some well distributed reference points, with these points as the direction of search in the evolutionary process of algorithm, finds and offer With reference to point set be associated Pareto optimal solutions or Pareto optimal solutions near solution, so as to get disaggregation have more excellent convergence Property and diversity.
Prior art one and prior art two, although the non-dominant disaggregation that can obtain algorithm has more preferable convergence And diversity, but it is directed to unconfined multi-objective optimization question.It is one for the workflow schedule problem under cloud environment The multi-objective optimization question of individual belt restraining, when problems are processed, conventional method is, based on multi-objective Evolutionary Algorithm, to use Constrained Optimization is converted into unconstrained optimization problem by static penalty, if penalty is too small, some it is non-can The fitness value of row solution will be above most of feasible solution, and population is likely to be evolved towards a non-feasible search space；But If penalty is too big, many more excellent individualities will be excluded, so as to cause Premature Convergence, to sum up, existing multiple target Evolution algorithm is easily trapped into local optimum when constraints is processed using static penalty.
The content of the invention
The technical problem to be solved in the present invention is to provide a kind of self adaptation multi-target evolution of belt restraining cloud workflow schedule Method, it is easy when constraints is processed using static penalty to solve the multi-objective Evolutionary Algorithm existing for prior art It is absorbed in the problem of local optimum.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of many mesh of self adaptation of belt restraining cloud workflow schedule Mark evolvement method, including：
S1, the number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution, root According to the population for detecting Evolving State residing during evolution, adaptively processed about using corresponding individuality assessment strategy Beam condition, and being ranked up to the individuality in population, wherein, in individual assessment strategy, using constraint violation processing method come Treatment constraints；
S2, according to individual ranking results, from population select individuality carry out genetic manipulation, obtain sub- population, wherein, entering During row genetic manipulation, evolution parameter is adaptively adjusted according to population Evolving State residing during evolution.
Further, the number solved according to Pareto and Pareto entropys detection population is residing during evolution enters Change state includes：
If not having Pareto to solve in population, population Evolving State residing during evolution is original state.
Further, the number solved according to Pareto and Pareto entropys detection population is residing during evolution enters Change state includes：
If the number of Pareto solutions is less than Population Size in population, population Evolving State residing during evolution is Convergence state；Or,
If the number of Pareto solutions is equal to Population Size in population, and population is in t+1 iteration, the number of Pareto solutions Change is there occurs, then population Evolving State residing during evolution is convergence state；Or,
If the number of Pareto solutions is equal to Population Size in population, in t+1 iteration, the number of Pareto solutions does not have population Change, andThen population Evolving State residing during evolution is receipts Hold back state；
Wherein, △ Entropy (t+1) represent the t+1 times iteration and the t times poor entropy of iteration, and M represents the individual of optimization aim Number, △ Entropymax-diverRepresent maximum difference entropy.
Further, the number solved according to Pareto and Pareto entropys detection population is residing during evolution enters Change state includes：
If the number of Pareto solutions is equal to Population Size in population, in t+1 iteration, the number of Pareto solutions does not have population Change, andThen population Evolving State residing during evolution is many Sample state；Or,
If the number of Pareto solutions is equal to Population Size in population, in t+1 iteration, the number of Pareto solutions does not have population Change, Pareto entropys also do not change, then population Evolving State residing during evolution is maturity state；
Wherein, △ Entropy (t+1) represent the t+1 times iteration and the t times poor entropy of iteration, and M represents the individual of optimization aim Number, △ Entropymax-diverRepresent maximum difference entropy.
Further, the △ Entropymax-diverIt is expressed as：
Wherein, EntropymaxPareto entropys when representing Pareto solutions for optimal distribution, EntropyminRepresent Pareto Pareto entropys during for worst distribution are solved, M represents the number of optimization aim, and K represents the grid division under every one-dimensional optimization aim, Cellk,mT () represents the t times iterationIn individual amount,It isThe Based on Integer Labelling in PCCS is mapped to,Represent the K m-th lattice coordinate components of Pareto solutions,Represent that k-th Pareto solves the value of corresponding m-th optimization aim, l tables Show EntropyminIn the case of individual that grid of only one of which coordinate.
Further, the basis is detected population Evolving State residing during evolution, is adaptive selected Corresponding individuality assessment strategy treatment constraints, and the individual in population is ranked up, including：
If population Evolving State residing during evolution is detected for original state, according to the first fitness function The ideal adaptation angle value of population is assessed, wherein, first fitness function is expressed as：
Size according to ideal adaptation angle value is ranked up to the individuality in population；
Wherein, Fa(xi) represent the fitness value of individuality i, xiRepresent individuality i, a and represent Fa(xi) it is after having added penalty Fitness value,It is normalized constraint violation.
Further, the basis is detected population Evolving State residing during evolution, is adaptive selected Corresponding individuality assessment strategy treatment constraints, and the individual in population is ranked up, including：
If population Evolving State residing during evolution is detected for convergence state, according to the second fitness function The ideal adaptation angle value of population is assessed, wherein, second fitness function is expressed as：
Size according to ideal adaptation angle value is ranked up to the individuality in population；
Wherein,Represent the fitness value of k-th target of individuality i, xiIndividuality i is represented, a is representedIt is to add Fitness value after penalty,Fitness value after the normalization of k-th target of expression individuality i,Table Show normalized constraint violation, rfRepresent possible ratios.
Further, the basis is detected population Evolving State residing during evolution, adaptively utilizes Corresponding individuality assessment strategy, the individuality in population is ranked up including：
If it is diversified state or maturity state to detect population Evolving State residing during evolution, based on about The dominated Sorting principle of beam is ranked up to the individuality in population.
Further, the evolution parameter includes：Crossover probability and mutation probability；
It is described when genetic manipulation is carried out, according to population Evolving State residing during evolution be adaptively adjusted into Changing parameter includes：
When genetic manipulation is carried out, the adjustment of self adaptation is carried out to the crossover probability and mutation probability in genetic manipulation, its In, the t times crossover probability p of iterationcT the regulation rule of () is expressed as：
The t times mutation probability p of iterationmT the regulation rule of () is expressed as：
Wherein,<·>Represent a holding pc(t) and pmT () is in the function among given border, work as pc(t) and pm T () is less than lower boundary, lower border value is given into they, works as pc(t) and pmT () is higher than coboundary, upper boundary values are given into it , pc(t-1) the t-1 times crossover probability of iteration, p are representedm(t-1) the t-1 times mutation probability of iteration, △ np (t) tables are represented Show the change number of the Pareto number of individuals in the t times iteration, △ Entropy (t-1) represent the t-1 times iteration and change for the t-2 times The poor entropy in generation, ciAnd mjRepresent adjusting step, i, j=1,2,3,4.
Above-mentioned technical proposal of the invention has the beneficial effect that：
In such scheme, the number and Pareto entropys solved according to Pareto detect population evolution residing during evolution State, according to the population for detecting Evolving State residing during evolution, adaptively using corresponding individual assessment plan Constraints is slightly processed, and the individuality in population is ranked up, wherein, in individual assessment strategy, at constraint violation Reason method processes constraints, can effectively process constraints；According to individual ranking results, individuality is selected to enter from population Row genetic manipulation, obtains sub- population, wherein, when genetic manipulation is carried out, according to population evolution shape residing during evolution State is adaptively adjusted evolution parameter, multi-target evolution method is had good convergence and diversity, is solving cloud work There is preferably performance when flowing the optimization problem dispatched so that can also have preferably global detection and part under constraints Producing capacity.
Brief description of the drawings
Fig. 1 is the flow of the self adaptation multi-target evolution method of belt restraining cloud workflow schedule provided in an embodiment of the present invention Schematic diagram；
Fig. 2 is the detailed of the self adaptation multi-target evolution method of belt restraining cloud workflow schedule provided in an embodiment of the present invention Schematic flow sheet；
Fig. 3 represents Epigenomics workflows total executory cost TEC and degree of unbalancedness DI under the constraints of time limit 1 Equilibrium solution；
Fig. 4 represents Epigenomics workflows total executory cost TEC and degree of unbalancedness DI under the constraints of time limit 2 Equilibrium solution；
Fig. 5 represents Epigenomics workflows total executory cost TEC and degree of unbalancedness DI under the constraints of time limit 3 Equilibrium solution；
Fig. 6 represents Epigenomics workflows total executory cost TEC and degree of unbalancedness DI under the constraints of time limit 4 Equilibrium solution；
Fig. 7 represents the equilibrium of Inspiral workflows total executory cost TEC and degree of unbalancedness DI under the constraints of time limit 1 Solution；
Fig. 8 represents the equilibrium of Inspiral workflows total executory cost TEC and degree of unbalancedness DI under the constraints of time limit 2 Solution；
Fig. 9 represents the equilibrium of Inspiral workflows total executory cost TEC and degree of unbalancedness DI under the constraints of time limit 3 Solution；
Figure 10 represent Inspiral workflows under the constraints of time limit 4 total executory cost TEC and degree of unbalancedness DI it is equal Weighing apparatus solution.
Specific embodiment
To make the technical problem to be solved in the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing and tool Body embodiment is described in detail.
The present invention easily falls into for existing multi-objective Evolutionary Algorithm when constraints is processed using static penalty Enter the problem of local optimum, there is provided a kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule.
As shown in figure 1, the self adaptation multi-target evolution method of belt restraining cloud workflow schedule provided in an embodiment of the present invention, Including：
S1, the number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution, root According to the population for detecting Evolving State residing during evolution, adaptively processed about using corresponding individuality assessment strategy Beam condition, and being ranked up to the individuality in population, wherein, in individual assessment strategy, using constraint violation processing method come Treatment constraints；
S2, according to individual ranking results, from population select individuality carry out genetic manipulation, obtain sub- population, wherein, entering During row genetic manipulation, evolution parameter is adaptively adjusted according to population Evolving State residing during evolution.
The self adaptation multi-target evolution method of the belt restraining cloud workflow schedule described in the embodiment of the present invention, according to Pareto The number and Pareto entropys of solution detect population Evolving State residing during evolution, are being evolved according to the population for detecting Residing Evolving State in journey, adaptively using corresponding individuality assessment strategy treatment constraints, and to population in Body is ranked up, wherein, in individual assessment strategy, constraints is processed using constraint violation processing method, can be effective Treatment constraints；According to individual ranking results, select individuality to carry out genetic manipulation from population, obtain sub- population, wherein, When carrying out genetic manipulation, evolution parameter is adaptively adjusted according to population Evolving State residing during evolution, makes many mesh Mark evolvement method has good convergence and diversity, has preferably property when the optimization problem of cloud workflow schedule is solved Can so that can also have preferably global detection and local producing capacity under constraints.
The self adaptation multi-target evolution method of belt restraining cloud workflow schedule provided in an embodiment of the present invention is properly termed as base In NSGA-II algorithms (the Pareto Entropy based on NSGA- with self adaptation individuality assessment strategy of Pareto entropys II with adaptive individual-assessment scheme, ai-NSGA-II-PE), ai-NSGA-II-PE with Multi-objective optimization algorithm (Non-dominated Sorting the Genetic Algorithm-II, NSGA- of non-dominated ranking II based on), as shown in Fig. 2 the self adaptation multi-target evolution side of belt restraining cloud workflow schedule provided in an embodiment of the present invention Method can specifically include：
S201, initializes population, including iterations, population scale etc.；
S202, the number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution, According to the population for detecting Evolving State residing during evolution, adaptively using corresponding individuality assessment strategy treatment Constraints, and the individuality in population is ranked up；Wherein, the number solved according to Pareto and the detection kind of Pareto entropys The residing Evolving State in NSGA-II evolutionary process of group includes：
(1) if not having Pareto to solve in population, it is initial shape to define population Evolving State residing during evolution State；
(2) following 3 kinds of situations, it is convergence state to define population Evolving State residing during evolution：
The number of Pareto solutions is less than Population Size in a, population；
The number of Pareto solutions is equal to Population Size in b, population, but population is in t+1 iteration, the number of Pareto solutions There occurs change；
The number of Pareto solutions is equal to Population Size in c, population, and in t+1 iteration, the number of Pareto solutions does not have population Change, andWherein, △ Entropy (t+1) represent the t+1 time iteration with The t times poor entropy of iteration, M represents the number of optimization aim, △ Entropymax-diverRepresent maximum difference entropy.
In the present embodiment, in order to more fully understand Pareto entropys, difference entropy, maximum difference entropy △ Entropymax-diver, it is right Pareto entropys, difference entropy, maximum difference entropy △ Entropymax-diverIt is described in detail：
In the present embodiment, can by parallel lattice coordinate system (Parallel Cell Coordinate System, PCCS) come describe Pareto entropys and its difference entropy.M is the number of optimization aim, and the grid division under every one-dimensional optimization aim is K,Represent that k-th Pareto solves the value of corresponding m-th optimization aim,Being mapped to one according to formula (1) has K × M lattice The two dimensional surface grid of son,It isThe Based on Integer Labelling in PCCS is mapped to, represents that m-th lattice of k-th Pareto solution are sat Mark component.The coordinate components of the solution in each cartesian coordinate system may be mapped to some in two dimensional surface grid In grid.
OrderWhereinIt is the function that rounds up, that is, returns to a smallest positive integral not less than x,WithIt is respectively the maximum and minimum value of m-th target in gathering for current Pareto solutions,IfWillIt is set to 1.
In the present embodiment, the t times Pareto entropys Entropy (t) of iteration can be expressed as：
Wherein, Cellk,mT () representsIn individual amount, poor entropy △ Entropy (t) of the t times and t-1 iteration can To be expressed as：
△ Entropy (t)=Entropy (t)-Entropy (t-1)
Maximum difference entropy △ Entropymax-diverA kind of difference of entropy under extreme case is referred to, i.e., in diversified state When, new explanation is in same dominance hierarchy with old solution, but during with more preferable crowding, old solution can be replaced, and makeWithIt is aobvious Change is write, causes solution to be changed into worst distribution from optimal distribution (i.e. the coordinate vector of each target occupies a grid in PCCS) (i.e. the coordinate vector of each target has K-1 to squeeze in a grid in PCCS, remaining to occupy at another), now Changing value (the △ Entropy of Pareto entropysmax-diver) can be expressed as：
Wherein, EntropymaxPareto entropys when representing Pareto solutions for optimal distribution, EntropyminRepresent Pareto Pareto entropys during for worst distribution are solved, M represents the number of optimization aim, and K represents the grid division under every one-dimensional optimization aim, Cellk,mT () represents the t times iterationIn individual amount,It isThe Based on Integer Labelling in PCCS is mapped to,Represent the K m-th lattice coordinate components of Pareto solutions,Represent that k-th Pareto solves the value of corresponding m-th optimization aim, l tables Show EntropyminIn the case of individual that grid of only one of which coordinate.
(3) if the number of Pareto solutions is equal to Population Size in population, population is in t+1 iteration, and it is individual that Pareto is solved Number does not change, andThen define population evolution residing during evolution State is diversified state；
(4) if the number of Pareto solutions is equal to Population Size in population, population is in t+1 iteration, and it is individual that Pareto is solved Number does not change, and Pareto entropys also do not change, then it is ripe shape to define population Evolving State residing during evolution State.
In the present embodiment, then, according to the population for detecting Evolving State residing during evolution, self adaptation land productivity Constraints is processed with corresponding individual assessment strategy, and the individuality in population is ranked up.In the individual assessment ring of self adaptation Section, mainly judges the quality of an individual by two aspects, and one is desired value that it is obtained in object function, another It is whether the constrained objective value that is obtained under constraints exceedes and conclude a contract or treaty beam.
In the present embodiment, individuality is assessed using different individual assessment strategies in different Evolving States, individuality is entered Row sequence, in individual assessment strategy, constraints is processed using constraint violation processing method according to population in evolutionary process In residing different Evolving State, constraints effectively can be processed using different constraint violation processing methods, make More excellent feasible individual is remained；Specifically：
(1) if detecting population Evolving State residing during evolution for original state, according to the first fitness The ideal adaptation angle value of function evaluation population, wherein, first fitness function is expressed as：
Size according to ideal adaptation angle value is ranked up to the individuality in population；
Wherein, Fa(xi) represent the fitness value of individuality i, xiRepresent individuality i, a and represent Fa(xi) it is after having added penalty Fitness value,Normalized constraint violation, will constrained objective value (constrained objective value is under constraints To) be normalized more than the part for concludeing a contract or treaty beam, sort then to according to normalized constraint violation individuality so that Remain the individuality of the constraint violation for possessing small.
(2) if detecting population Evolving State residing during evolution at least one in convergence state, i.e. population After individual feasible solution (feasible solution is referred to as feasible individual), then the ideal adaptation of population is assessed according to the second fitness function Angle value, wherein, second fitness function is expressed as：
Size according to ideal adaptation angle value is ranked up to the individuality in population；
Wherein,Represent the fitness value of k-th target of individuality i, xiIndividuality i is represented, a is representedIt is to add Fitness value after penalty,Fitness value after the normalization of k-th target of expression individuality i,Table Show normalized constraint violation, rfRepresent possible ratios, rfIt is feasible individual quantity and the ratio of population scale.
(3) if it is diversified state or maturity state, this reality to detect population Evolving State residing during evolution Example is applied using another individuality assessment strategy come the individuality that sorts, according to defining 1, the dominated Sorting principle based on constraint is as follows, makes Group hunting must be planted and possess the feasible individual of more preferable crowding, and prevent infeasible solutions from entering population：
Define 1：If following any one condition is true, individuality S is illustrated1Than individual S2It is outstanding：
(1) if S1It is feasible solution and S2It is infeasible solutions；
(2)S1、S2All it is infeasible solutions, S1There is smaller constraint violation；
(3)S1、S2All it is feasible solution, S1Compare S2There is non-dominant grade higher；Or, working as S1And S2There is identical to arrange etc. Level, S1Compare S2There is more excellent crowding, individuality is ranked up by non-dominant grade, wherein, non-dominant grade sequence and gather around Squeeze degree and calculate similar with employed in NSGA-II algorithms；
Non-dominant grade sequence refers to：If one or more optimization target values of certain individual i are superior to other one Individual j, and other optimization target values are equal, then individual i domination individualities j.According to dominance relation to individual distribution non-dominant Grade, the low individuality of grade individual dominance hierarchy high, if individuality m can not be arranged by any individuality, i.e., resulting solution Optimal, then individuality m is called Pareto optimal solutions or non-domination solution.
Crowding, for being ranked up to the individuality in identical non-dominant grade, its basic thought is according to all The optimization target values of body calculate the theorem in Euclid space distance with individual two adjacent individualities.
S203, according to individual ranking results, selects individuality to carry out genetic manipulation from population, obtains sub- population.Possess row There is sequence individuality higher bigger probability to be selected into sub-group.
It is to select the individuality to carry out heredity using different individual assessment strategies in different Evolving States in the present embodiment Operation.
When carrying out corresponding genetic manipulation according to population Evolving State residing during evolution, population is according to residing for it Evolving State adaptively carry out the adjustment of evolution parameter, i.e., the crossover probability and mutation probability in genetic manipulation are carried out from The adjustment of adaptation, so as to improve global detection and local producing capacity.
In the present embodiment, crossover probability and mutation probability in genetic manipulation can be according to population institutes during evolution The number and Pareto entropys that the Evolving State at place, Pareto are solved carry out self-adaptative adjustment, wherein, the t times crossover probability of iteration pcT the regulation rule of () is expressed as：
The t times mutation probability p of iterationmT the regulation rule of () is expressed as：
Wherein,<·>Represent a holding pc(t) and pmT () is in the function among given border, work as pc(t) and pm T () is less than lower boundary, lower border value is given into they, works as pc(t) and pmT () is higher than coboundary, upper boundary values are given into it , pc(t-1) the t-1 times crossover probability of iteration, p are representedm(t-1) the t-1 times mutation probability of iteration, △ np (t) tables are represented Show the change number of the Pareto number of individuals in the t times iteration, △ Entropy (t-1) represent the t-1 times iteration and change for the t-2 times The poor entropy in generation, ciAnd mjRepresent adjusting step, i, j=1,2,3,4；Wherein, ciAnd mjIt is expressed as：
Wherein, gmaxIt is expressed as maximum iteration.
S204, according to father population and sub- population Evolving State residing during evolution, adaptive polo placement father population and The ideal adaptation angle value of sub- population, the same S202 of specific steps.
S205, according to father population and the ideal adaptation angle value of sub- population, determines the individual sequence of father population and sub- population.
In the present embodiment, the individuality according to ideal adaptation angle value from small to large to father population and sub- population is ranked up, its In, during sequence, father population and sub- population are placed on what is operated together.
S206, from father population and sub- population after sequence, selection top n is individual as new population, calculates new population Pareto entropys and detection Population status, return and perform S202, and new population participates in next iteration, until iterations is equal to default Maximum iteration, wherein, the value of N determines according to actual conditions.
In the present embodiment, from father population and sub- population after sequence, selection top n individuality is used as new population.During sequence, Father population and sub- population are placed on what is operated together, and it is also the top n for selecting two populations (father population and sub- population) to select individuality Body.
In the case where sensitive cloud computing environment is calculated, to solve the problems, such as its workflow schedule, such as drag is set up：
Optimization aim (Minimize) is the total executory cost (Total Execution Cost, TEC) of minimum and injustice Weighing apparatus degree (Degree of Imbalance, DI), constraints (Subject to) performs the time (Total Execution for total Time, TET) it is less than time limit dw, it is expressed as
Minimize:TEC
DI
Subject to:TET≤dw
Wherein, | VM | represents the number of virtual machine,The virtual machine of execution task i is represented,It is virtual machine unit Time cost,It isThe actual run time of execution task i, tiExpression task i, τ are virtual machine processed in units energy Power,Transmission time between expression task i and task j, subscript ei,jWhat expression task i and task j were connected to,It isData conversion cost, the T in i ∈ T, j ∈ T refers to all of task subscript, TmaxAnd TminAll virtual machines are represented respectively Maximum run time and minimum run time, TavgIt is the average operating time of virtual machine, V is set of tasks,Represent and appoint The end time of business i.
In the present embodiment,The actual run time of execution task iIt is expressed as：
Wherein,It is the size of task i,It isDisposal ability.
In the present embodiment, the transmission time between task i and task jIt is expressed as：
Wherein,WithVirtual machine VM is represented respectivelyiAnd VMjBetween data transfer bandwidth, subscript ei,jExpression task What i and task j were connected to,It is the output data size produced by task i, if task i and task j are in same virtual machine Perform, then transmission time is 0；
In the present embodiment, at the beginning of task i betweenIt is expressed as：
Wherein,It isThe termination time is rented,Refer to the virtual machine of execution task i, taExpression task a, parent(ti) all female task of task i is represented,The end time of expression task a,Expression task a and task i it Between transmission time, subscript ea,iWhat expression task a and task i were connected to.
In the present embodiment, the end time of task iIt is expressed as：
Wherein,Between at the beginning of expression task i,RepresentThe actual run time of execution task i.
In the present embodiment, 4 time limit value are defined when being tested, when these time limit value are in most fast and most slow operation Between between, most fast run time completes algorithm (Heterogeneous Earliest by heterogeneous computing environment earliest time Finish Time, HEFT) traffic control stream obtained, and most slow run time performs all of task and is obtained by a virtual machine , the time limit gradually becomes strict with the increase of numbering.
The present embodiment employs two kinds of workflows of Epigenomics and Inspiral and is tested, wherein, Epigenomics workflows are a kind of highly pipelined application programs, and plurality of line concurrently enters to independent data block Row operation；Inspiral is made up of several sub- workflows.Test result such as Fig. 3, Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8, Fig. 9, Figure 10 It is shown.
Implementation method one
In the present embodiment, time limit value is respectively provided with for the time limit 1, in the time limit 2, in the time limit 3, in the time limit 4, choose research-on-research flow model Epigenomics, by ai-NSGA-II-PE and NSGA-II algorithms (the ParetoEntropy based on based on Pareto entropys NSGA-II, NSGA-II-PE) (NSGA-II-PE uses self-adaptative adjustment evolution parameter proposed by the invention, but uses tradition Static penalty processes constraints), NSGA-II, based on the multi-objective Evolutionary Algorithm (Multi-Objective for decomposing Evolutionary Algorithm based on Decomposition, MOEA/D), strength Pareto evolutionary algorithm (Strength Pareto Evolutionary Algorithm 2, SPEA2), multi-objective particle (Multi- Objective Particle Swarm Optimization, MOPSO) it is compared respectively；Simulation result such as Fig. 3, Fig. 4, figure 5th, Fig. 6, from Fig. 3, Fig. 4, Fig. 5, Fig. 6, compared to other algorithms, ai-NSGA-II-PE energy under strict constraints Find global detection and the more excellent Pareto forward positions of local exploitation effect.
Implementation method two
In the present embodiment, time limit value is respectively provided with for the time limit 1, in the time limit 2, in the time limit 3, in the time limit 4, choose research-on-research flow model Inspiral, by ai-NSGA-II-PE and NSGA-II-PE (use self-adaptative adjustment evolution parameter proposed by the invention, but Use traditional static penalty to process constraints), NSGA-II, MOEA/D, SPEA2, MOPSO evolution algorithm carries out respectively Compare；Simulation result such as Fig. 7, Fig. 8, Fig. 9, Figure 10, from Fig. 7, Fig. 8, Fig. 9, Figure 10, compared to other algorithms, ai- NSGA-II-PE can find global detection and the more excellent Pareto forward positions of local exploitation effect under strict constraints.
The above is the preferred embodiment of the present invention, it is noted that for those skilled in the art For, on the premise of principle of the present invention is not departed from, some improvements and modifications can also be made, these improvements and modifications Should be regarded as protection scope of the present invention.

Claims (9)

1. a kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule, it is characterised in that including：
S1, the number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution, according to inspection The population for measuring Evolving State residing during evolution, adaptively using corresponding individuality assessment strategy treatment constraint bar Part, and the individuality in population is ranked up, wherein, in individual assessment strategy, processed using constraint violation processing method Constraints；
S2, according to individual ranking results, from population select individuality carry out genetic manipulation, obtain sub- population, wherein, lost When passing operation, evolution parameter is adaptively adjusted according to population Evolving State residing during evolution.
2. the self adaptation multi-target evolution method of belt restraining cloud workflow schedule according to claim 1, it is characterised in that The number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution include：
If not having Pareto to solve in population, population Evolving State residing during evolution is original state.
3. the self adaptation multi-target evolution method of belt restraining cloud workflow schedule according to claim 1, it is characterised in that The number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution include：
If the number of Pareto solutions is less than Population Size in population, population Evolving State residing during evolution is convergence State；Or,
If the number of Pareto solutions is equal to Population Size in population, and population is in t+1 iteration, and the number of Pareto solutions occurs Change, then population Evolving State residing during evolution is convergence state；Or,
If the number of Pareto solutions is equal to Population Size in population, population in t+1 iteration, do not send out by the number of Pareto solutions Changing, andThen population Evolving State residing during evolution is convergence shape State；
Wherein, △ Entropy (t+1) represent the t+1 times iteration and the t times poor entropy of iteration, and M represents the number of optimization aim, △Entropymax-diverRepresent maximum difference entropy.
4. the self adaptation multi-target evolution method of belt restraining cloud workflow schedule according to claim 1, it is characterised in that The number solved according to Pareto and Pareto entropys detection population Evolving State residing during evolution include：
If the number of Pareto solutions is equal to Population Size in population, population in t+1 iteration, do not send out by the number of Pareto solutions Changing, andThen population Evolving State residing during evolution is variation State；Or,
If the number of Pareto solutions is equal to Population Size in population, population in t+1 iteration, do not send out by the number of Pareto solutions Changing, Pareto entropys also do not change, then population Evolving State residing during evolution is maturity state；
Wherein, △ Entropy (t+1) represent the t+1 times iteration and the t times poor entropy of iteration, and M represents the number of optimization aim, △Entropymax-diverRepresent maximum difference entropy.
5. the self adaptation multi-target evolution method of the belt restraining cloud workflow schedule according to claim 3 or 4, its feature exists In the △ Entropymax-diverIt is expressed as：
&Delta;Entropy max - d i v e r = Entropy max - Entropy min = - &Sigma; k = 1 K &Sigma; m = 1 M 1 K M log 1 K M + &Sigma; k = 1 K - 1 &Sigma; m = 1 M Cell k &NotEqual; l , m ( t ) K M log Cell k &NotEqual; l , m ( t ) K M + &Sigma; m = 1 M Cell k = l , m ( t ) K M log Cell k = l , m ( t ) K M = log K M - ( K - 1 K log K M K - 1 + 1 K log K M ) = K - 1 K log K M - K - 1 K log K M K - 1 = K - 1 K log ( K - 1 )
Wherein, EntropymaxPareto entropys when representing Pareto solutions for optimal distribution, EntropyminRepresent that Pareto solutions are Pareto entropys during worst distribution, M represents the number of optimization aim, and K represents the grid division under every one-dimensional optimization aim, Cellk,mT () represents the t times iterationIn individual amount,It isThe Based on Integer Labelling in PCCS is mapped to,Represent the K m-th lattice coordinate components of Pareto solutions,Represent that k-th Pareto solves the value of corresponding m-th optimization aim, l tables Show EntropyminIn the case of individual that grid of only one of which coordinate.
6. the self adaptation multi-target evolution method of belt restraining cloud workflow schedule according to claim 1, it is characterised in that The population that the basis is detected Evolving State residing during evolution, is adaptive selected corresponding individuality assessment strategy Treatment constraints, and the individuality in population is ranked up, including：
If detecting population Evolving State residing during evolution for original state, assessed according to the first fitness function The ideal adaptation angle value of population, wherein, first fitness function is expressed as：
F a ( x i ) = E ~ ( x i )
Size according to ideal adaptation angle value is ranked up to the individuality in population；
Wherein, Fa(xi) represent the fitness value of individuality i, xiRepresent individuality i, a and represent Fa(xi) be added it is suitable after penalty Answer angle value,It is normalized constraint violation.
7. the self adaptation multi-target evolution method of belt restraining cloud workflow schedule according to claim 1, it is characterised in that The population that the basis is detected Evolving State residing during evolution, is adaptive selected corresponding individuality assessment strategy Treatment constraints, and the individuality in population is ranked up, including：
If detecting population Evolving State residing during evolution for convergence state, assessed according to the second fitness function The ideal adaptation angle value of population, wherein, second fitness function is expressed as：
Size according to ideal adaptation angle value is ranked up to the individuality in population；
Wherein,Represent the fitness value of k-th target of individuality i, xiIndividuality i is represented, a is representedIt is to have added punishment Fitness value after function,Fitness value after the normalization of k-th target of expression individuality i,Represent normalizing The constraint violation of change, rfRepresent possible ratios.
8. the self adaptation multi-target evolution method of belt restraining cloud workflow schedule according to claim 1, it is characterised in that The population that the basis is detected Evolving State residing during evolution, adaptively using corresponding individual assessment plan Slightly, the individuality in population is ranked up including：
If it is diversified state or maturity state to detect population Evolving State residing during evolution, based on constraint Dominated Sorting principle is ranked up to the individuality in population.
9. the self adaptation multi-target evolution method of belt restraining cloud workflow schedule according to claim 1, it is characterised in that The evolution parameter includes：Crossover probability and mutation probability；
It is described when genetic manipulation is carried out, according to population Evolving State residing during evolution be adaptively adjusted evolution ginseng Number includes：
When genetic manipulation is carried out, the adjustment of self adaptation is carried out to the crossover probability and mutation probability in genetic manipulation, wherein, the The t crossover probability p of iterationcT the regulation rule of () is expressed as：
The t times mutation probability p of iterationmT the regulation rule of () is expressed as：
Wherein,<·>Represent a holding pc(t) and pmT () is in the function among given border, work as pc(t) and pmT () is low In lower boundary, lower border value is given to them, works as pc(t) and pmT () is higher than coboundary, upper boundary values are given into they, pc (t-1) the t-1 times crossover probability of iteration, p are representedm(t-1) the t-1 times mutation probability of iteration is represented, △ np (t) is represented The change number of Pareto number of individuals in the t times iteration, △ Entropy (t-1) represent the t-1 times iteration and the t-2 times iteration Difference entropy, ciAnd mjRepresent adjusting step, i, j=1,2,3,4.
CN201710046208.0A 2017-01-22 2017-01-22 A kind of adaptive multi-target evolution method of belt restraining cloud workflow schedule Active CN106845642B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710046208.0A CN106845642B (en) 2017-01-22 2017-01-22 A kind of adaptive multi-target evolution method of belt restraining cloud workflow schedule

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710046208.0A CN106845642B (en) 2017-01-22 2017-01-22 A kind of adaptive multi-target evolution method of belt restraining cloud workflow schedule

Publications (2)

Publication Number Publication Date
CN106845642A true CN106845642A (en) 2017-06-13
CN106845642B CN106845642B (en) 2019-03-29

Family

ID=59120017

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710046208.0A Active CN106845642B (en) 2017-01-22 2017-01-22 A kind of adaptive multi-target evolution method of belt restraining cloud workflow schedule

Country Status (1)

Country Link
CN (1) CN106845642B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108650030A (en) * 2018-05-07 2018-10-12 上海电力学院 The multiple convergence node dispositions methods of the water surface of underwater wireless sensor network
CN109361725A (en) * 2018-08-14 2019-02-19 西安电子科技大学 Car networking cloud system resource allocation methods based on multi-objective genetic algorithm
CN109992355A (en) * 2019-01-30 2019-07-09 北京理工大学 A kind of multiple target cloud workflow schedule method based on the non-dominant genetic algorithm of improvement
CN110111146A (en) * 2019-04-30 2019-08-09 东华大学 A kind of multiple-objection optimization method for solving of facing cloud example selection
CN110119558A (en) * 2019-04-30 2019-08-13 广州大学 A kind of circuit parameter optimization method based on difference optimization algorithm
CN112907057A (en) * 2021-02-06 2021-06-04 北京理工大学 Production scheduling optimization method and system based on improved MOPSO algorithm
CN113419854A (en) * 2021-06-23 2021-09-21 东北大学 Cloud resource scheduling method oriented to unbalanced multi-objective optimization
CN113947705A (en) * 2021-10-25 2022-01-18 四川大学 Target detection method based on evolutionary neural network under constraint condition
US11379728B2 (en) 2019-01-09 2022-07-05 Red Hat, Inc. Modified genetic recombination operator for cloud optimization

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120271935A1 (en) * 2011-04-19 2012-10-25 Moon Billy G Coordinating data center compute and thermal load based on environmental data forecasts
CN103019822A (en) * 2012-12-07 2013-04-03 北京邮电大学 Large-scale processing task scheduling method for income driving under cloud environment
CN105046408A (en) * 2015-06-25 2015-11-11 国网山东省电力公司 Configurable workflow realization method and system
CN105354085A (en) * 2015-10-30 2016-02-24 广东石油化工学院 Scheduling method for cloud workflow jobs
CN105790266A (en) * 2016-04-27 2016-07-20 华东交通大学 Microgrid parallel multi-target robust optimization scheduling integrated control method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120271935A1 (en) * 2011-04-19 2012-10-25 Moon Billy G Coordinating data center compute and thermal load based on environmental data forecasts
CN103019822A (en) * 2012-12-07 2013-04-03 北京邮电大学 Large-scale processing task scheduling method for income driving under cloud environment
CN105046408A (en) * 2015-06-25 2015-11-11 国网山东省电力公司 Configurable workflow realization method and system
CN105354085A (en) * 2015-10-30 2016-02-24 广东石油化工学院 Scheduling method for cloud workflow jobs
CN105790266A (en) * 2016-04-27 2016-07-20 华东交通大学 Microgrid parallel multi-target robust optimization scheduling integrated control method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
LI LIU等: "A Survey on Virtual Machine Scheduling in Cloud Computing", 《IEEE》 *
YUNJING XU: "Small is better:avoiding latency trap in virtualized data centers", 《ACM》 *
李克武等: "云环境中基于分解的多目标工作流调度算法", 《计算机工程与科学》 *
王海军等: "求多目标优化问题Pareto最优解集的方法", 《大学数学》 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108650030B (en) * 2018-05-07 2021-05-11 上海电力学院 Water surface multi-sink node deployment method of underwater wireless sensor network
CN108650030A (en) * 2018-05-07 2018-10-12 上海电力学院 The multiple convergence node dispositions methods of the water surface of underwater wireless sensor network
CN109361725A (en) * 2018-08-14 2019-02-19 西安电子科技大学 Car networking cloud system resource allocation methods based on multi-objective genetic algorithm
CN109361725B (en) * 2018-08-14 2020-12-08 西安电子科技大学 Internet of vehicles cloud system resource allocation method based on multi-target genetic algorithm
US11379728B2 (en) 2019-01-09 2022-07-05 Red Hat, Inc. Modified genetic recombination operator for cloud optimization
CN109992355A (en) * 2019-01-30 2019-07-09 北京理工大学 A kind of multiple target cloud workflow schedule method based on the non-dominant genetic algorithm of improvement
CN110111146A (en) * 2019-04-30 2019-08-09 东华大学 A kind of multiple-objection optimization method for solving of facing cloud example selection
CN110119558A (en) * 2019-04-30 2019-08-13 广州大学 A kind of circuit parameter optimization method based on difference optimization algorithm
CN110119558B (en) * 2019-04-30 2023-11-14 广州大学 Circuit parameter optimization method based on differential optimization algorithm
CN110111146B (en) * 2019-04-30 2023-07-11 东华大学 Multi-objective optimization solving method oriented to cloud instance selection
CN112907057B (en) * 2021-02-06 2022-11-11 北京理工大学 Production scheduling optimization method and system based on improved MOPSO algorithm
CN112907057A (en) * 2021-02-06 2021-06-04 北京理工大学 Production scheduling optimization method and system based on improved MOPSO algorithm
CN113419854A (en) * 2021-06-23 2021-09-21 东北大学 Cloud resource scheduling method oriented to unbalanced multi-objective optimization
CN113419854B (en) * 2021-06-23 2023-09-22 东北大学 Unbalanced multi-objective optimization-oriented cloud resource scheduling method
CN113947705B (en) * 2021-10-25 2022-08-12 四川大学 Target detection method based on evolutionary neural network under constraint condition
CN113947705A (en) * 2021-10-25 2022-01-18 四川大学 Target detection method based on evolutionary neural network under constraint condition

Also Published As

Publication number Publication date
CN106845642B (en) 2019-03-29

Similar Documents

Publication Publication Date Title
CN106845642A (en) A kind of self adaptation multi-target evolution method of belt restraining cloud workflow schedule
Liaw et al. Evolutionary manytasking optimization based on symbiosis in biocoenosis
CN103217960B (en) Automatic selection method of dynamic scheduling strategy of semiconductor production line
Eskandari et al. FastPGA: A dynamic population sizing approach for solving expensive multiobjective optimization problems
CN104035816A (en) Cloud computing task scheduling method based on improved NSGA-II
CN102789599B (en) Operation shop bottleneck recognition method based on cluster analysis and multiple attribute decision making
CN105929690B (en) A kind of Flexible Workshop Robust Scheduling method based on decomposition multi-objective Evolutionary Algorithm
CN108133260A (en) The workflow schedule method of multi-objective particle swarm optimization based on real-time status monitoring
CN106897821A (en) A kind of transient state assesses feature selection approach and device
Zeng et al. A dynamic multi-objective evolutionary algorithm based on an orthogonal design
CN105654196A (en) Adaptive load prediction selection method based on electric power big data
Shao et al. A multi-objective discrete invasive weed optimization for multi-objective blocking flow-shop scheduling problem
CN104427505B (en) A kind of method and device of cell scenario division
CN108564592A (en) Based on a variety of image partition methods for being clustered to differential evolution algorithm of dynamic
CN107016461A (en) One kind mixing multi-target evolution method
CN102629305A (en) Feature selection method facing to SNP (Single Nucleotide Polymorphism) data
CN105205052B (en) A kind of data digging method and device
CN108573274A (en) A kind of selective clustering ensemble method based on data stability
CN109659972A (en) Multiple spot photovoltaic power output probability forecasting method and system based on Pair-copula theory
CN111460001A (en) Theoretical line loss rate evaluation method and system for power distribution network
CN105184486A (en) Power grid business classification method based on directed acyclic graphs support vector machine
CN105024645B (en) A kind of photovoltaic array Fault Locating Method that is evolved based on matrix
CN104778368A (en) Pareto set individual ranking method aiming at high-dimensional multi-objective optimization problem
CN103353895A (en) Pre-processing method of power distribution network line loss data
Korejo et al. Multi-population methods with adaptive mutation for multi-modal optimization problems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right

Effective date of registration: 20211105

Address after: 1109, Beichen times building, No. 8, Beichen East Road, Chaoyang District, Beijing 100020

Patentee after: Beijing mingyida Technology Co.,Ltd.

Address before: 100083 No. 30, Haidian District, Beijing, Xueyuan Road

Patentee before: University OF SCIENCE AND TECHNOLOGY BEIJING

Patentee before: BEIJING INSTITUTE OF TECHNOLOGY

TR01 Transfer of patent right