CN101794115B - Scheduling rule intelligent excavating method based on rule parameter global coordination optimization - Google Patents

Abstract

The invention relates to a scheduling rule intelligent excavating method based on rule parameter global coordination optimization, belonging to the automatic control, information technology and advanced manufacturing field, in particular relates to a complex production process oriented scheduling rule intelligent excavating method for scheduling environment in real time. The invention is characterized in that the method includes the following steps: a complex production process oriented scheduling rule intelligent excavating frame for scheduling environment in real time is built, a scheduling problem instance classification model is build, and scheduling rule parameter global coordination optimization problem is constructed and solved. The invention is based on the scheduling rule intelligent excavating frame provided by the invention and adopts double-layer fuzzy C-means clustering method to classify scheduling problem instances. Rule parameter global coordination optimization problem is constructed directing at scheduling problem instance in each class and linear partition based particle swarm optimization is adopted to solve and optimize the problem, wherein Bayes estimation method is adopted to carry out comprehensive evaluation on scheduling rule performance. The obtained scheduling rule has better scheduling effect on different problem instances in similar scheduling environment.

Description

A kind of scheduling rule intelligent excavating method of rule-based parameter global coordination optimization

Technical field

The invention belongs to automatic control, infotech and advanced manufacturing field.Be specifically related to a kind of scheduling rule intelligent excavating method towards complex process Real-Time Scheduling environment.

Background technology

The production run scheduling problem is the focus of academia and industry member research always.In the past few decades, people have proposed much to find the solution the method for production run scheduling problem, as based on the method for operational research, based on didactic method, based on soft Calculation Method etc.Have the knowledge that can incorporate people's (experienced scheduler), express characteristics such as simple, that calculated amount is little based on the method for scheduling rule, in the Real-Time Scheduling environment, have remarkable advantages.But to the complex process scheduling problem, be difficult to obtain both had better scheduling performance according to artificial experience or scheduling expertise merely the scheduling rule that has than strong adaptability is arranged, the scheduling rule method for digging of the facing production course scheduling problem that proposes is mainly based on the local data (local dispatch environment parameter or local scheduling decision parameter) of scheduling problem example in the different scheduling decisions moment, and only carry out scheduling rule performance evaluation at the incidence relation between the part property value before and after the scheduling rule, mining process is not considered the coupling between the corresponding scheduling rule between the different machines group, do not consider the requirement of scheduling problem optimization aim yet, thereby exist overall performance undesirable based on the scheduling rule that this obtained, be difficult to be adapted to the defective of different dispatch environment.

Summary of the invention

For overcoming the defective of above-mentioned production run scheduling rule method for digging, the present invention proposes a kind of intelligent excavating method based on the scheduling rule parameter global coordination optimization towards complex process Real-Time Scheduling environment.The present invention at first sets up towards the scheduling rule intelligent excavating framework of complex process Real-Time Scheduling environment, and its core is that the scheduling rule mining process is converted into scheduling rule parameter global coordination optimization process.This framework comprises structure scheduling problem case library, the division of scheduling problem example, makes up scheduling problem case library of the same type, makes up and find the solution the rule parameter global coordination optimization problem at every class problem-instance.On above-mentioned scheduling rule intelligent excavating frame foundation, the present invention adopts double-deck fuzzy C-means clustering method that different scheduling problem examples is classified.To every class scheduling problem example, the rule parameter global coordination optimization method that employing is divided PSO (particle cluster algorithm) based on linearity is optimized parameter of regularity, performance when wherein, adopting the Bayesian Estimation method that given scheduling rule is acted on a plurality of similar scheduling problem example is carried out comprehensive evaluation.Finally excavate the scheduling rule that is applicable to every class problem-instance.Thereby the scheduling rule that makes the present invention excavate is more targeted, dispatching effect is better.

The scheduling rule intelligent excavating method of rule-based parameter global coordination optimization is characterized in that, described method is excavated on computing machine and the production scheduling information collecting device in scheduling rule and realized according to the following steps successively:

The first step: definition scheduling problem example

The scheduling problem example is designated as Problem, and it can be defined as follows:

Existing N workpiece formed workpiece set J={1 ..., N}.M machine group formed machine group set M={M ₁, M ₂..., M _m, machine group M wherein _iBy N (M _i) the identical parallel machine of platform forms.S manufacturing procedure S={S arranged ₁, S ₂..., S _s, operation S wherein _iThe machine group set that is comprised is MS (S _i).The release of workpiece i is R constantly _i, the deadline is C _i, be D delivery date _i, its process is by n _iIndividual operation O _I1, O _I2...,

Form operation O _IjCan

Process on arbitrary machine in the machine group, be p (O its process time _Ij), operation O _IjDirect subsequent operation set next (O in processing route _Ij) expression, directly preceding continuous operational set prev (O _Ij) expression.A (O _Ij), b (O _Ij) and c (O _Ij) be respectively and operate O _IjTime of arrival, processing start time and process finishing time.

Process satisfies following constraint:

● can not interrupt constraint: operation Once you begin processing just can not stop, until machining;

● the processing route constraint:

Each operation that is workpiece must be processed by predetermined processing route requirement.

Satisfying under the condition of above-mentioned constraint, the regulation goal that the present invention sets has two kinds:

● minimize and drag issue:

$\min \left\{{\mathrm{\Σ}}_{i=1}^N\mathrm{sign}(C_i-D_i)\right\}$

Wherein, drag issue to be meant in a scheduling problem example deadline C _iBe later than D at delivery date _iThe workpiece sum, can be expressed as

$\mathrm{sign}(C_i-D_i)=\left\{\begin{array}{c}1,C_i-D_i>0\\ 0,C_i-D_i\≤0\end{array}\right.;$

● minimize the manufacturing cycle:

min{max{C _i|i＝1，2，…，N}}

Wherein the manufacturing cycle is meant in a scheduling problem example, processes the used time of all workpiece, can be expressed as max{C _i| i=1,2 ..., N}.

Second step: scheduling rule intelligent excavating software is installed on described rule digging computing machine

Scheduling rule intelligent excavating software is realized by the scheduling rule intelligent excavating framework towards complex process Real-Time Scheduling environment.The core of this framework is that the scheduling rule mining process is converted into scheduling rule parameter global coordination optimization process.This framework comprises structure scheduling problem case library, divides the scheduling problem example, makes up scheduling problem case library of the same type, makes up and find the solution the rule parameter global coordination optimization problem at every class problem-instance.The scheduling rule intelligent excavating framework is at first according to the actual production data, make up the scheduling problem example, and the scheduling problem example is kept in the scheduling problem case library, afterwards the scheduling problem example types is divided and produced to the relevant scheduling problem example in the scheduling problem case library, to the relevant scheduling problem example of every class, produce a plurality of scheduling problem examples of the same type and make up scheduling rule parameter global coordination optimization problem, on this basis, adopt the rule parameter global coordination optimization algorithm to find the solution above-mentioned parameter of regularity optimization problem, in each parameter iteration optimizing process, the scheduling rule that parameter is given acts on above-mentioned a plurality of scheduling problem examples of the same type and carries out industrial process simulation, obtain corresponding a plurality of scheduling index, and adopt the scheduling rule integrated evaluating method that the current scheduling rule is estimated, the evaluation of estimate that is obtained is as the target function value of rule parameter global coordination optimization problem, final scheduling rule after obtaining to optimize.Excavate framework according to above-mentioned scheduling rule, scheduling rule intelligent excavating software comprises structure scheduling problem case library module, scheduling problem example division module, makes up scheduling problem case library module of the same type, makes up scheduling rule parameter global coordination optimization problem module, finds the solution scheduling rule parameter global coordination optimization problem module.

The 3rd step: produce real-time information collection

Gather the production real-time information with the production scheduling information collecting device, comprise and produce workpiece information, manufacturing schedule information, facility information, and above-mentioned information is sent to scheduling rule by netting twine excavates computing machine, wherein, producing workpiece information comprises and respectively operates process time, workpiece release time, workpiece delivery date, workpiece processing route information in workpiece quantity, the workpiece; Manufacturing schedule information comprises on-stream time, concluding time, the time out that operates on the machine; Facility information comprises failure message, maintenance information.

The 4th step: make up the scheduling problem case library

In described scheduling rule excavation computing machine, call and make up scheduling problem case library module, set up the scheduling problem case library.Go on foot the production real-time information of gathering according to the first step defined scheduling problem example implication and the 3rd in particular moment, produce the scheduling problem example of particular moment, the particular moment of being adopted among the present invention is 8:00,12:00, the 16:00 of every day, can produce 3 scheduling problem examples like this every day, the scheduling problem example that will set up every day is stored in the scheduling problem case library successively.Press the said process certain time, the time that continues among the present invention is 3 months.

The 5th step: divide the scheduling problem example

In described scheduling rule excavation computing machine, call and divide the scheduling problem example module, the scheduling problem example that is kept in the scheduling problem case library is classified.In this module, adopt scheduling problem example division methods, to being stored in K scheduling problem example { Problem in the scheduling problem case library based on double-deck fuzzy C-means clustering ₁..., Problem _KDivide, the setting according to the 4th step has 270 scheduling problem examples, i.e. K=270 in the case library.In the method, lower floor's cluster is responsible for the information relevant with the scheduling problem example is carried out cluster, extracts characteristic information; The characteristic information that the upper strata cluster is responsible for extracting carries out cluster, to be used for the division of scheduling problem example.This division scheduling problem example module is divided the scheduling problem-instance as follows:

The 5.1st step: realize the fuzzy C-means clustering algorithm

In the levels cluster, all need use following fuzzy C-means clustering method: suppose that sample set is X={x ₁, x ₂..., x _n, be divided into C ambiguity group, and ask every group cluster centre c _j, j=1 ..., C makes following defined desired value J _CReach minimum:

$J_C=\underset{j=1}{\overset{C}{\mathrm{\Σ}}}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\μ}}_{\mathrm{ij}}^{\mathrm{\α}}{\left|\right|x_i-c_j\left|\right|}^2,$ 1≤α≤∞

And need to satisfy: $\underset{j=1}{\overset{C}{\mathrm{\Σ}}}{\mathrm{\μ}}_{\mathrm{ij}}=1,$ $\∀i=1,\·\·\·,n$

Wherein, μ _IjI data point of ∈ [0,1] expression belongs to the degree of membership of j cluster centre, c _jBe j cluster centre, α is a weighted index.Fuzzy membership μ _IjAnd c _jCan obtain with following formula respectively:

${\mathrm{\μ}}_{\mathrm{ij}}=1/\underset{k=1}{\overset{C}{\mathrm{\Σ}}}{\left(\frac{{\left|\right|x_i-c_i\left|\right|}^2}{{\left|\right|x_i-c_k\left|\right|}^2}\right)}^{2/(\mathrm{\α}-1)}$

$c_j=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\μ}}_{\mathrm{ij}}^m{x}_i/\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\μ}}_{\mathrm{ij}}^m$

The 5.2nd step: lower floor's cluster, it extracts the scheduling problem characteristic information from the various information of scheduling problem

Lower floor's cluster is responsible for extracting the scheduling problem characteristic information from the various information of scheduling problem, and it is realized as follows:

The 5.2.1 step: extract the workpiece technology characteristics

Adopt following ternary representation to represent the technology characteristics of workpiece i:

Process _i＝{ML _i，MF _i，AF _i}

Wherein:

● ML _iBe the longest path in the processing route of workpiece i, note arrives operation O _IjThe farm labourer at the place path of planting is ML (O _Ij), ML (O then _Ij) can obtain by the following formula iteration:

And have

● MF _iBe the maximum branch amount in the processing route of workpiece i, note operation O _IjThe maximum branch amount at place is MF (O _Ij), MF (O then _Ij) can obtain by the following formula iteration:

And MF is arranged _i=max (MF (O _Ij) | j=1 ..., n _i)

● AF _iBe the average of branch amount before and after each operation of workpiece i, that is:

$A{F}_i=\frac{1}{n_i}\underset{j=1}{\overset{n_i}{\mathrm{\Σ}}}\left(\right|\mathrm{prev}\left(O_{\mathrm{ij}}\right)|+|\mathrm{next}\left(O_{\mathrm{ij}}\right)\left|\right)$

Calculate { Problem successively ₁..., Problem _KThe Process value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Process value is carried out cluster.Note cluster centre quantity is C ^p, the Process cluster centre of workpiece is

J=1,2 ..., C ^p, the Process of workpiece i _iThe degree of membership that belongs to each center is

J=1 ..., C ^pThe workpiece technology characteristics vector of definition scheduling problem example is:

[PP(1)，…，PP(C ^p)]

Wherein

J=1 ..., C ^p, belong to the average degree of membership value of j Process cluster centre for the Process of workpiece in this scheduling problem example.

The 5.2.2 step: extract the workpiece processing temporal characteristics

The present invention adopts theoretical process time of the Makespan of workpiece to characterize feature process time, and this Makespan represents that a workpiece finishes under the situation about waiting for the required shortest time of processing not having.Use Makespan the theoretical process time of workpiece i _iExpression, it can obtain by the following formula iterative computation:

Calculate { Problem successively ₁..., Problem _KThe Makespan value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Makespan value is carried out cluster.Note cluster centre quantity is C ^m, the cluster centre of Makespan is respectively

J=1,2 ..., C ^m, the Makespan of workpiece i _iThe degree of membership that belongs to each center is designated as

J=1 ..., C ^mThe workpiece processing temporal characteristics vector of definition scheduling problem example is:

[PM(1)，…，PM(C ^m)]

Wherein

J=1 ..., C ^m, belong to the average degree of membership value of j Makespan cluster centre for the Makespan of workpiece in this scheduling problem example.

The 5.2.3 step: extract workpiece and hand over phase tightness feature

The processing tightness that the present invention characterizes a workpiece with the process time and the difference between the friendship phase of a workpiece, with the friendship phase tightness of following formula definition workpiece i:

Slack _i＝D _i-Makespan _i

Calculate { Problem successively ₁..., Problem _KThe Slack value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Slack value is carried out cluster.Note cluster centre quantity is C ^d, hand over phase tightness Slack cluster centre to be

J=1,2 ..., C ^d, the Slack of workpiece i _iThe degree of membership that belongs to each center is

J=1 ..., C ^dThe workpiece of definition scheduling problem example hands over phase tightness proper vector to be:

[PD(1)，PD(2)，…，PD(C ^d)]

Wherein

J=1 ..., C ^d, belong to the average degree of membership value of j Slack cluster centre for the Slack of workpiece in this scheduling problem example.

The 5.2.4 step: extract the machine process capacity characteristic

The present invention represents with the contained machine quantity of each operation in the scheduling problem example, that is:

Capability＝{SC ₁，…，SC _s}

Wherein I=1 ..., s is the total quantity of i machine that operation comprises in this scheduling problem example.

Calculate { Problem successively ₁..., Problem _KThe Capability of all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Capability value is carried out cluster.Note cluster centre quantity is C ^c, machinery processing capacity Capability cluster centre is designated as

J=1,2 ..., C ^c, the Capability of k scheduling problem example _kThe degree of membership that belongs to each center is designated as J=1 ..., C ^cThe machinery processing capacity proper vector of scheduling problem example correspondence is:

[PC(1)，PC(2)，…，PC(C ^c)]

Wherein $\mathrm{PC}\left(j\right)={\mathrm{\μ}}_{\mathrm{kj}}^c,$ j＝1，…，C ^c

The 5.3rd step: the upper strata cluster, it divides the scheduling problem example according to the scheduling problem characteristic information

In the clustering method of upper strata, the technology characteristics vector that the present invention obtained for the 5.2nd step, workpiece processing temporal characteristics vector, workpiece hand over phase tightness proper vector, machinery processing capacity proper vector to merge, total characteristic vector as a scheduling problem example is designated as:

X=[PP (1) ..., PP (C ^p), PM (1) ..., PM (C ^m), PD (1) ..., PD (C ^d), PC (1) ..., PC (C ^c)] with above-mentioned vector as sample, then { a Problem in the clustering algorithm ₁..., Problem _KTotal K sample, use the fuzzy C-means clustering method to carry out cluster to these samples, note cluster centre quantity is C ^r, then resulting scheduling problem example cluster centre is

J=1,2 ..., C ^r, the degree of membership that k scheduling problem example belongs to each scheduling problem example cluster centre is designated as J=1 ..., C ^rThe scheduling problem example is divided into C ^rClass makes in k scheduling problem example

Maximum classification is the affiliated type of this scheduling problem example, in case study on implementation of the present invention, the scheduling problem example is divided into 14 scheduling problem example types.

The 6th step: make up scheduling problem case library of the same type

In described scheduling rule excavation computing machine, call and make up similar scheduling problem case library module, make up scheduling problem case library of the same type.Because the scheduling problem example quantity in the scheduling problem case library is limited after all, and the scheduling rule excavation needs a large amount of scheduling problem examples in order to estimate the quality of the scheduling rule of being excavated, so need initiatively produce the scheduling problem example of the same type of sufficient amount at each scheduling problem example types.The present invention is directed to each scheduling problem example types, produce 100 scheduling problem examples of the same type, its method is as follows: divide the result according to the scheduling problem that the 5th step was determined, at the scheduling problem example { Problem of scheduling problem case library preservation ₁..., Problem _KIn, choose a scheduling problem example that belongs to current scheduling problem-instance type wantonly, and the correlated characteristic information of this scheduling problem example is carried out random fluctuation, to produce a plurality of different scheduling problem examples of the same type.(α β) is uniform random number in [α, β] interval to note randu, and the present invention gets α, and β is respectively 0.9 and 1.1.Need the information of change to have:

● workpiece quantity: the workpiece quantity after the variation is

● operate process time: the operation after the variation is process time

● workpiece delivery date: the workpiece after the variation is delivery date

● comprise number of machines in each machine group: comprising number of machines in each the machine group after the variation is $\tilde{N}\left(M_i\right)=N\left(M_i\right)\×p\left(\mathrm{0.9,1.1}\right)$

After the information change of above-mentioned scheduling problem example, promptly produced a new scheduling problem example, this example has been saved in the current scheduling problem-instance type of scheduling problem case library of the same type.

According to said method, can produce a plurality of different scheduling problem examples to each scheduling problem type, these scheduling problem examples all are saved in the scheduling problem case library of the same type.

The 7th step: make up scheduling rule parameter global coordination optimization problem

In described scheduling rule excavation computing machine, call and make up scheduling rule parameter global coordination optimization problem module.In this module, scheduling rule is defined as Rule={RS, RP}, wherein RS is the scheduling rule structure, RP is the scheduling rule parameter sets under specific RS.Scheduling rule is input as operational attribute and machine group attribute two category informations, and is output as the processing priority of workpiece, wherein:

● the processing priority of workpiece is high more, and workpiece just should be processed more in advance.

● operational attribute has: process time, time of arrival, prediction time of arrival, haulage time, delivery date, residue process time, residue processing step number, subsequent operation number, subsequent handling bottleneck degree.

● machine group attribute has: comprise operation number to be processed in number of machines, machine group working ability, machine group bottleneck degree, the machine group, interior operation summation process time to be processed of machine group, estimate to arrive summation process time of operating;

The present invention adopts Adaptive Neuro-fuzzy Inference (ANFIS:Adaptive Neuro-Network Fuzzy InferenceSystem) in order to the expression scheduling rule, be above-mentioned workpiece attribute of being input as of ANFIS and machine group attribute, be output as workpiece processing priority.Then the structure RS of scheduling rule is ANFIS, its concrete structure as shown in Figure 4, scheduling rule parameters R P is the parameter among the ANFIS, is designated as RP={rp ₁, rp ₂..., rp _ZExpression, wherein Z is the number of parameter among the ANFIS.On this basis, constructed scheduling rule parameter global coordination optimization problem is: coordination optimization RP={rp ₁, rp ₂..., rp _Z, so that by RP={rp ₁, rp ₂..., rp _ZPerformance the best of the scheduling rule determined.

Adopt said method,, make up corresponding scheduling rule parameter global coordination optimization problem respectively at each scheduling problem example types.

The 8th step: find the solution scheduling rule parameter global coordination optimization problem

In described scheduling rule excavation computing machine, call and find the solution scheduling rule parameter global coordination optimization problem module, the pairing scheduling rule parameter global coordination optimization of each scheduling problem example types problem is found the solution.This module adopts based on linearity division PSO (particle group optimizing) method each scheduling rule parameter global coordination optimization problem is found the solution, to obtain the scheduling rule corresponding with each scheduling problem example types.This method realizes as follows:

The 8.1st step: for reducing the number of parameter to be optimized, with parameter of regularity RP={rp to be optimized ₁, rp ₂..., rp _ZExpress with the slope and the displacement of a plurality of different line segments.At first the parameter among the RP is pressed parameter value scope [min (rp separately _i), max (rp _i)] normalization, i.e. rp ' _i=(rp _i-min (rp _i))/(max (rp _i)-min (rp _i)), i=1 ..., Z, obtain RP '=rp ' ₁, rp ' ₂..., rp ' _Z, in the present invention, the parameter value scope of ANFIS is decided to be [1000,1000], i.e. min (rp _i)=-1000, max (rp _i)=1000.Simultaneously, suppose that slope and displacement with Q bar line segment are k _j, b _j, j=1,2 ..., Q represents the parameter among the RP ', with Q=10 bar line segment, the denotable parameter set of j bar line segment is designated as Ie among the present invention _j, the division of above-mentioned parameter need be satisfied Ie ₁∪ Ie ₂∪ ... ∪ Ie _Q=RP ' and

I ≠ j, i=1 ..., Q, j=1 ..., Q.Then the relation of parameter among the RP ' and Q bar line segment can be represented by the formula:

rp′ _(i)＝k _j×(i)+b _j-[k _j×(i)+b _j] ⁺， $r{p}_{\left(i\right)}^{\′}\⋐S_j,$ j＝1，…，Q

The 8.2nd step: adopt the PSO method to k _j, b _j, j=1,2 ..., Q is optimized.With k _jAnd b _jAs decision variable, then each particle position is X={k in the PSO method _j, b _j| j=1,2 ..., Q} also can be expressed as X={x ₁, x ₂..., x _2Q, each particle's velocity is V={v ₁, v ₂..., v _2Q.Optimize in the iteration each, each particle upgrades by following formula:

$v_{\mathrm{id}}^{k+1}=w\&times;v_{\mathrm{id}}^k+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="HSA00000031368300022.png"\; state="\{\{state\}\}">c</figure-callout>}}_1\&times;\mathrm{randu}\left(\mathrm{0,1}\right)\&times;({\mathrm{pbest}}_{\mathrm{id}}-x_{\mathrm{id}}^k)+{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HSA00000031368300022.png,HSA00000031368300031.png"\; state="\{\{state\}\}">c</figure-callout>}}_2\&times;\mathrm{randu}\left(\mathrm{0,1}\right)\&times;({\mathrm{pbest}}_{\mathrm{gd}}-x_{\mathrm{id}}^k)$

$x_{\mathrm{id}}^{k+1}=x_{\mathrm{id}}^k+v_{\mathrm{id}}^{k+1}$

In following formula, i=1,2 ..., 2Q;

It is the d dimension component of particle i velocity in the k time iteration;

It is the d dimension component of particle i position vector in the k time iteration; Pbest _IdThe d dimension component of the desired positions that is arrived for particle i; Pbest _GdD dimension component for desired positions that population arrives; c ₁, c ₂Be weight factor, be respectively 0.5 in the present invention; W is the inertia weight function, is 3 in the present invention.

When the each iteration of PSO, all need calculate the pairing adaptive value F of particle current location (X), also promptly be quality by the determined scheduling rule of particle current location.From X={x ₁, x ₂..., x _2QCan obtain RP ', again according to min (rp _i) and max (rp _i), obtain RP.Based on scheduling rule Rule={RS, RP} carries out industrial process simulation to the scheduling problem-instance, can obtain the scheduling index f of this rule correspondence.According to performance index f, can try to achieve particle position X={x ₁, x ₂..., x _2QPairing adaptive value.

The present invention makes the scheduling rule of being excavated to different scheduling problem examples of the same type dispatching effect preferably be arranged all, with X={x ₁, x ₂..., x _2QPairing scheduling rule acts on a plurality of similar scheduling problem examples, obtain a plurality of scheduling schemes, and adopt scheduling rule integrated evaluating method that the performance of above-mentioned given scheduling rule is carried out comprehensive evaluation, thereby obtain the adaptive value F (X) of X correspondence based on Bayesian Estimation.

This method is at first with X={x ₁, x ₂..., x _2QCorresponding given scheduling rule acts on h of the same type the scheduling problem example (h=100) that is obtained by the 6th step, obtains h scheduling index (f of above-mentioned scheduling problem example ₁, f ₂..., f _h).Consider that above-mentioned scheduling problem example is of the same type, the present invention sees above-mentioned prior probability distribution by the scheduling index that produces under the same given scheduling rule effect as a normal distribution, R～N (μ, σ ²).According to the Bayesian Estimation theory, when an above-mentioned h scheduling index is (f ₁, f ₂..., f _h) time, when this scheduling rule was acted on above-mentioned a plurality of scheduling problem example of the same type, the posterior probability of the scheduling index that is obtained also satisfied normal distribution, promptly

$R|({\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000031368300022.png"\; state="\{\{state\}\}">f</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="HSA00000031368300022.png,HSA00000031368300031.png"\; state="\{\{state\}\}">f</figure-callout>}}_2,\&CenterDot;\&CenterDot;\&CenterDot;,f_h)~N(\stackrel{\&OverBar;}{\mathrm{\&mu;}},{\stackrel{\&OverBar;}{\mathrm{\&sigma;}}}^2)$

Wherein $\stackrel{\&OverBar;}{\mathrm{\&mu;}}=w{\mathrm{\&Sigma;}}_{i=1}^h{f}_i/h+(1-w)\&CenterDot;\mathrm{\&mu;},$ w＝1/(1+se ²/σ ²)， $\stackrel{\&OverBar;}{\mathrm{\&sigma;}}=\sqrt{1/(1/s{\mathrm{<figure-callout\; id="2"\; label="e"\; filenames="HSA00000031368300022.png,HSA00000031368300031.png"\; state="\{\{state\}\}">e</figure-callout>}}^2+1/{\mathrm{\&sigma;}}^2)},$ $\mathrm{se}=\mathrm{\&sigma;}/\sqrt{h}.$

According to above-mentioned probability distribution, the fiducial interval when fiducial probability is 0.9 can obtain by following formula:

Because above-mentioned probability distribution is normal distribution, can obtain by looking into the normal probability paper table, that is:

$a=1.63\&CenterDot;\stackrel{\&OverBar;}{\mathrm{\&sigma;}}$

So the scheduling problem example performance index of given scheduling rule correspondence are that 0.9 o'clock fiducial interval is at fiducial probability

The adaptive value that the present invention defines particle position is:

$F\left(X\right)=\stackrel{\&OverBar;}{\mathrm{\&mu;}}-1.63\&CenterDot;\stackrel{\&OverBar;}{\mathrm{\&sigma;}}$

In view of above-mentioned analysis, carry out the scheduling rule parameter global coordination optimization as follows, thereby excavate the pairing scheduling rule of each scheduling problem example types:

Step (8.2.1): each parameter Q of initialization, X, V, pbest and gbest, it is 1 that the current iteration number of times is set;

Step (8.2.2): according to particle more new formula upgrade particle position and speed;

Step (8.2.3): the pbest and the gbest that upgrade each particle according to the adaptive value computing formula of particle;

Step (8.2.4): judge whether to satisfy the algorithm stop condition, if the algorithm iteration number of times reaches K _Max, then stop, according to the gbest of particle, determine the scheduling rule parameter, thereby generate scheduling rule; As not satisfying, the current iteration number of times adds 1, returns the 8.2.2 step.

Scheduling rule intelligent excavating method according to above-mentioned rule-based parameter global coordination optimization, the present invention has done a large amount of l-G simulation tests, also said method is applied to simultaneously in actual the process of textile production scheduling and the scheduling of microelectronics production run (scheduling rule is excavated running software figure as shown in Figure 8).Can find out that from simulation result and practical application effect the scheduling rule that adopts the present invention to excavate can provide complex process scheduling scheme preferably at short notice, can be adapted to complex process Real-Time Scheduling environment.

Description of drawings

Fig. 1: scheduling rule intelligent excavating hardware system structure figure.

Fig. 2: scheduling rule intelligent excavating method structural drawing.

Fig. 3: the rule parameter global coordination optimization method flow diagram of dividing PSO based on linearity.

Fig. 4: have L bar ANFIS structural drawing.

Fig. 5: the processing route figure of certain workpiece in the actual production, one of them circle is represented an operation, and horizontal line top is operation number in the circle, and the below is the affiliated operation of this operation.

Fig. 6: scheduling rule is excavated the running software interface.

Embodiment

The intelligent excavating method of scheduling rule depends on scheduling rule intelligent excavating hardware system and the realization of scheduling rule intelligent excavating software in the complex process of the present invention.Its hardware system is formed (structure as shown in Figure 1) by rule digging computing machine and production scheduling information collecting device.The rule digging computing machine reads the relevant real-time information of production scheduling from the production scheduling information collecting device, the scheduling rule intelligent excavating software that is installed on the rule digging computing machine carries out the scheduling rule excavation according to these information.

Below the related step of scheduling rule intelligent excavating method of above-mentioned rule-based parameter global coordination optimization that the present invention is proposed be elaborated:

The first step: definition scheduling problem example

The scheduling problem example is designated as Problem, and it can be defined as follows:

Existing N workpiece formed workpiece set J={1 ..., N}.M machine group formed machine group set M={M ₁, M ₂..., M _m, machine group M wherein _iBy N (M _i) the identical parallel machine of platform forms.S manufacturing procedure S={S arranged ₁, S ₂..., S _s, operation S wherein _iThe machine group set that is comprised is MS (S _i).The release of workpiece i is R constantly _i, the deadline is C _i, be D delivery date _i, its process is by n _iIndividual operation O _I1, O _I2...,

Form operation O _IjCan

Process on arbitrary machine in the machine group, be p (O its process time _Ij), operation O _IjDirect subsequent operation set next (O in processing route _Ij) expression, directly preceding continuous operational set prev (O _Ij) expression.A (O _Ij), b (O _Ij) and c (O _Ij) be respectively and operate O _IjTime of arrival, processing start time and process finishing time.

Process satisfies following constraint:

● can not interrupt constraint: operation Once you begin processing just can not stop, until machining;

● the processing route constraint:

Each operation that is workpiece must be processed by predetermined processing route requirement.

Satisfying under the condition of above-mentioned constraint, the regulation goal that the present invention sets has two kinds:

● minimize and drag issue:

$\min \left\{{\mathrm{\&Sigma;}}_{i=1}^N\mathrm{sign}(C_i-D_i)\right\}$

Wherein, drag issue to be meant in a scheduling problem example deadline C _iBe later than D at delivery date _iThe workpiece sum, can be expressed as

$\mathrm{sign}(C_i-D_i)=\left\{\begin{array}{c}1,C_i-D_i>0\\ 0,C_i-D_i\&le;0\end{array}\right.;$

● minimize the manufacturing cycle:

min{max{C _i|i＝1，2，…，N}}

Wherein the manufacturing cycle is meant in a scheduling problem example, processes the used time of all workpiece, can be expressed as max{C _i| i=1,2 ..., N}.

Second step: scheduling rule intelligent excavating software is installed on described rule digging computing machine

Scheduling rule intelligent excavating software is realized by the scheduling rule intelligent excavating framework towards complex process Real-Time Scheduling environment.The core of this framework is that the scheduling rule mining process is converted into scheduling rule parameter global coordination optimization process.This framework comprises structure scheduling problem case library, divides the scheduling problem example, makes up scheduling problem case library of the same type, makes up and find the solution the rule parameter global coordination optimization problem at every class problem-instance.The scheduling rule intelligent excavating framework is at first according to the actual production data, make up the scheduling problem example, and the scheduling problem example is kept in the scheduling problem case library, afterwards the scheduling problem example types is divided and produced to the relevant scheduling problem example in the scheduling problem case library, to the relevant scheduling problem example of every class, produce a plurality of scheduling problem examples of the same type and make up scheduling rule parameter global coordination optimization problem, on this basis, adopt the rule parameter global coordination optimization algorithm to find the solution above-mentioned parameter of regularity optimization problem, in each parameter iteration optimizing process, the scheduling rule that parameter is given acts on above-mentioned a plurality of scheduling problem examples of the same type and carries out industrial process simulation, obtain corresponding a plurality of scheduling index, and adopt the scheduling rule integrated evaluating method that the current scheduling rule is estimated, the evaluation of estimate that is obtained is as the target function value of rule parameter global coordination optimization problem, final scheduling rule after obtaining to optimize.Excavate framework according to above-mentioned scheduling rule, scheduling rule intelligent excavating software comprises structure scheduling problem case library module, scheduling problem example division module, makes up scheduling problem case library module of the same type, makes up scheduling rule parameter global coordination optimization problem module, finds the solution scheduling rule parameter global coordination optimization problem module.

The 3rd step: produce real-time information collection

Gather the production real-time information with the production scheduling information collecting device, comprise and produce workpiece information, manufacturing schedule information, facility information, and above-mentioned information is sent to scheduling rule by netting twine excavates computing machine, wherein, producing workpiece information comprises and respectively operates process time, workpiece release time, workpiece delivery date, workpiece processing route information in workpiece quantity, the workpiece; Manufacturing schedule information comprises on-stream time, concluding time, the time out that operates on the machine; Facility information comprises failure message, maintenance information.

The 4th step: make up the scheduling problem case library

In described scheduling rule excavation computing machine, call and make up scheduling problem case library module, set up the scheduling problem case library.Go on foot the production real-time information of gathering according to the first step defined scheduling problem example implication and the 3rd in particular moment, produce the scheduling problem example of particular moment, the particular moment of being adopted among the present invention is 8:00,12:00, the 16:00 of every day, can produce 3 scheduling problem examples like this every day, the scheduling problem example that will set up every day is stored in the scheduling problem case library successively.Press the said process certain time, the time that continues among the present invention is 3 months.

The 5th step: divide the scheduling problem example

In described scheduling rule excavation computing machine, call and divide the scheduling problem example module, the scheduling problem example that is kept in the scheduling problem case library is classified.In this module, adopt scheduling problem example division methods, to being stored in K scheduling problem example { Problem in the scheduling problem case library based on double-deck fuzzy C-means clustering ₁..., Problem _KDivide, the setting according to the 4th step has 270 scheduling problem examples, i.e. K=270 in the case library.In the method, lower floor's cluster is responsible for the information relevant with the scheduling problem example is carried out cluster, extracts characteristic information; The characteristic information that the upper strata cluster is responsible for extracting carries out cluster, to be used for the division of scheduling problem example.This division scheduling problem example module is divided the scheduling problem-instance as follows:

The 5.1st step: realize the fuzzy C-means clustering algorithm

In the levels cluster, all need use following fuzzy C-means clustering method: suppose that sample set is X={x ₁, x ₂..., x _n, be divided into C ambiguity group, and ask every group cluster centre c _j, j=1 ..., C makes following defined desired value J _CReach minimum:

$J_C=\underset{j=1}{\overset{C}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}^{\mathrm{\&alpha;}}{\left|\right|x_i-c_j\left|\right|}^2,$ 1≤α≤∞

And need to satisfy: $\underset{j=1}{\overset{C}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}=1,$ $\&ForAll;i=1,\&CenterDot;\&CenterDot;\&CenterDot;,n$

Wherein, μ _IjI data point of ∈ [0,1] expression belongs to the degree of membership of j cluster centre, c _jBe j cluster centre, α is a weighted index.Fuzzy membership μ _IjAnd c _jCan obtain with following formula respectively:

${\mathrm{\&mu;}}_{\mathrm{ij}}=1/\underset{k=1}{\overset{C}{\mathrm{\&Sigma;}}}{\left(\frac{{\left|\right|x_i-c_i\left|\right|}^2}{{\left|\right|x_i-c_k\left|\right|}^2}\right)}^{2/(\mathrm{\&alpha;}-1)}$

$c_j=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}^m{x}_i/\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}^m$

The 5.2nd step: lower floor's cluster, it extracts the scheduling problem characteristic information from the various information of scheduling problem

Lower floor's cluster is responsible for extracting the scheduling problem characteristic information from the various information of scheduling problem, and it is realized as follows:

The 5.2.1 step: extract the workpiece technology characteristics

Adopt following ternary representation to represent the technology characteristics of workpiece i:

Process _i＝{ML _i，MF _i，AF _i}

Wherein:

● ML _iBe the longest path in the processing route of workpiece i, note arrives operation O _IjThe farm labourer at the place path of planting is ML (O _Ij), ML (O then _Ij) can obtain by the following formula iteration:

And have

● MF _iBe the maximum branch amount in the processing route of workpiece i, note operation O _IjThe maximum branch amount at place is MF (O _Ij), MF (O then _Ij) can obtain by the following formula iteration:

And MF is arranged _i=max (MF (O _Ij) | j=1 ..., ni)

● AF _iBe the average of branch amount before and after each operation of workpiece i, that is:

$A{F}_i=\frac{1}{n_i}\underset{j=1}{\overset{n_i}{\mathrm{\&Sigma;}}}\left(\right|\mathrm{prev}\left(O_{\mathrm{ij}}\right)|+|\mathrm{next}\left(O_{\mathrm{ij}}\right)\left|\right)$

Calculate { Problem successively ₁..., Problem _KThe Process value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Process value is carried out cluster.Note cluster centre quantity is C ^p, the Process cluster centre of workpiece is

J=1,2 ..., C ^p, the Process of workpiece i _iThe degree of membership that belongs to each center is

J=1 ..., C ^pThe workpiece technology characteristics vector of definition scheduling problem example is:

[PP(1)，…，PP(C ^p)]

Wherein

J=1 ..., C ^p, belong to the average degree of membership value of j Process cluster centre for the Process of workpiece in this scheduling problem example.

The 5.2.2 step: extract the workpiece processing temporal characteristics

The present invention adopts theoretical process time of the Makespan of workpiece to characterize feature process time, and this Makespan represents that a workpiece finishes under the situation about waiting for the required shortest time of processing not having.Use Makespan the theoretical process time of workpiece i _iExpression, it can obtain by the following formula iterative computation:

Calculate { Problem successively ₁..., Problem _KThe Makespan value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Makespan value is carried out cluster.Note cluster centre quantity is C ^m, the cluster centre of Makespan is respectively

J=1,2 ..., C ^m, the Makespan of workpiece i _iThe degree of membership that belongs to each center is designated as J=1 ..., C ^mThe workpiece processing temporal characteristics vector of definition scheduling problem example is:

[PM(1)，…，PM(C ^m)]

Wherein

J=1 ..., C ^m, belong to the average degree of membership value of j Makespan cluster centre for the Makespan of workpiece in this scheduling problem example.

The 5.2.3 step: extract workpiece and hand over phase tightness feature

The processing tightness that the present invention characterizes a workpiece with the process time and the difference between the friendship phase of a workpiece, with the friendship phase tightness of following formula definition workpiece i:

Slack _i＝D _i-Makespan _i

Calculate { Problem successively ₁..., Problem _KThe Slack value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Slack value is carried out cluster.Note cluster centre quantity is C ^d, hand over phase tightness Slack cluster centre to be

J=1,2 ..., C ^d, the Slack of workpiece i _iThe degree of membership that belongs to each center is

J=1 ..., C ^dThe workpiece of definition scheduling problem example hands over phase tightness proper vector to be:

[PD(1)，PD(2)，…，PD(C ^d)]

Wherein

J=1 ..., C ^d, belong to the average degree of membership value of j Slack cluster centre for the Slack of workpiece in this scheduling problem example.

The 5.2.4 step: extract the machine process capacity characteristic

The present invention represents with the contained machine quantity of each operation in the scheduling problem example, that is:

Capability＝{SC ₁，…，SC _s}

Wherein

I=1 ..., s is the total quantity of i machine that operation comprises in this scheduling problem example.

Calculate { Problem successively ₁..., Problem _KThe Capability of all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Capability value is carried out cluster.Note cluster centre quantity is C ^c, machinery processing capacity Capability cluster centre is designated as

J=1,2 ..., C ^c, the Capability of k scheduling problem example _kThe degree of membership that belongs to each center is designated as

J=1 ..., C ^cThe machinery processing capacity proper vector of scheduling problem example correspondence is:

[PC(1)，PC(2)，…，PC(C ^c)]

Wherein $\mathrm{PC}\left(j\right)={\mathrm{\&mu;}}_{\mathrm{kj}}^c,$ j＝1，…，C ^c

The 5.3rd step: the upper strata cluster, it divides the scheduling problem example according to the scheduling problem characteristic information

In the clustering method of upper strata, the technology characteristics vector that the present invention obtained for the 5.2nd step, workpiece processing temporal characteristics vector, workpiece hand over phase tightness proper vector, machinery processing capacity proper vector to merge, total characteristic vector as a scheduling problem example is designated as:

X=[PP (1) ..., PP (C ^p), PM (1) ..., PM (C ^m), PD (1) ..., PD (C ^d), PC (1) ..., PC (C ^c)] with above-mentioned vector as sample, then { a Problem in the clustering algorithm ₁..., Problem _KTotal K sample, use the fuzzy C-means clustering method to carry out cluster to these samples, note cluster centre quantity is C ^r, then resulting scheduling problem example cluster centre is

J=1,2 ..., C ^r, the degree of membership that k scheduling problem example belongs to each scheduling problem example cluster centre is designated as J=1 ..., C ^rThe scheduling problem example is divided into C ^rClass makes in k scheduling problem example

Maximum classification is the affiliated type of this scheduling problem example, in case study on implementation of the present invention, the scheduling problem example is divided into 14 scheduling problem example types.

The 6th step: make up scheduling problem case library of the same type

In described scheduling rule excavation computing machine, call and make up similar scheduling problem case library module, make up scheduling problem case library of the same type.Because the scheduling problem example quantity in the scheduling problem case library is limited after all, and the scheduling rule excavation needs a large amount of scheduling problem examples in order to estimate the quality of the scheduling rule of being excavated, so need initiatively produce the scheduling problem example of the same type of sufficient amount at each scheduling problem example types.The present invention is directed to each scheduling problem example types, produce 100 scheduling problem examples of the same type, its method is as follows: divide the result according to the scheduling problem that the 5th step was determined, at the scheduling problem example { Problem of scheduling problem case library preservation ₁..., Problem _KIn, choose a scheduling problem example that belongs to current scheduling problem-instance type wantonly, and the correlated characteristic information of this scheduling problem example is carried out random fluctuation, to produce a plurality of different scheduling problem examples of the same type.(α β) is uniform random number in [α, β] interval to note randu, and the present invention gets α, and β is respectively 0.9 and 1.1.Need the information of change to have:

● workpiece quantity: the workpiece quantity after the variation is

● operate process time: the operation after the variation is process time

● workpiece delivery date: the workpiece after the variation is delivery date

● comprise number of machines in each machine group: comprising number of machines in each the machine group after the variation is $\tilde{N}\left(M_i\right)=N\left(M_i\right)\&times;p\left(\mathrm{0.9,1.1}\right)$

After the information change of above-mentioned scheduling problem example, promptly produced a new scheduling problem example, this example has been saved in the current scheduling problem-instance type of scheduling problem case library of the same type.

According to said method, can produce a plurality of different scheduling problem examples to each scheduling problem type, these scheduling problem examples all are saved in the scheduling problem case library of the same type.

The 7th step: make up scheduling rule parameter global coordination optimization problem

In described scheduling rule excavation computing machine, call and make up scheduling rule parameter global coordination optimization problem module.In this module, scheduling rule is defined as Rule={RS, RP}, wherein RS is the scheduling rule structure, RP is the scheduling rule parameter sets under specific RS.Scheduling rule is input as operational attribute and machine group attribute two category informations, and is output as the processing priority of workpiece, wherein:

● the processing priority of workpiece is high more, and workpiece just should be processed more in advance.

● operational attribute has: process time, time of arrival, prediction time of arrival, haulage time, delivery date, residue process time, residue processing step number, subsequent operation number, subsequent handling bottleneck degree.

● machine group attribute has: comprise operation number to be processed in number of machines, machine group working ability, machine group bottleneck degree, the machine group, interior operation summation process time to be processed of machine group, estimate to arrive summation process time of operating;

The present invention adopts Adaptive Neuro-fuzzy Inference (ANFIS:Adaptive Neuro-Network Fuzzy InferenceSystem) in order to the expression scheduling rule, be above-mentioned workpiece attribute of being input as of ANFIS and machine group attribute, be output as workpiece processing priority.Then the structure RS of scheduling rule is ANFIS, its concrete structure as shown in Figure 4, scheduling rule parameters R P is the parameter among the ANFIS, is designated as RP={rp ₁, rp ₂..., rp _ZExpression, wherein Z is the number of parameter among the ANFIS.On this basis, constructed scheduling rule parameter global coordination optimization problem is: coordination optimization RP={rp ₁, rp ₂..., rp _Z, so that by RP={rp ₁, rp ₂..., rp _ZPerformance the best of the scheduling rule determined.

Adopt said method,, make up corresponding scheduling rule parameter global coordination optimization problem respectively at each scheduling problem example types.

The 8th step: find the solution scheduling rule parameter global coordination optimization problem

In described scheduling rule excavation computing machine, call and find the solution scheduling rule parameter global coordination optimization problem module, the pairing scheduling rule parameter global coordination optimization of each scheduling problem example types problem is found the solution.This module adopts based on linearity division PSO (particle group optimizing) method each scheduling rule parameter global coordination optimization problem is found the solution, to obtain the scheduling rule corresponding with each scheduling problem example types.This method realizes as follows:

The 8.1st step: for reducing the number of parameter to be optimized, with parameter of regularity RP={rp to be optimized ₁, rp ₂..., rp _ZExpress with the slope and the displacement of a plurality of different line segments.At first the parameter among the RP is pressed parameter value scope [min (rp separately _i), max (rp _i)] normalization, i.e. rp ' _i=(rp _i-min (rp _i))/(max (rp _i)-min (rp _i)), i=1 ..., Z, obtain RP '=rp ' ₁, rp ' ₂..., rp ' _Z, in the present invention, the parameter value scope of ANFIS is decided to be [1000,1000], i.e. min (rp _i)=-1000, max (rp _i)=1000.Simultaneously, suppose that slope and displacement with Q bar line segment are k _j, b _j, j=1,2 ..., Q represents the parameter among the RP ', with Q=10 bar line segment, the denotable parameter set of j bar line segment is designated as Ie among the present invention _j, the division of above-mentioned parameter need be satisfied Ie ₁∪ Ie ₂∪ ... ∪ Ie _Q=RP ' and

I ≠ j, i=1 ..., Q, j=1 ..., Q.Then the relation of parameter among the RP ' and Q bar line segment can be represented by the formula:

rp′ _(i)＝k _j×(i)+b _j-[k _j×(i)+b _j] ⁺， $r{p}_{\left(i\right)}^{\&prime;}\&Subset;S_j,$ j＝1，…，Q

The 8.2nd step: adopt the PSO method to k _J,b _j, j=1,2 ..., Q is optimized.With k _jAnd b _jAs decision variable, then each particle position is X={k in the PSO method _j, b _j| j=1,2 ..., Q} also can be expressed as X={x ₁, x ₂..., x _2Q, each particle's velocity is V={v ₁, v ₂..., v _2Q.Optimize in the iteration each, each particle upgrades by following formula:

$v_{\mathrm{id}}^{k+1}=w\&times;v_{\mathrm{id}}^k+{\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="HSA00000031368300022.png"\; state="\{\{state\}\}">c</figure-callout>}}_1\&times;\mathrm{randu}\left(\mathrm{0,1}\right)\&times;({\mathrm{pbest}}_{\mathrm{id}}-x_{\mathrm{id}}^k)+{\mathrm{<figure-callout\; id="2"\; label="c"\; filenames="HSA00000031368300022.png,HSA00000031368300031.png"\; state="\{\{state\}\}">c</figure-callout>}}_2\&times;\mathrm{randu}\left(\mathrm{0,1}\right)\&times;({\mathrm{pbest}}_{\mathrm{gd}}-x_{\mathrm{id}}^k)$

$x_{\mathrm{id}}^{k+1}=x_{\mathrm{id}}^k+v_{\mathrm{id}}^{k+1}$

In following formula, i=1,2 ..., 2Q; It is the d dimension component of particle i velocity in the k time iteration;

It is the d dimension component of particle i position vector in the k time iteration; Pbest _IdThe d dimension component of the desired positions that is arrived for particle i; Pbest _GdD dimension component for desired positions that population arrives; c ₁, c ₂Be weight factor, be respectively 0.5 in the present invention; W is the inertia weight function, is 3 in the present invention.

When the each iteration of PSO, all need calculate the pairing adaptive value F of particle current location (X), also promptly be quality by the determined scheduling rule of particle current location.From X={x ₁, x ₂..., x _2QCan obtain RP ', again according to min (rp _i) and max (rp _i), obtain RP.Based on scheduling rule Rule={RS, RP} carries out industrial process simulation to the scheduling problem-instance, can obtain the scheduling index f of this rule correspondence.According to performance index f, can try to achieve particle position X={x ₁, x ₂..., x _2QPairing adaptive value.

The present invention makes the scheduling rule of being excavated to different scheduling problem examples of the same type dispatching effect preferably be arranged all, with X={x ₁, x ₂..., x _2QPairing scheduling rule acts on a plurality of similar scheduling problem examples, obtain a plurality of scheduling schemes, and adopt scheduling rule integrated evaluating method that the performance of above-mentioned given scheduling rule is carried out comprehensive evaluation, thereby obtain the adaptive value F (X) of X correspondence based on Bayesian Estimation.

This method is at first with X={x ₁, x ₂..., x _2QCorresponding given scheduling rule acts on h of the same type the scheduling problem example (h=100) that is obtained by the 6th step, obtains h scheduling index (f of above-mentioned scheduling problem example ₁, f ₂..., f _h).Consider that above-mentioned scheduling problem example is of the same type, the present invention sees above-mentioned prior probability distribution by the scheduling index that produces under the same given scheduling rule effect as a normal distribution, R～N (μ, σ ²).According to the Bayesian Estimation theory, when an above-mentioned h scheduling index is (f ₁, f ₂..., f _h) time, when this scheduling rule was acted on above-mentioned a plurality of scheduling problem example of the same type, the posterior probability of the scheduling index that is obtained also satisfied normal distribution, promptly

$R|({\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="HSA00000031368300022.png"\; state="\{\{state\}\}">f</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="f"\; filenames="HSA00000031368300022.png,HSA00000031368300031.png"\; state="\{\{state\}\}">f</figure-callout>}}_2,\&CenterDot;\&CenterDot;\&CenterDot;,f_h)~N(\stackrel{\&OverBar;}{\mathrm{\&mu;}},{\stackrel{\&OverBar;}{\mathrm{\&sigma;}}}^2)$

Wherein $\stackrel{\&OverBar;}{\mathrm{\&mu;}}=w{\mathrm{\&Sigma;}}_{i=1}^h{f}_i/h+(1-w)\&CenterDot;\mathrm{\&mu;},$ w＝1/(1+se ²/σ ²)， $\stackrel{\&OverBar;}{\mathrm{\&sigma;}}=\sqrt{1/(1/s{\mathrm{<figure-callout\; id="2"\; label="e"\; filenames="HSA00000031368300022.png,HSA00000031368300031.png"\; state="\{\{state\}\}">e</figure-callout>}}^2+1/{\mathrm{\&sigma;}}^2)},$ $\mathrm{se}=\mathrm{\&sigma;}/\sqrt{h}.$

According to above-mentioned probability distribution, the fiducial interval when fiducial probability is 0.9 can obtain by following formula:

Because above-mentioned probability distribution is normal distribution, can obtain by looking into the normal probability paper table, that is:

$a=1.63\&CenterDot;\stackrel{\&OverBar;}{\mathrm{\&sigma;}}$

So the scheduling problem example performance index of given scheduling rule correspondence are that 0.9 o'clock fiducial interval is at fiducial probability

The adaptive value that the present invention defines particle position is:

$F\left(X\right)=\stackrel{\&OverBar;}{\mathrm{\&mu;}}-1.63\&CenterDot;\stackrel{\&OverBar;}{\mathrm{\&sigma;}}$

In view of above-mentioned analysis, carry out the scheduling rule parameter global coordination optimization as follows, thereby excavate the pairing scheduling rule of each scheduling problem example types:

Step (8.2.1): each parameter Q of initialization, X, V, pbest and gbest, it is 1 that the current iteration number of times is set;

Step (8.2.2): according to particle more new formula upgrade particle position and speed;

Step (8.2.3): the pbest and the gbest that upgrade each particle according to the adaptive value computing formula of particle;

Step (8.2.4): judge whether to satisfy the algorithm stop condition, if the algorithm iteration number of times reaches K _Max, then stop, according to the gbest of particle, determine the scheduling rule parameter, thereby generate scheduling rule; As not satisfying, the current iteration number of times adds 1, returns the 8.2.2 step.

The scheduling rule intelligent excavating method flow process of the rule-based parameter global coordination optimization that the present invention proposes as shown in Figure 3.

Scheduling rule intelligent excavating method (SRDM:Scheduling Rule Data Mining) according to the above-mentioned rule-based parameter global coordination optimization that proposes, the present invention obtains a large amount of actual production data from case study on implementation, on this basis, carried out sufficient numerical experiment.Method of the present invention is applied to the actual production process case, has obtained effect preferably.

Specific implementation process is as follows:

At first scheduling rule is installed and excavates the software and hardware system according to the requirement of this instructions.

Secondly, from the production scheduling computing machine, read nearest 3 months production history data, every days 8 point, 12 and 16 obtain production run scheduling problem example respectively, thereby form 270 actual schedule problem-instance.The scheduling problem example that provides according to this instructions is divided into 14 classes based on the corresponding division methods parameter that is adopted in the division methods of double-deck fuzzy C-means clustering algorithm and the table 1 with 270 scheduling problem examples, is designated as C01～C14.And,, produce 100 scheduling problem examples respectively at 14 class scheduling problem examples according to the scheduling problem example producing method that this instructions provides.

Afterwards, excavate in the computing machine in scheduling rule, the scheduling rule parameter that provides according to this instructions is divided the corresponding optimization method parameter that is adopted in the global coordination optimization method of PSO and the table 1 based on linearity, at the different scheduling problem examples in the 14 class scheduling types, carry out rule digging respectively, and the pairing scheduling rule of every class scheduling problem example is reached the production scheduling computing machine.

At last, the production scheduling computing machine is determined the affiliated type of this scheduling problem example, and is obtained the scheduling rule corresponding with the type according to scheduling problem example current to be found the solution, and uses it for finding the solution of scheduling problem example, finally obtains scheduling scheme preferably.

The used main algorithm parameter of numerical experiment of the present invention is as shown in table 1.

Table 1 parameter list

Sequence number	Parameter name	Parameter declaration	Parameter value
					1.	C p	The technology characteristics number of clusters	4
2.	C m	Process time feature clustering quantity	4
				3.	C d	Friendship phase tightness feature clustering quantity	4
4.	C c	The process capacity characteristic number of clusters	4
				5.	C r	The scheduling problem number of clusters	14
6.	α	The clustering algorithm weighted index	2
				7.	L	ANFIS rule bar number	20
8.	Q	The linear dividing line hop count of PSO amount	10
				9.	K max	PSO algorithm maximum iteration time	1000

The present invention adopts following scheduling rule to be used for comparison with the scheduling rule excavated:

1) to minimize the scheduling rule that the manufacturing cycle is a target:

● SPT (Shortest Processing Time): short more preferential more processing process time;

● LPT (Longest Processing Time): long more preferential more processing process time;

● SRPT (Shortest Remaining Processing Time): residue short more preferential more processing process time, operation O _IjResidue rp process time (O _Ij) be

$\mathrm{rp}\left(O_{\mathrm{ij}}\right)=\left\{\begin{array}{c}\max \{p\left(O_{\mathrm{ik}}\right)+\mathrm{rp}\left(O_{\mathrm{ik}}\right)|O_{\mathrm{ik}}\&Element;\mathrm{next}\left(O_{\mathrm{ij}}\right)\},\mathrm{next}\left(O_{\mathrm{ij}}\right)\&NotEqual;\mathrm{\&phi;}\\ 0,\mathrm{next}\left(O_{\mathrm{ij}}\right)=\mathrm{\&phi;}\end{array}\right.$

● WINQ (Workload in the Next Queue): the more little preferential more processing of the machine group load at subsequent operation place, it is calculated as follows:

$\mathrm{wn}\left(O_{\mathrm{ij}}\right)=\underset{O_{\mathrm{pq}}\&Element;\{a\left(O_{\mathrm{pq}}\right)\&le;t,c\left(O_{\mathrm{pq}}\right)\&GreaterEqual;t,O_{\mathrm{pq}}\&Element;M\left(O_{\mathrm{ik}}\right),O_{\mathrm{ik}}\&Element;\mathrm{next}\left(O_{\mathrm{ij}}\right)\}}{\mathrm{\&Sigma;}}p\left(O_{\mathrm{pq}}\right)$

Wherein, t is the current scheduling rule decision moment, promptly begins the machine moment on the selection operation when the free time appears in a machine.

2) to drag issue be the scheduling rule of target to minimize:

● EDD (Earliest Due Date): delivery date is more early preferential more;

● SLACK: slack time rule, the operation O _IjSl (O slack time _Ij) more little preferential more processing, be calculated as follows slack time:

sl(O _ij)＝D _i-rp(O _ij)

●?MOD(Modified?Operation?Due?date)：

Phase first more processing is more early handed in amended operation, and the amended operation friendship phase is defined as follows:

d′(O _ij)＝max{d(O _ij)，t+p(O _ij)}

Wherein, d (O _Ij) for operating O _IjOperation hand over the phase, it is defined as:

$d\left(O_{\mathrm{ij}}\right)=\left\{\begin{array}{c}\min \{d\left(O_{\mathrm{ik}}\right)+D_i\&CenterDot;p\left(O_{\mathrm{ik}}\right)/M{K}_i|O_{\mathrm{ik}}\&Element;\mathrm{prev}\left(O_{\mathrm{ij}}\right)\},\mathrm{prev}\left(O_{\mathrm{ij}}\right)\&NotEqual;\mathrm{\&phi;}\\ D_i\&CenterDot;p\left(O_{\mathrm{ij}}\right)/M{K}_i,\mathrm{prev}\left(O_{\mathrm{ij}}\right)=\mathrm{\&phi;}\end{array}\right.$

MK _iFor the minimum of workpiece i (each operation all do not have etc. the bide one's time) manufacturing cycle, be defined as

MK _i＝max{p(O _ik)+rp(O _ik)|prev(O _ik)＝φ}

●?CR+SPT(Critical?Ratio+SPT)：

The operation of this rule definition is delivery date:

d″(O _ij)＝max{t+β(t)·p(O _ij)，t+p(O _ij)}

β wherein _i(t)=(D _i-t)/MK _i

The present invention is based on above-mentioned scheduling problem example types (C01～C14), drag the issue regulation goal at minimizing the manufacturing cycle regulation goal and minimizing, with this algorithm SRDM and aforementioned to minimize four kinds of heuristic rule methods (SPT, LPT, SRPT and WINQ) that the manufacturing cycle is an optimization aim and to drag issue be that four kinds of heuristic rule methods (EDD, SLACK, MOD and CR+SPT) of optimization aim have been carried out numerical evaluation relatively to minimize, correlation computations result lists in table 2 and table 3 respectively respectively.From table 2 and table 3, can find out, no matter regulation goal still minimizes and drags issue for minimizing the manufacturing cycle, the scheduling rule of excavating based on the SRDM method all is better than above-mentioned heuristic scheduling rule, be 4.60% (with respect to for using WINQ in the scheduling problem Type C 06) at the minimum improvement rate that minimizes the manufacturing cycle regulation goal wherein, most improvement rates are between 7%～12%; At minimizing the minimum improvement rate that drags the issue regulation goal is 5.3% (with respect to for using CR+SPT in the scheduling problem Type C 03), and most improvement rates are between 9%～14%.Also can find out from table 2, for different scheduling problem examples, the scheduling performance of four kinds of heuristic rules differs greatly, promptly has clear superiority without any the relative Else Rule of a kind of heuristic rule, and the scheduling rule of excavating based on the SRDM method, mean value to scheduling problem example optimal effect of the same type all is better than four kinds of heuristic rules, has demonstrated SRDM to the stability in the different scheduling problem examples under the similar dispatch environment; Same conclusion also can draw from table 3.

The result that table 2 minimizes the manufacturing cycle compares

R0, R1, R2, R3, R4 are respectively the scheduling result of SRDM, SPT, LPT, SRPT, WINQ

I1, I2, I3, I4 are respectively the improvement rate of R0 with respect to R1, R2, R3, R4

Table 3 minimizes the result of calculation of dragging issue

R0, R1, R2, R3, R4 are respectively the scheduling result of SRDM, EDD, SLACK, MOD, CR+SPT,

I1, I2, I3, I4 are respectively the improvement rate of R0 with respect to R1, R2, R3, R4.

Claims (1)

1. the scheduling rule intelligent excavating method of a rule-based parameter global coordination optimization is characterized in that, described method is excavated on computing machine and the production scheduling information collecting device in scheduling rule and realized according to the following steps successively:

The first step: definition scheduling problem example

The scheduling problem example is designated as Problem, and it can be defined as follows:

Existing N workpiece formed workpiece set J={1 ..., N}, m machine group formed machine group set M={M ₁, M ₂..., M _m, machine group M wherein _iBy N (M _i) the identical parallel machine of platform forms, and s manufacturing procedure S={S arranged ₁, S ₂..., S _s, operation S wherein _iThe machine group set that is comprised is MS (S _i), the release of workpiece i is R constantly _i, the deadline is C _i, be D delivery date _i, its process is by n _iIndividual operation O _I1, O _I2...,

Form operation O _IjCan

Process on arbitrary machine in the machine group, be p (O its process time _Ij), operation O _IjDirect subsequent operation set next (O in processing route _Ij) expression, directly preceding continuous operational set prev (O _Ij) expression, a (O _Ij), b (O _Ij) and c (O _Ij) be respectively and operate O _IjTime of arrival, processing start time and process finishing time, process satisfies following constraint:

● can not interrupt constraint: operation Once you begin processing just can not stop, until machining;

● the processing route constraint:

Each operation that is workpiece must be processed by predetermined processing route requirement, is satisfying under the condition of above-mentioned constraint, and the regulation goal that the present invention sets has two kinds:

● minimize and drag issue:

$\min \left\{{\mathrm{\&Sigma;}}_{i=1}^N\mathrm{sign}(C_i-D_i)\right\}$

Wherein, drag issue to be meant in a scheduling problem example deadline C _iBe later than D at delivery date _iThe workpiece sum, can be expressed as

$\mathrm{sign}(C_i-D_i)=\left\{\begin{array}{c}1,C_i-D_i>0\\ 0,C_i-D_i\&le;0\end{array}\right.;$

● minimize the manufacturing cycle:

min{max{C _i|i＝1，2，…，N}}

Wherein the manufacturing cycle is meant in a scheduling problem example, processes the used time of all workpiece, can be expressed as max{C _i| i=1,2 ..., N};

Second step: scheduling rule intelligent excavating software is installed on described rule digging computing machine

Scheduling rule intelligent excavating software is realized by the scheduling rule intelligent excavating framework towards complex process Real-Time Scheduling environment, the core of this framework is that the scheduling rule mining process is converted into scheduling rule parameter global coordination optimization process, and it comprises structure scheduling problem case library, divides the scheduling problem example, makes up scheduling problem case library of the same type, makes up and find the solution the rule parameter global coordination optimization problem at every class problem-instance; The scheduling rule intelligent excavating framework is at first according to the actual production data, make up the scheduling problem example, and the scheduling problem example is kept in the scheduling problem case library, afterwards the scheduling problem example types is divided and produced to the relevant scheduling problem example in the scheduling problem case library, to the relevant scheduling problem example of every class, produce a plurality of scheduling problem examples of the same type and make up scheduling rule parameter global coordination optimization problem, on this basis, adopt the rule parameter global coordination optimization algorithm to find the solution above-mentioned parameter of regularity optimization problem, in each parameter iteration optimizing process, the scheduling rule that parameter is given acts on above-mentioned a plurality of scheduling problem examples of the same type and carries out industrial process simulation, obtain corresponding a plurality of scheduling index, and adopt the scheduling rule integrated evaluating method that the current scheduling rule is estimated, the evaluation of estimate that is obtained is as the target function value of rule parameter global coordination optimization problem, scheduling rule after final acquisition is optimized, excavate framework according to above-mentioned scheduling rule, scheduling rule intelligent excavating software comprises structure scheduling problem case library module, the scheduling problem example is divided module, make up scheduling problem case library module of the same type, make up scheduling rule parameter global coordination optimization problem module, find the solution scheduling rule parameter global coordination optimization problem module;

The 3rd step: produce real-time information collection

Gather the production real-time information with the production scheduling information collecting device, comprise and produce workpiece information, manufacturing schedule information, facility information, and above-mentioned information is sent to scheduling rule by netting twine excavates computing machine, wherein, producing workpiece information comprises and respectively operates process time, workpiece release time, workpiece delivery date, workpiece processing route information in workpiece quantity, the workpiece; Manufacturing schedule information comprises on-stream time, concluding time, the time out that operates on the machine; Facility information comprises failure message, maintenance information;

The 4th step: make up the scheduling problem case library

In described scheduling rule excavation computing machine, call and make up scheduling problem case library module, set up the scheduling problem case library, go on foot the production real-time information of gathering according to the first step defined scheduling problem example implication and the 3rd in particular moment, produce the scheduling problem example of particular moment, the particular moment of being adopted among the present invention is the 8:00 of every day, 12:00,16:00, can produce 3 scheduling problem examples like this every day, the scheduling problem example that to set up every day is stored in the scheduling problem case library successively, press the said process certain time, the time that continues among the present invention is 3 months;

The 5th step: divide the scheduling problem example

In described scheduling rule excavation computing machine, call and divide the scheduling problem example module, the scheduling problem example that is kept in the scheduling problem case library is classified, in this module, adopt scheduling problem example division methods, to being stored in K scheduling problem example { Problem in the scheduling problem case library based on double-deck fuzzy C-means clustering ₁..., Problem _KDivide, setting according to the 4th step, have 270 scheduling problem examples in the case library, be K=270, in the method, lower floor's cluster is responsible for the information relevant with the scheduling problem example is carried out cluster, extract characteristic information, the characteristic information that the upper strata cluster is responsible for extracting carries out cluster, and to be used for the division of scheduling problem example, this division scheduling problem example module is divided the scheduling problem-instance as follows:

The 5.1st step: realize the fuzzy C-means clustering algorithm

In the levels cluster, all need use following fuzzy C-means clustering method: suppose that sample set is X={x ₁, x ₂..., x _n, be divided into C ambiguity group, and ask every group cluster centre c _j, j=1 ..., C makes following defined desired value J _CReach minimum:

$J_C=\underset{j=1}{\overset{C}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}^{\mathrm{\&alpha;}}{\left|\right|x_i-c_j\left|\right|}^2,$ 1≤α≤∞

And need to satisfy: $\underset{j=1}{\overset{C}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}=1,$ $\&ForAll;i=1,\&CenterDot;\&CenterDot;\&CenterDot;,n$

Wherein, μ _IjI data point of ∈ [0,1] expression belongs to the degree of membership of j cluster centre, c _jBe j cluster centre, α is a weighted index, fuzzy membership μ _IjAnd c _jCan obtain with following formula respectively:

${\mathrm{\&mu;}}_{\mathrm{ij}}=1/\underset{k=1}{\overset{C}{\mathrm{\&Sigma;}}}{\left(\frac{{\left|\right|x_i-c_i\left|\right|}^2}{{\left|\right|x_i-c_k\left|\right|}^2}\right)}^{2/(\mathrm{\&alpha;}-1)}$

$c_j=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}^m{x}_i/\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_{\mathrm{ij}}^m$

The 5.2nd step: lower floor's cluster, it extracts the scheduling problem characteristic information from the various information of scheduling problem

Lower floor's cluster is responsible for extracting the scheduling problem characteristic information from the various information of scheduling problem, and it is realized as follows:

The 5.2.1 step: extract the workpiece technology characteristics

Adopt following ternary representation to represent the technology characteristics of workpiece i:

Process _i＝{ML _i，MF _i，AF _i}

Wherein:

● ML _iBe the longest path in the processing route of workpiece i, note arrives operation O _IjThe farm labourer at the place path of planting is ML (O _Ij), ML (O then _Ij) can obtain by the following formula iteration:

And have

● MF _iBe the maximum branch amount in the processing route of workpiece i, note operation O _IjThe maximum branch amount at place is MF (O _Ij), MF (O then _Ij) can obtain by the following formula iteration:

And MF is arranged _i=max (MF (O _Ij) | j=1 ..., n _i)

● AF _iBe the average of branch amount before and after each operation of workpiece i, that is:

$A{F}_i=\frac{1}{n_i}\underset{j=1}{\overset{n_i}{\mathrm{\&Sigma;}}}\left(\right|\mathrm{prev}\left(O_{\mathrm{ij}}\right)|+|\mathrm{next}\left(O_{\mathrm{ij}}\right)\left|\right)$

Calculate { Problem successively ₁..., Problem _KThe Process value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Process value is carried out cluster, establishing cluster centre quantity is C ^p, the Process cluster centre of workpiece is J=1,2 ..., C ^p, the Process of workpiece i _iThe degree of membership that belongs to each center is J=1 ..., C ^p, the workpiece technology characteristics vector of definition scheduling problem example is:

[PP(1)，…，PP(C ^p)]

Wherein

J=1 ..., C ^p, belong to the average degree of membership value of j Process cluster centre for the Process of workpiece in this scheduling problem example;

The 5.2.2 step: extract the workpiece processing temporal characteristics

The present invention adopts theoretical process time of the Makespan of workpiece to characterize feature process time, and this Makespan represents that a workpiece finishes under the situation about waiting for the required shortest time of processing not having, and uses Makespan the theoretical process time of workpiece i _iExpression, it can obtain by the following formula iterative computation:

Calculate { Problem successively ₁..., Problem _KThe Makespan value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Makespan value is carried out cluster, remember that cluster centre quantity is C ^m, the cluster centre of Makespan is respectively

J=1,2 ..., C ^m, the Makespan of workpiece i _iThe degree of membership that belongs to each center is designated as

J=1 ..., C ^m, the workpiece processing temporal characteristics vector of definition scheduling problem example is:

[PM(1)，…，PM(C ^m)]

Wherein

J=1 ..., C ^m, belong to the average degree of membership value of j Makespan cluster centre for the Makespan of workpiece in this scheduling problem example;

The 5.2.3 step: extract workpiece and hand over phase tightness feature

The processing tightness that the present invention characterizes a workpiece with the process time and the difference between the friendship phase of a workpiece, with the friendship phase tightness of following formula definition workpiece i:

Slack _i＝D _i-Makespan _i

Calculate { Problem successively ₁..., Problem _KThe Slack value of each workpiece in all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Slack value is carried out cluster, establishing cluster centre quantity is C ^d, hand over phase tightness Slack cluster centre to be

J=1,2 ..., C ^d, the Slack of workpiece i ⁱThe degree of membership that belongs to each center is J=1 ..., C ^d, the workpiece of definition scheduling problem example hands over phase tightness proper vector to be:

[PD(1)，PD(2)，…，PD(C ^d)]

Wherein

J=1 ..., C ^d, belong to the average degree of membership value of j Slack cluster centre for the Slack of workpiece in this scheduling problem example;

The 5.2.4 step: extract the machine process capacity characteristic

The present invention represents with the contained machine quantity of each operation in the scheduling problem example, that is:

Capability＝{SC ₁，…，SC _s}

Wherein

I=1 ..., s is the total quantity of i machine that operation comprises in this scheduling problem example, calculates { Problem successively ₁..., Problem _KThe Capability of all scheduling problem examples, and adopt the fuzzy C-means clustering method that above-mentioned Capability value is carried out cluster, establishing cluster centre quantity is C ^c, machinery processing capacity Capability cluster centre is designated as

J=1,2 ..., C ^c, the Capability of k scheduling problem example _kThe degree of membership that belongs to each center is designated as

J=1 ..., C ^c, the machinery processing capacity proper vector of scheduling problem example correspondence is:

[PC(1)，PC(2)，…，PC(C ^c)]

Wherein $\mathrm{PC}\left(j\right)={\mathrm{\&mu;}}_{\mathrm{kj}}^c,$ j＝1，…，C ^c；

The 5.3rd step: the upper strata cluster, it divides the scheduling problem example according to the scheduling problem characteristic information

In the clustering method of upper strata, the technology characteristics vector that the present invention obtained for the 5.2nd step, workpiece processing temporal characteristics vector, workpiece hand over phase tightness proper vector, machinery processing capacity proper vector to merge, total characteristic vector as a scheduling problem example is designated as:

X＝[PP(1)，…，PP(C ^p)，PM(1)，…，PM(C ^m)，PD(1)，…，PD(C ^d)，PC(1)，…，PC(C ^c)]

With above-mentioned vector as sample, then { a Problem in the clustering algorithm ₁..., Problem _KTotal K sample, use the fuzzy C-means clustering method to carry out cluster to these samples, note cluster centre quantity is C ^r, then resulting scheduling problem example cluster centre is

J=1,2 ..., C ^r, the degree of membership that k scheduling problem example belongs to each scheduling problem example cluster centre is designated as

J=1 ..., C ^r, the scheduling problem example is divided into C ^rClass makes in k scheduling problem example

Maximum classification is the affiliated type of this scheduling problem example, in case study on implementation of the present invention, the scheduling problem example is divided into 14 scheduling problem example types;

The 6th step: make up scheduling problem case library of the same type

In described scheduling rule excavation computing machine, call and make up similar scheduling problem case library module, make up scheduling problem case library of the same type, because the scheduling problem example quantity in the scheduling problem case library is limited after all, and the scheduling rule excavation needs a large amount of scheduling problem examples in order to estimate the quality of the scheduling rule of being excavated, so need be at each scheduling problem example types, initiatively produce the scheduling problem example of the same type of sufficient amount, the present invention is directed to each scheduling problem example types, produce 100 scheduling problem examples of the same type, its method is as follows: divide the result according to the scheduling problem that the 5th step was determined, at the scheduling problem example { Problem of scheduling problem case library preservation ₁..., Problem _KIn, optional scheduling problem example that belongs to current scheduling problem-instance type, and the correlated characteristic information of this scheduling problem example carried out random fluctuation, to produce a plurality of different scheduling problem examples of the same type, (α is at [α β) to note randu, β] uniform random number in the interval, the present invention gets α, and β is respectively 0.9 and 1.1, needs the information of change to have:

● workpiece quantity: the workpiece quantity after the variation is

● operate process time: the operation after the variation is process time

● workpiece delivery date: the workpiece after the variation is delivery date

● comprise number of machines in each machine group: comprising number of machines in each the machine group after the variation is $\tilde{N}\left(M_i\right)=N\left(M_i\right)\&times;p\left(\mathrm{0.9,1.1}\right)$

After the information change of above-mentioned scheduling problem example, promptly produced a new scheduling problem example, this example has been saved in the current scheduling problem-instance type of scheduling problem case library of the same type;

According to said method, can produce a plurality of different scheduling problem examples to each scheduling problem type, these scheduling problem examples all are saved in the scheduling problem case library of the same type;

The 7th step: make up scheduling rule parameter global coordination optimization problem

In described scheduling rule excavation computing machine, call and make up scheduling rule parameter global coordination optimization problem module, in this module, scheduling rule is defined as Rule={RS, RP}, wherein RS is the scheduling rule structure, and RP is the scheduling rule parameter sets under specific RS, and scheduling rule is input as operational attribute and machine group attribute two category informations, and be output as the processing priority of workpiece, wherein:

● the processing priority of workpiece is high more, and workpiece just should be processed more in advance;

● operational attribute has: process time, time of arrival, prediction time of arrival, haulage time, delivery date, residue process time, residue processing step number, subsequent operation number, subsequent handling bottleneck degree;

● machine group attribute has: comprise operation number to be processed in number of machines, machine group working ability, machine group bottleneck degree, the machine group, interior operation summation process time to be processed of machine group, estimate to arrive summation process time of operating;

The present invention adopts Adaptive Neuro-fuzzy Inference (ANFIS:Adaptive Neuro-Network Fuzzy InferenceSystem) in order to the expression scheduling rule, be above-mentioned workpiece attribute of being input as of ANFIS and machine group attribute, be output as workpiece processing priority, then the structure RS of scheduling rule is ANFIS, scheduling rule parameters R P is the parameter among the ANFIS, is designated as RP={rp ₁, rp ₂..., rp _ZExpression, wherein Z is the number of parameter among the ANFIS, on this basis, constructed scheduling rule parameter global coordination optimization problem is: coordination optimization RP={rp ₁, rp ₂..., rp _Z, so that by RP={rp ₁, rp ₂..., rp _ZPerformance the best of the scheduling rule determined, adopt said method, respectively at each scheduling problem example types, make up corresponding scheduling rule parameter global coordination optimization problem;

The 8th step: find the solution scheduling rule parameter global coordination optimization problem

In described scheduling rule excavation computing machine, call and find the solution scheduling rule parameter global coordination optimization problem module, the pairing scheduling rule parameter global coordination optimization of each scheduling problem example types problem is found the solution, this module adopts based on linearity division PSO (particle group optimizing) method each scheduling rule parameter global coordination optimization problem is found the solution, to obtain the scheduling rule corresponding with each scheduling problem example types, this method realizes as follows:

The 8.1st step: for reducing the number of parameter to be optimized, with parameter of regularity RP={rp to be optimized ₁, rp ₂..., rp _ZExpress with the slope and the displacement of a plurality of different line segments, at first the parameter among the RP is pressed parameter value scope [min (rp separately _i), max (rp _i)] normalization, i.e. rp ' _i=(rp _i-min (rp _i))/(max (rp _i)-min (rp _i)), i=1 ..., Z, obtain RP '=rp ' ₁, rp ' ₂..., rp ' _Z, in the present invention, the parameter value scope of ANFIS is decided to be [1000,1000], i.e. min (rp _i)=-1000, max (rp _i)=1000 simultaneously, suppose that slope and the displacement with Q bar line segment is k _j, b _j, j=1,2 ..., Q represents the parameter among the RP ', with Q=10 bar line segment, the denotable parameter set of j bar line segment is designated as Ie among the present invention _j, the division of above-mentioned parameter need be satisfied Ie ₁∪ Ie ₂∪ ... ∪ Ie _Q=RP ' and

I ≠ j, i=1 ..., Q, j=1 ..., Q, then the relation of parameter among the RP ' and Q bar line segment can be represented by the formula:

rp′ _(i)＝k _j×(i)+b _j-[k _j×(i)+b _j] ⁺， $r{p}_{\left(i\right)}^{\&prime;}\&Subset;S_j,$ j＝1，…，Q

The 8.2nd step: adopt the PSO method to k _j, b _j, j=1,2 ..., Q is optimized, with k _jAnd b _jAs decision variable, then each particle position is X={k in the PSO method _j, b _j| j=1,2 ..., Q} also can be expressed as X={x ₁, x ₂..., x _2Q, each particle's velocity is V={v ₁, v ₂..., v _2Q, to optimize in the iteration each, each particle upgrades by following formula:

$v_{\mathrm{id}}^{k+1}=w\&times;v_{\mathrm{id}}^k+c_1\&times;\mathrm{randu}\left(\mathrm{0,1}\right)\&times;({\mathrm{pbest}}_{\mathrm{id}}-x_{\mathrm{id}}^k)+c_2\&times;\mathrm{randu}\left(\mathrm{0,1}\right)\&times;({\mathrm{pbest}}_{\mathrm{gd}}-x_{\mathrm{id}}^k)$

$x_{\mathrm{id}}^{k+1}=x_{\mathrm{id}}^k+v_{\mathrm{id}}^{k+1}$

In following formula, i=1,2 ..., 2Q;

It is the d dimension component of particle i velocity in the k time iteration;

It is the d dimension component of particle i position vector in the k time iteration; Pbest _IdThe d dimension component of the desired positions that is arrived for particle i; Pbest _GdD dimension component for desired positions that population arrives; c ₁, c ₂Be weight factor, be respectively 0.5 in the present invention; W is the inertia weight function, is 3 in the present invention;

When the each iteration of PSO, all need calculate the pairing adaptive value F of particle current location (X), also promptly be quality, from X={x by the determined scheduling rule of particle current location ₁, x ₂..., x _2QCan obtain RP ', again according to min (rp _i) and max (rp _i), obtain RP, based on scheduling rule Rule={RS, RP} carries out industrial process simulation to the scheduling problem-instance, can obtain the scheduling index f of this rule correspondence, according to performance index f, can try to achieve particle position X={x ₁, x ₂..., x _2QPairing adaptive value;

The present invention makes the scheduling rule of being excavated to different scheduling problem examples of the same type dispatching effect preferably be arranged all, with X={x ₁, x ₂..., x _2QPairing scheduling rule acts on a plurality of similar scheduling problem examples, obtain a plurality of scheduling schemes, and adopt scheduling rule integrated evaluating method that the performance of above-mentioned given scheduling rule is carried out comprehensive evaluation, thereby obtain the adaptive value F (X) of X correspondence based on Bayesian Estimation;

This method is at first with X={x ₁, x ₂..., x _2QCorresponding given scheduling rule acts on h of the same type the scheduling problem example (h=100) that is obtained by the 6th step, obtains h scheduling index (f of above-mentioned scheduling problem example ₁, f ₂..., f _h), considering that above-mentioned scheduling problem example is of the same type, the present invention sees above-mentioned prior probability distribution by the scheduling index that produces under the same given scheduling rule effect as a normal distribution, R～N (μ, σ ²), according to the Bayesian Estimation theory, when an above-mentioned h scheduling index is (f ₁, f ₂..., f _h) time, when this scheduling rule was acted on above-mentioned a plurality of scheduling problem example of the same type, the posterior probability of the scheduling index that is obtained also satisfied normal distribution, promptly

$R|(f_1,f_2,\&CenterDot;\&CenterDot;\&CenterDot;,f_h)~N(\stackrel{\&OverBar;}{\mathrm{\&mu;}},{\stackrel{\&OverBar;}{\mathrm{\&sigma;}}}^2)$

Wherein $\stackrel{\&OverBar;}{\mathrm{\&mu;}}=w{\mathrm{\&Sigma;}}_{i=1}^h{f}_i/h+(1-w)\&CenterDot;\mathrm{\&mu;},$ w＝1/(1+se ²/σ ²)， $\stackrel{\&OverBar;}{\mathrm{\&sigma;}}=\sqrt{1/(1/s{e}^2+1/{\mathrm{\&sigma;}}^2)},$ $\mathrm{se}=\mathrm{\&sigma;}/\sqrt{h},$

According to above-mentioned probability distribution, the fiducial interval when fiducial probability is 0.9 can obtain by following formula:

Because above-mentioned probability distribution is normal distribution, can obtain by looking into the normal probability paper table, that is:

$a=1.63\&CenterDot;\stackrel{\&OverBar;}{\mathrm{\&sigma;}}$

So the scheduling problem example performance index of given scheduling rule correspondence are that 0.9 o'clock fiducial interval is at fiducial probability

The adaptive value that the present invention defines particle position is:

$F\left(X\right)=\stackrel{\&OverBar;}{\mathrm{\&mu;}}-1.63\&CenterDot;\stackrel{\&OverBar;}{\mathrm{\&sigma;}}$

In view of above-mentioned analysis, carry out the scheduling rule parameter global coordination optimization as follows, thereby excavate the pairing scheduling rule of each scheduling problem example types:

Step (8.2.1): each parameter Q of initialization, X, V, pbest and gbest, it is 1 that the current iteration number of times is set;

Step (8.2.2): according to particle more new formula upgrade particle position and speed;

Step (8.2.3): the pbest and the gbest that upgrade each particle according to the adaptive value computing formula of particle;

Step (8.2.4): judge whether to satisfy the algorithm stop condition, if the algorithm iteration number of times reaches K _Max, then stop, according to the gbest of particle, determine the scheduling rule parameter, thereby generate scheduling rule; As not satisfying, the current iteration number of times adds 1, returns the 8.2.2 step.

Images