CN105512954A - Integrated search method for large-scale flexible job shop scheduling - Google Patents

Abstract

The invention relates to an integrated search method for large-scale flexible job shop scheduling. The integrated search method comprises the following steps that: 1) a solution vector of FJSS (flexible job shop scheduling) to be solved is defined; 2) a solution vector in hybrid harmony search is converted into the codes of a feasible solution of the FJSS; 3) a hybrid harmony search algorithm is adopted to perform harmony search; 4) the solution vector of the harmony search is converted into the scheduling solution of an FJSP problems, and corresponding neighborhoods are designed based on the scheduling solution, and local search is performed through using the designed neighborhoods, and an improved scheduling solution which is obtained through the local search is converted into a solution vector, and the solution vector is utilized to replace the solution vector before the local search, and the search process of the harmony search is continued, so that an optimal solution can be obtained; and 5) the optimal solution which is found based on the hybrid harmony search algorithm is adopted as the initial solution of a large-scale search algorithm, and the large-scale search algorithm is adopted to continue search until a termination condition is satisfied, and scheduling of which the solution is completion time can be obtained.

Description

A kind of integration search method for extensive flexible job shop scheduling based

Technical field

The present invention relates to computer utility and production scheduling technical field, particularly about a kind of integration search method for extensive flexible job shop scheduling based.

Background technology

In musical performance, musicians rely on oneself memory, and by repeatedly adjusting the tone of each musical instrument in band, finally reach a beautiful harmony state, the people such as Korea S scholar GeemZW, by the inspiration of this phenomenon, propose harmonic search algorithm.By musical instrument i (i=1,2 ..., n) be analogous to i-th variable in optimization problem, the tone of each musical instrument is analogous to the value of each variable, the harmony R of each musical instrument tone _j(j=1,2 ..., m) be analogous to the jth group solution vector of optimization problem, the evaluation effect of music is analogous to objective function.Basic harmonic search algorithm produces m group initial solution (harmony) and puts into harmony data base (HarmonyMemory, HM), searches for new variable with probability HMCR in HM, searches in the scope allowed in HM exogenousd variables with probability 1-HMCR; Then algorithm produces local dip with probability P AR to new variable; Judge whether the objective function of new explanation belongs to the poorest solution in HM, if so, then replaces it; Then continuous iteration is until meet stopping criterion.Major parameter based on harmonic search algorithm has harmony data base size m, data base retains probability (HarmonyMemoryConsideringRate, HMCR) harmony conciliation rate (PitchAdjustingRate, PAR), wherein arranging of algorithm parameter affects algorithm convergence efficiency and speed of convergence, and basic harmony search algorithm process is:

Step 1) initiation parameter data base size m, data base retain probability HMCR, harmony conciliation rate PAR, disturbance factor bw, algorithm iteration times N I;

Step 2) the random m group initial solution that produces puts into HM, the wherein row vector of HM matrix represent a solution of optimization problem;

$HM=\left[\begin{array}{cccc}{x}_1^1& x_2^1& ...& x_n^1\\ x_1^2& x_2^2& ...& x_n^2\\ ...& ...& ...& ...\\ x_1^m& x_2^m& ...& x_n^m\end{array}\right]---\left(1\right)$

Step 3) separate generation new explanation [x by the current m group being in HM ₁' x ₂' ... x _n'], for each decision variable x _i' concrete the rule produced has following 3: data base is selected, tone adjusts and Stochastic choice.First an equally distributed random number rand () between [0,1] is produced, if rand () is less than HMCR, x _i' produced by data base selective rule, otherwise produced by random rule.Secondly, a decision variable x _iif ' produced by data base rule, then also need to adjust through tone with the probability of PAR.The rule that data base is selected, tone adjusts is respectively as shown in formula (2), formula (3) and formula (4):

wherein a ∈ (1,2 ..., m) (2)

x _i′＝x _i′±rand()×bw(3)

x _i′＝LB _i+rand()×(UB _i-LB _i)(4)

In formula, UB _iand LB _ibe respectively the bound scope of i-th decision variable, a refers to and separates x _ii-th decision variable, bw is that algorithm parameter refers to bandwidth.

Step 4) upgrade HM.If the new solution vector [x produced ₁' x ₂' ... x _n'] be better than solution the worst in HM, then replace this worst solution in HM with current new explanation.

Step 5) repeat step 3) and step 4) until reach the iterations of specifying.

In recent years, harmonic search algorithm was widely used in solving combinatorial optimization problem as a kind of global optimization method, in the important combinatorial optimization problems such as traveling salesman problem, pipe laying problem, bus routing problem, be obtained for successful application.In the evolution of harmony search algorithm, many mutation are there is.By the inspiration of the swarm intelligence thought in particle cluster algorithm, Omran and Mahdavi proposes global optimum's harmonic search algorithm (gHS), and the tone regulation rule of this algorithm to basic harmonic search algorithm changes.Suppose that the optimum solution in HM is:

$\left[\begin{array}{cccc}{x}_1^{best}& x_2^{best}& ...& x_n^{best}\end{array}\right]$

Then tone regulation rule is:

wherein, k is the random integers (5) between 1 to n

The people such as LingWang consider the structure of keeping optimization better, are revised as further by tone regulation rule:

${x_i}^{\′}=x_i^{best}---\left(6\right)$

LNS (large neighborhood search) be a kind of combination constraint planning and Local Search to solve the powerful technology of optimization problem.Typical Local Search only makes little change on the basis of current solution, as one or two operation mobile in scheduling solution.But LNS is different with it, the subset of LNS select permeability variable relaxes to problem, and therefore it likely does larger change on current solution basis.In LNS, first select the variable relaxed, then redistribute the value of these variablees, keep the value of other variable constant simultaneously, this step is referred to as to destroy.After this step, by again optimizing the variable of unallocated value to find more excellent solution, this step is referred to as to build.Destroy and build these two steps iteration in LNS to perform, until end condition meets.As shown in Figure 1, the main thought of LNS is operated by destruction, and original problem, by yojan, then runs constraint planning in the problem of yojan, effectively can overcome constraint planning and explore the scarce capacity problem that large search volume exists.

HHS (HybridHarmonySearch, HHS mixing harmony search algorithm) be method based on evolutionary computation, its advantage is the solution that can produce high-quality very soon, but explore development ability when solving extensive problem not enough, when algorithm performs to a certain extent, even if increase iterations further again, acquired results also can not improve further; LNS is the method based on constraint planning, and its advantage is to have very strong search exploitation (climbing the mountain) ability, but shortcoming is that search performance depends on initial solution, and search development ability is sharply successively decreased with the increase of problem scale.

Summary of the invention

For the problems referred to above, the object of this invention is to provide a kind of integration search method that effectively can solve extensive Flexible Job-shop Scheduling Problems.

For achieving the above object, the present invention takes following technical scheme: a kind of integration search method for extensive flexible job shop scheduling based, is characterized in that comprising the following steps: 1) define the solution vector that certain will solve FJSP; 2) solution vector in mixing harmony search algorithm is converted to the coding of FJSP feasible solution; 3) adopt mixing harmonic search algorithm to carry out harmony search algorithm, i.e. first some solution vectors of random initializtion, setting maximum iteration time, per in generation, produces a new solution vector by mixing harmony search algorithm operator; 4) solution vector of mixing harmony search algorithm is converted into the scheduling solution of FJSP problem, Local Search is carried out in field according to arranging, and the scheduling solution after Local Search resulting improvement is converted into solution vector again, solution vector before utilizing this solution vector to replace Local Search, the search procedure continuing harmony search algorithm obtains optimum solution; 5) using based on the initial solution of the optimum solution that finds of mixing harmonic search algorithm as large field searching algorithm, adopt large field searching algorithm to proceed search, until meet end condition, obtain up-to-date scheduling solution and completion date.

Further, described step 1) define the solution vector that certain will solve FJSS, be specially: solution vector is expressed as n representation dimension amount, the variable range of every one dimension is all [-1,1], and the dimension n of solution vector is the twice of solved FJSS scheduling problem operation number summation, wherein, the first half of solution vector represent the machine choice information of all operations, latter half represent the sequencing information of all operations.

Further, described step 2) solution vector in harmony search algorithm is converted to the coding of FJSS feasible solution, transfer algorithm adopts different switching strategies respectively for machine choice part and operation sequencing part, is specially: 2.1) in machine choice part, specific practice: first by [-1,1] the linear transformed mappings of the real number in the real number in [1, m], and then gets immediate integer, to any x ∈ [-1,1], the integer z in corresponding [1, m] is:

$z=round(\frac{m-1}{2}(x+1)+1)$

Wherein, round () is for getting an immediate integer of real number, and special circumstances are as m=1, to any x ∈ [-1,1], and z=1; 2.2) in operation sequencing part, the conversion method based on LPV rule is used.

Further, described step 5) using the optimum solution that finds based on the harmonic search algorithm initial solution as large field searching algorithm, large field searching algorithm is adopted to proceed search, until end condition meets, the solution obtained is the scheduling of completion date, and detailed process is: 5.1) set up the model based on constraint; 5.2) destroy step, namely disturbance is carried out to current solution, in subsequent builds step, generate new more excellent solution; 5.3) construction step, namely utilizes constraint plan search on the basis of current solution, form new scheduling solution.

The present invention is owing to taking above technical scheme, it has the following advantages: 1, the present invention is directed to completion date is that the FJSP of optimization aim is integrated with two algorithms, namely harmony search algorithm and large neighborhood search is mixed, HHS a kind ofly has the evolution algorithmic of mould because of algorithm demeanour, LNS is then a kind of typically based on the method for constraint, in order to form a more effective search mechanisms, first the present invention runs HHS, then operation LNS improves the last solution that HHS obtains further, before operation LNS, elite's solution that this integration search method also utilizes HHS to obtain, extract effective machine assignment information, the solution space of further reduction problem, therefore HHS and LNS is made to have complementary advantages, thus effectively solve extensive Flexible Job-shop Scheduling Problems.2, the present invention utilizes the integration search method proposed can complete the setting of flexible job shop system job number, machine number, basic parameter such as operation number, each operation available machines used situation etc., but also can according to different device contexts and process requirements, configure various machining path for each operating flexibility, dispatching effect is better than current some existing algorithms.The present invention can be widely used in extensive scheduling involved in the commercial production such as semiconductor production, automobile assembling, weaving.

Accompanying drawing explanation

Fig. 1 is the LNS principle schematic of prior art;

Fig. 2 is operation sequencing part translation example of the present invention.

Embodiment

Below in conjunction with accompanying drawing, detailed description is carried out to the present invention.But should be appreciated that being provided only of accompanying drawing understands the present invention better, they not should be understood to limitation of the present invention.

FJSP (FlexibleJobShopScheduling, FJSP) can formalized description as follows: have one group of n independently operation J={J mutually ₁, J ₂..., J _n, one group of m platform machine M={M ₁, M ₂..., M _m.Each operation J _icomprise one group of sequence of operation having priority to retrain operation J _iits all operations is complete with appointment order to be done that and if only if, can be expressed as each operation O _i,j, i.e. operation J _ij operation, can in subset in any machine on perform.The execution time of each operation depends on machine, adopts p _{i, j, k}represent operation O _i,jat machine M _kon processing time.Scheduling is divided into two subproblems: route subproblem, is assigned on a suitable machine by each operation; Sequence subproblem, namely determines the sequence that all machines operate, and the target solved finds one to dispatch solution to minimize completion date.Completion date refers to the time having needed All Jobs, can be defined as C _max=max _1≤i≤n{ C _i, wherein, C _ioperation J _ideadline.In addition, following hypothesis can be done: all machines are all available in 0 moment; All Jobs is all in the release of 0 moment; Every platform machine once can only process an operation; Each operation, once execution, must interruptedly not complete; The order that each work operates in the industry is that predefined is good and can not be modified; The setup time of machine and the transmission time of operation negligible.

In order to clearly describe this problem, table 1 gives the sample of a FJSP problem, and wherein, row correspond to operation, and row correspond to machine.Each entry in form represents each processing time operated on corresponding machine, and in this form, mark " – " expression operation can not run on corresponding machine.

Table 1FJSP sample treatment schedule

Integration search method for extensive flexible job shop scheduling based provided by the invention, comprises following content:

1, the solution vector that certain will solve FJSP is defined.

In harmonic search algorithm, solution vector can be expressed as it is the real number vector of a n dimension, and the variable range conveniently setting every one dimension is all [-1,1].The dimension n of solution vector is the twice of solved FJSP scheduling problem operation number summation.Wherein, the first half of solution vector represent the machine choice information of all operations, and latter half represent the sequencing information of all operations.

2, the solution vector in harmony search algorithm is converted to the coding of FJSP feasible solution, i.e. two vector codings, the present invention adopts different switching strategies respectively for machine choice part and operation sequencing part, is specially:

1) in machine choice part, during solution vector represents, solution vector is often tieed up variable range and be all defined as [-1,1], and the scope of machine choice is the integer of 1 to m in feasible solution coding, wherein m is the optional machine number of the corresponding operation in this position.Therefore need the integer real number in [-1,1] being mapped as 1 to m, specific practice be first by the linear transformed mappings of real number in [-1,1] to the real number in [1, m], and then get immediate integer.So to any x ∈ [-1,1], the integer z in corresponding [1, m] is:

$z=round(\frac{m-1}{2}(x+1)+1)---\left(7\right)$

In formula, round () is for getting an immediate integer of real number, and special circumstances are as m=1, to any x ∈ [-1,1], and z=1.

2) in operation sequencing part, the conversion method based on LPV (maximum position) rule is used.

First the integer ID number that each operation imparting one is fixing is given.Each Action number is given successively by the order operated in workpiece and workpiece, for table 1, O ₁₁, O ₁₂, O ₂₁, O ₂₂, O ₂₃fixing No. ID be respectively 1,2,3,4,5.Solution vector is also [-1 in the often dimension of operation sequencing part, 1] real number in, first LPV rule is utilized the operation sequencing part of solution vector to be converted to an arrangement of No. ID, operation, specific practice is by the solution vector of operation sequencing part together with location number, arranges with non-decreasing order by the value often tieed up.The arrangement that location number is originally formed is just as an arrangement of No. ID, operation.This arrangement is equivalent to arbitrary arrangement of all operations, likely corresponding infeasible solution, so be converted to the workpiece number at place by each operation No. ID again, the solution vector of such operation sequencing part is just converted into the operation sequencing part of FJSP feasible solution coding.For the FJSP problem of table 1, then the signal of concrete operations as shown in Figure 2.

3, adopt mixing harmonic search algorithm to carry out harmony search algorithm, i.e. first some solution vectors of random initializtion, setting maximum iteration time, per in generation, produces a new solution vector by mixing harmony search algorithm operator.

4, Local Search is embedded in the algorithm of mixing harmony search algorithm and more careful search is carried out to solution space, the solution vector being about to mixing harmony search algorithm is converted into the scheduling solution of FJSP problem, neighborhood based on setting carries out Local Search, scheduling solution after Local Search resulting improvement is converted into solution vector again, the solution vector before Local Search is replaced by this solution vector, the search procedure continuing harmony search algorithm obtains optimum solution, and specific implementation process is:

Current major part is all based on critical path about the neighborhood of FJSP scheduling problem, and major part reaches destruction critical path based on the one or more operations in mobile critical path, reduces the object of completion date.The present invention proposes a FJSP scheduling problem neighborhood based on public key operation, and wherein, public key operation is exactly the operation in all critical paths.Because can find that a lot of of FJSP problem separate the corresponding more than critical path of extracting figure, if the operation in the critical path of mobile not common key operation, then this operation is not at least in a certain bar critical path, so this operation mobile can not destroy this critical path, the completion date of gained solution also can not reduce.

Neighbour structure specific design based on public key operation is as follows: read public key operation successively, first this operation is deleted from extracting figure each operation, recalculates the earliest start time of each operation in each extracting figure; Then be current completion date with the completion date before deletion action, calculate the late start time of each operation from back to front; Whether the time slot formed between the last operation of examination successively and operation, see and can insert in suitable gap by operation, if can insert, current solution is exactly the neighborhood produced, and exits.Otherwise the operation of reading in next critical path, then do identical process.

5, using based on the initial solution of the optimum solution that finds of mixing harmony search algorithm as LNS, adopt LNS to proceed to search for and improve this solution further, until meet end condition, the solution obtained is the scheduling of minimum completion date.

LNS of the present invention realizes on the basis of COMET system, in COMET system, use LNS, needs first to carry out modeling by setting up constraint to problem to be solved; In optimizing process, extra constraint will constantly be added; Once produce a new constraint, the constraint planning mechanism of transmission provided in COMET internal system will be triggered, and this means that current all constraints will be used to participate in filtering disaggregation one by one, until in the territory of separating more value can be removed.In addition, in search procedure, each once find a solution, dynamically will increase a constraint in system, this constraint limits the solution next time found and must be better than the current solution found.To the LNS search procedure solving FJSP be structured in COMET system be described in detail below:

1) model based on constraint is set up.

In order to set forth the model based on constraint for FJSP, the present invention defines two mark σ _ij, μ _ijrepresent operation O respectively _ijstart time in scheduling and the machine of selection.So, a solution of FJSP can use (the σ of all operations _ij, μ _ij) value is to representing, this solution is that feasible and if only if that it meets following three kinds of constraints:

1. priority constraint: the operation done in the industry at, must meet the priority of specifying, it can be expressed as formula formally: represent operation O _ijat machine μ _ijon process time.

2. resource constraint: any one operation can only be processed on its available machine, namely

3. capacity-constrained: machine can only single treatment one operation, form can turn to following constraint: if μ _x,y=μ _{α, β}, so ${\mathrm{\σ}}_{x,y}+p_{x,y,{\mathrm{\μ}}_{xy}}\≤{\mathrm{\σ}}_{\mathrm{\α},\mathrm{\β}}$ Or ${\mathrm{\σ}}_{\mathrm{\α},\mathrm{\β}}+p_{\mathrm{\α},\mathrm{\β},{\mathrm{\μ}}_{\mathrm{\α},\mathrm{\β}}}\≤{\mathrm{\σ}}_{\mathrm{\α},\mathrm{\β}},$ Wherein, α, β, x, y are all subscripts of operation, be equivalent, just represent different operations with the implication of i, j.

FJSP problem is exactly minimize completion date on the basis of as above three constraints:

$C_{max}={\max }_{1\&le;i\&le;n,1<j\&le;n_i}\{{\mathrm{\&sigma;}}_{i,j}+p_{i,j,{\mathrm{\&mu;}}_{ij}}\}.$

2) destroy step, that is: disturbance is carried out to current solution, in subsequent builds step, generate new more excellent solution.

In destruction step, select some variablees to relax, other variable remains unchanged simultaneously.For FJSP, here adopt partial order scheduling slack (POS), first select one group of operation, represent with Ω, then remaining each operation is fixed on its current machine, and the operation on same machine keeps their relative priority orders in current solution.If (σ, μ) is current scheduling solution, (σ ', μ ') and be the solution needing to build.

POS lax can formalization representation as follows: and μ _x,y=μ _{α, β}if, so wherein, σ ' refers to the start time set of all operations in the new solution built, σ ' _{α, β}then concrete expression operates O _{α, β}start time in new structure solution, and μ ' _x,y=μ _x,y, μ ' _{α, β}=μ _{α, β}.POS is relaxed, only keeps the relative priority order between remaining operation, instead of the real start time, benefiting very much, more space can be left for again optimizing.If Υ is the subset of Ω, central each operation is fixed on original machine, so in LNS, how to select Υ to be called that neighborhood inspires.Time window neighborhood is adopted to inspire in the present invention, each generation time window [t _min, t _max], so Υ represents at time window [t _min, t _max] between process one group of operation.In order to further enhanced search, another one neighbour structure is built by time window neighbour structure and obtains, and it is such subset and the machine assignment operated in Υ is fixing, namely

3) construction step, namely utilizes constraint plan search on the basis of current solution, to form new scheduling solution, namely minimum completion date.

The summation of all operations time period on the longest path that minimum completion date equals precedence constraints chart corresponding to dispatching office (disconnected figure).In simple terms, search performs as follows: operation be assigned on the machine of specifying, first consider that those have the operation of less machine choice; Then on each machine, operation is sorted respectively, pay the utmost attention to the machine that those have less free time; After ordered steps, according to precedence constraints chart, calculate each operation the earliest with Late Start and completion date; Finally the earliest start time of each operation is set to its start time.In order to avoid the search of long time, in the construction step that LNS is each, set the maximum frequency of failure.In addition, the search controller of LNS adopts depth-first search, and because search realizes in the system based on constraint, it can inherit the advantage of the constraint propagation in constraint planning well.

In sum, the present invention is integrated with the feature of two aspects: the optimum solution that the first, HHS finds is that LNS provides a good starting point; The second, the search of LNS is limited in a region more likely by the information of the machine choice extracted, and can strengthen the ability that LNS concentrates search.Consider the complicacy of two existing algoritic modules (HHS and LNS), algorithm of the present invention is relatively simple, first HHS algorithm is performed, owing to mixing the Exchange rings of harmony search algorithm, namely each iteration all replaces harmony the poorest in harmony data base, and so harmony data base (reaching maximum iteration time) after HHS execution terminates can be counted as an elite Xie Chi.Because in FJSP, most important to the machine assignment of operation, so before entering into LNS search, some excellent machine assignment information can be extracted from elite's solution of these harmony data bases.The step of information extraction is as follows: for each operation, and that machine selected by harmony best in harmony data base will be added in the available machines used list of this machine; In remaining machine, selected that maximum machines by other elite's solution, if the frequency selected is not less than τ, τ is adjustable parameter, so also will be added in alternative machines list simultaneously.So the extraction of machine choice information makes each operation have the alternative machines set of a yojan.

The various embodiments described above are only for illustration of the present invention, and wherein each step etc. of method all can change to some extent, and every equivalents of carrying out on the basis of technical solution of the present invention and improvement, all should not get rid of outside protection scope of the present invention.

Claims (5)

1., for an integration search method for extensive flexible job shop scheduling based, it is characterized in that comprising the following steps:

1) solution vector that certain will solve FJSP is defined;

2) solution vector in mixing harmony search algorithm is converted to the coding of FJSP feasible solution;

3) adopt mixing harmonic search algorithm to carry out harmony search algorithm, i.e. first some solution vectors of random initializtion, setting maximum iteration time, per in generation, produces a new solution vector by mixing harmony search algorithm operator;

4) solution vector of mixing harmony search algorithm is converted into the scheduling solution of FJSP problem, Local Search is carried out in field according to arranging, and the scheduling solution after Local Search resulting improvement is converted into solution vector again, solution vector before utilizing this solution vector to replace Local Search, the search procedure continuing harmony search algorithm obtains optimum solution;

5) using based on the initial solution of the optimum solution that finds of mixing harmonic search algorithm as large field searching algorithm, adopt large field searching algorithm to proceed search, until meet end condition, obtain up-to-date scheduling solution and completion date.

2., as claimed in claim 1 for the integration search method of extensive flexible job shop scheduling based, it is characterized in that: described step 1) define the solution vector that certain will solve FJSS, be specially: solution vector is expressed as n representation dimension amount, the variable range of every one dimension is all [-1,1], and the dimension n of solution vector is the twice of solved FJSS scheduling problem operation number summation, wherein, the first half of solution vector represent the machine choice information of all operations, latter half represent the sequencing information of all operations.

3. as claimed in claim 1 a kind of based on mixing harmony search algorithm flexible job shop scheduling based method, it is characterized in that: described step 2) solution vector in harmony search algorithm is converted to the coding of FJSS feasible solution, transfer algorithm adopts different switching strategies respectively for machine choice part and operation sequencing part, is specially:

2.1) in machine choice part, specific practice: first by the linear transformed mappings of real number in [-1,1] to [1, m] in real number, and then get immediate integer, to any x ∈ [-1,1], the integer z in corresponding [1, m] is:

$z=round\left(\frac{m-1}{2}\right(x+1)+1)$

Wherein, round () is for getting an immediate integer of real number, and special circumstances are as m=1, to any x ∈ [-1,1], and z=1;

2.2) in operation sequencing part, the conversion method based on LPV rule is used.

4. as claimed in claim 2 a kind of based on mixing harmony search algorithm flexible job shop scheduling based method, it is characterized in that: described step 2) solution vector in harmony search algorithm is converted to the coding of FJSS feasible solution, transfer algorithm adopts different switching strategies respectively for machine choice part and operation sequencing part, is specially:

2.1) in machine choice part, specific practice: first by the linear transformed mappings of real number in [-1,1] to [1, m] in real number, and then get immediate integer, to any x ∈ [-1,1], the integer z in corresponding [1, m] is:

$z=round\left(\frac{m-1}{2}\right(x+1)+1)$

Wherein, round () is for getting an immediate integer of real number, and special circumstances are as m=1, to any x ∈ [-1,1], and z=1;

2.2) in operation sequencing part, the conversion method based on LPV rule is used.

5. a kind of flexible job shop scheduling based method based on mixing harmony search algorithm as described in any one of Claims 1 to 4, it is characterized in that: described step 5) using the optimum solution that finds based on the harmonic search algorithm initial solution as large field searching algorithm, large field searching algorithm is adopted to proceed search, until end condition meets, the solution obtained is the scheduling of completion date, and detailed process is:

5.1) model based on constraint is set up;

5.2) destroy step, namely disturbance is carried out to current solution, in subsequent builds step, generate new more excellent solution;

5.3) construction step, namely utilizes constraint plan search on the basis of current solution, form new scheduling solution.