CN1948559A - Design method of composite sample store and its system - Google Patents

Abstract

The invention offered a method to design combined library of samples, which facilitated people to reduce the frequency of sample experiments and increase effective amount of information in fixed frequency of sample experiments by applying acquired knowledge and given assumption extremely. The steps of this method were as follows: (1) Many components to compose samples were supplied. (2) Variables were offered for each component, which were sampled at intervals. (3) At least one constraint condition was set for at least one variable. (4) Pseudo-samples were produced. (5) Pseudo-samples were checked to confirm whether they were qualified samples. (6) Steps (4) and (5) were repeated until at least one qualified sample was found. The method of this invention was efficient and accurate, which could avoid system deviation in sample design.

Description

The method of design of combined sample library and system

Technical field

The present invention relates to a kind of test method efficiently-high-throughput experimental technique, more particularly, relate to the design field of combined sample library wherein.

Background technology

At the numerous characteristics of material, as heat conductivity, luminous, catalytic activity etc., all can utilize the discover method and the system of composition material, differentiate novel material or optimize existing material.Present combination research method by grid search, is synthesized a large amount of samples ' using rough power ' in sample space, screen these samples according to desired characteristic then.Yet this method is not almost considered the known experimental knowledge of the relevant component of sample.Even consider such experimental knowledge, also lack appropriate means design in sample space fully randomization sample library.

Therefore, be necessary to develop a kind of brand-new combination experiment design system and method, effectively experimental knowledge being incorporated in the sample library design, outside this experimental knowledge the sampling of sample then should be completely random to avoid the interference of human factor.In addition, we need know that needing the synthetic sample is which, the sum of representative sample information such as what are.

Summary of the invention

The object of the present invention is to provide a kind of brand-new combination experiment design system and method, experimental knowledge is incorporated in the design in sample storehouse effectively.

The invention provides a kind of method that experimental knowledge is come the composite design sample library of integrating.Particularly, this experimental knowledge can be presented as all component of sample, all the variable relevant with component and the constraint condition of these variablees.

The present invention provides a kind of method of the design sample storehouse that may further comprise the steps simultaneously:

(1) provides component in the target sample;

(2) set variable for each component;

(3) be at least one constraint condition of above-mentioned specification of variables;

(4) produce pseudo-sample library;

(5) choose the qualified samples that satisfies constraint condition in the pseudo-sample;

In one embodiment, the constraint condition of variable is by the relation between the variable of experience or known knowledge decision in the past.In yet another embodiment, experimental knowledge is physics or the chemical law of nature of being represented by variable.

In one embodiment, pseudo-sample can produce by random sampling.In one embodiment, above-mentioned random sampling is by adopting one group of variable, wherein the corresponding component of each variable and in certain intervals random value, carry out.An example of random sampling comprises Monte Carlo (Monte Carlo) simulation.In another certain embodiment, random value is by producing by random number generator.In yet another embodiment, random value is relevant with probability distribution or probability density.This probability distribution is uniform distribution (uniform distribution) or non-uniform Distribution.Wherein, non-uniform Distribution comprises Bernoulli distribution (Bernoulli distribution), beta distribution (beta distribution), X square distribution (Chi-square distribution), exponential distribution, F-distribution, gamma distributes, Gaussian distribution (Gaussiandistribution), normal distribution (lognormality for example, multivariate normal distribution and single argument normal distribution), non-central X square distribution, noncentral f distribution, binomial distribution, negative binomial distributes, polynomial expression distributes, Pareto distribution (Pareto distribution), Bai Song distribute (Poisson distribution), student's t-distribution and Sa Lisi distribution (Tsallis distribution).Probability distribution comprises uniform distribution, normal distribution and Gaussian distribution.

Another aspect of the present invention provides a kind of method that obtains the sample of desired number in sample library, and this method may further comprise the steps:

(1) provides all component forming sample;

(2) set a variable for each component;

(3) be at least one constraint condition of specification of variables;

(4) desired number of sampling;

(5) generation pseudo-sample at random;

(6) whether the variable according to pseudo-sample satisfies constraint condition, determines whether this puppet sample is qualified samples;

(7) repeating step (5) and (6) reach desired number until the quantity of qualified samples.

Again on the one hand, the invention provides a kind of calculate need design and/or the method for synthetic sample optimal number, this method may further comprise the steps:

(1) provides all component forming sample;

(2) set a variable for each component;

(3) be at least one constraint condition of specification of variables;

(4) between the given area, provide the required section of cutting apart for each variable;

(5) produce pseudo-sample;

(6) select to satisfy the pseudo-sample of constraint condition as qualified samples;

(7) determine the qualified samples ratio by qualified samples quantity divided by pseudo-sample size;

(8) quantity of calculation sample; And

(9) determine the optimal number of sample, wherein, the optimal number of sample can multiply by the qualified samples ratio by the quantity with sample and calculate and get.

This method further comprises the step of determining to be drawn divided by pseudo-sample size by qualified samples quantity the qualified samples ratio.

On the other hand, the present invention also provides a kind of computer product that comprises computer software.This computer software is in case operation can be carried out method of the present invention and calculating.For example.This computer software can carry out random sampling.

Brief description

Fig. 1 is the two component sample synoptic diagram that produce by the Monte Carlo simulation method, and each sample is made up of cerium (Ce) and iron (Fe).The pseudo-sample that cerium variable that all round dots among the figure (comprising hollow, grey, black) representative is produced by equally distributed, independent random separately and iron variable constitute; Between cerium variable and the iron variable without any constraint.The pseudo-sample of first constraint condition is satisfied in ash point and stain representative among the figure.The pseudo-sample (describing reference example 1 in detail) of the first and second two constraint conditions is satisfied in stain representative simultaneously.

Fig. 2 is that each sample is made up of cerium (Ce), iron (Fe), tungsten (W) and nickel (Ni) by the three-dimensional plot of four component sample of Monte Carlo simulation method generation.The pseudo-sample that representative is equally distributed separately by four variablees, independent random produces of having a few is therebetween without any constraint.

Fig. 3 is the pseudo-sample synoptic diagram that satisfies first constraint condition among Fig. 2.

Fig. 4 is the pseudo-sample synoptic diagram that further satisfies second constraint condition among Fig. 3.

Fig. 5 allows the user to design the graphic user interface (GUI) in multicomponent sample storehouse, and it provides a variable, range of variables and required compartmented for each component.

Fig. 6 is the graphic user interface that provides after a selected component and the variables corresponding.

Fig. 7 allows the user variable to be stipulated the graphic user interface of one or more constraint conditions.

Graphic user interface shown in Figure 8 allows the user to select 1) whether carry out the Monte Carlo simulation method; 2) how to carry out the Monte Carlo simulation method.

Graphic user interface shown in Figure 9 allows the user to import specified need sample number that obtains and the specified number of components of importing sample.

Graphic user interface shown in Figure 10 allows the user to specify each component.

Graphic user interface shown in Figure 11 allows the user to define constraint condition with variable.

Graphic user interface shown in Figure 12 allows the user to specify the constraint condition tolerance.

Embodiment

The present invention relates to a kind of layout strategy of combined sample library so that character designs per sample, synthesize, screen also measure sample storehouse.

One aspect of the present invention provides the method for design of sample library, comprises a plurality of components of sampling.Here said " combined sample storehouse " is meant the set that comprises a plurality of samples, and " sample " is meant the material that comprises various ingredients." component " is meant a kind of material, comprise as element, molecule, compound, material, thing piece etc., or the combination of these materials.

In an embodiment of the present invention, certain sample comprises the component that the n kind is different, C ₁, C ₂, C ₃... C _i... C _n, wherein n is an integer, refers to the quantity of different components in the sample.Each component C _iThe quality representation that is had is MW _i, wherein i ∈ 0,1,2...n}, composing quantity is expressed as X in the sample _i, corresponding composition ratiometer is shown R _iMass M W _iThe molecular weight or the nucleidic mass that refer to this component.So-called composing quantity X _iBe meant the quantity of i component in the sample, so this sample can be expressed as (C ₁) _X1(C ₂) _X2... (C _i) _Xi... (C _n) _Xn, wherein i ∈ 0,1,2...n}, composition ratio can be characterized by the relative weight quantity of a kind of component in the sample, its available formula 1 expression:

$R_i=\frac{({\mathrm{MW}}_i\×X_i)}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{MW}}_i\×X_i}$ Formula 1

Composing quantity X _iThe molar ratio that also can refer to i component in the sample.In this case, composition ratio also can be expressed as the molar fraction of a certain component in the sample, and its value can be defined by following formula 2 between 0 to 1:

$R_i=\frac{X_i}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{X}_i}$ Formula 2

Composition ratio also can further be expressed as a certain percentages of ingredients in the sample, and its value is between 0% to 100%.

In any sample in storehouse, whole composition ratio sums of all components are 1.As shown in Equation 3:

$\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{R}_i=1$ Formula 3

For example, glucose molecule C ₆H ₁₂O ₆Can regard the sample that comprises three components as: carbon (C), protium (H) and oxygen element (O), each component has composing quantity, as C be 6, H's is 12, O's is 6.The material mass of each component (MW) can be drawn by each atomic mass, and C is 12, and H is 1, and O is 16.Therefore, (weight) composition ratio of C is 0.4 or 40%, (12*6/ (12*6+1*12+16*6)); H is 0.067 or 6.7%; O be 0.533 or the summation of the composition ratio of three components of 53.3%. be 1.

Another feature of combined sample library is, each sample all is made up of the component of same type in the sample library, but these components have different composition ratios.

Each component that the method for design of the combined sample library that the present invention provides on the other hand is included as multicomponent sample provides a variable.In other words, component is corresponding one by one in variable and the sample.Suppose that variable V is at interval [V _Min, V _Max] in a random value, V wherein _MinBe not less than 0, V _MaxBe not more than 1, and V _Min≤ V _MaxIn one embodiment, this interval is [0,1].If supposing variable V is { V between discrete regions ₁, V ₂... V _xIn value, then V can disperse, wherein this discrete value drops on interval [V _Min, V _Max] in (as [0,1]).If set V is interval [V _Min, V _Max] random value, then it can be a successive value.When variable is relevant with component and do not have any constraint condition or all irrelevant with other variable, then the setting of the first variable random value is not subjected to the constraint of second variable supposition.If variable is a successive,, depend on the distribution probability or the probability density of this variable possibility value then from the interval interior process of setting random value of same variable.If this variable disperses, then the particular probability of discrete value is separately depended in the interval in the setting of random value.

For example, V _iBe the first component C _iVariable, V _jBe component the 2nd C _jVariable.V _iCan be set at [V _{I, min}, V _{I, max}] interval random value, V _jCan be set at [V _{J, min}, V _{J, max}] interval random value, V _iValue and V _jIrrelevant.Work as C _i, C _jBe by C ₁, C ₂... C _i... C _j... C _nComponent in the sample of forming, I wherein, j ∈ 0,1,2...n}, this compositional variable V _iBecome component C _iComposition ratio R _i, compositional variable V _jBecome composition ratio R _j, and the variable summation of all components satisfies following formula 4 in the sample:

$\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{V}_i=1$ Formula 4

Another aspect of the present invention is: provide or set at least one constraint condition at least one variable in the sample in the method for design of the combined sample library that is provided." constraint condition " speech refers to the condition of at least one variable or the relation between the variable.Particularly, constraint condition is the restricted condition that a variable or a plurality of variable must satisfy in the sample.In other words, one group of value { V in effective or qualified sample _iVariable must satisfy at least one constraint condition or one group of specific constraint condition.For example, suppose that sample comprises component C ₁, C ₂... C _n, and each component C _iHas variable V _i, wherein i ∈ 0,1,2...n}, so, in effective sample, the summation of component variable must satisfy following constraint condition, formula 5:

$1-\mathrm{\Δ}\≤\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{V}_i\≤1+\mathrm{\Δ}$ Formula 5

Wherein Δ is error (as constraint tolerance or a constraint deviation), and Δ is the value that changes between 0 to 0.2.In preferred embodiment, Δ is 0.01,0.02,0.03,0.04,0.05,0.06,0.07,0.08,0.09, or 0.10.When Δ approaches 0, this constraint approaches equation as shown in formula 5.

In addition, in the sample library design, can be presented as about the experimental knowledge of the component of sample relation between a plurality of variablees also to can be understood as constraint.In other words, we can be by setting the relation that retrains between several components that realize in the experimental knowledge.For example, according to experience before, by C ₁, C ₂... C _nIn the sample of forming, C perhaps _iWith C _jThe ratio of composition ratio should be 2: 1, wherein i ∈ 0,1,2...n}, j ∈ 0,1,2...n}, and i ≠ j.At this moment, except that the constraint of the inherent shown in the formula 4, the variable of the component of an effective sample must satisfy the second constraint V _i: V _j=2: 1.Again for example, reach C _iAnd C _jThe composition ratio sum be x, wherein x is the value between 0 to 1.Herein, the variable of the component of an effective sample must satisfy the second constraint V _i+ V _j=x, i wherein, j ∈ 0,1,2...n} and i ≠ j.

The method of design of combined sample library provided by the invention comprises the pseudo-sample of generation." pseudo-sample " is meant a multi-component supposition sample, its each component all have one independently variable make that any assignment of some variablees is all the be under no restraint restriction of condition of entire variable with respect to the independent events of any assignment of another variable and pseudo-sample.In other words, its variable can satisfy or not satisfy constraint condition.For example, a pseudo-sample comprises C ₁, C ₂... C _i... C _j... C _n, i wherein, j ∈ 0,1,2...n} and i ≠ j, V _iAnd V _jBe the random value in [0,1] interval, V _iSome values between desirable [0,1], V _jAnother value between desirable [0,1].Requirement these variate-values and that need not coincidence formula 4 and 5.

Pseudo-sample can correspondence also can not corresponding real physical sample, pseudo-sample is the sample spot of independently supposing variable-value without any constraint.Therefore a large amount of pseudo-samples are formed a sample space.

Pseudo-sample can adopt random sampling to produce.Random sampling methods is a kind of by being the method that each component random assignment of a sample spot produces sample spot.At interval [V _Min, V _Max] in produce random number method be clear and definite, V wherein _MinBe not less than 0, and V _MaxBe not more than 1.Please refer to Carter " generation of random digit and application (fourth dimension) " ( " The Generation and Application of Random Numbers(Fourth Dimensions) ", Vol.XVI, 1994).The algorithm of random number generator and computation program are that computer science is known, please refer to " art-seminumerical algorithm of computer program " (the 2nd volume, A Disen-Wei Sili second edition published in 1981) of D.E.Knuth (" The Art of Computer Programming-Seminumerical Algorithms" Vol.2,2 ^NdEd.Addison-Wesley, 1981); " art that numerical method prescription-science is calculated " of work such as Press (" Numerical Recipes:The Art of Scientific Computing "Cambridge University Press, 1986; And " Numerical Recipes (FORTRAN) ", 191-225 page or leaf, 1988); And S.L.Anderson show " random number generator on vectorial supercomputer and other advanced system " (" Random Number Generators on Vector Supercomputers and Other Advanced Architectures ", SIAM Rev,, 32:221-225, nineteen ninety).

At random the numerical value of Chan Shenging be the generation of incident or between given zone [V _Min, V _Max] in the variable (V) of possible assignment, wherein the probability that takes place of this incident depends on the probability density or the probability distribution of this variable.Therefore, variable is further defined by the assignment of probability function to the value that the range of variables comprised.For example, discrete variable can specify dependent probability to define by giving each discrete value in the interval.Continuous variable can specify probability distribution to define by giving the interval that comprises all variablees possibility values.

Here the probability distribution that is adopted refers to reflect the arrangement of the variable-value of its observation or the theoretical frequency of occurrences.The probability distribution of knowing in this technical field comprises that equiblibrium mass distribution and lack of balance distribute.Lack of balance distributes and comprises that Bernoulli distribution, beta distribution, X square distribution, exponential distribution, F-distribution, gamma distribution, Gaussian distribution, normal distribution (for example, lognormality, multivariate normal distribution and single argument normal distribution), non-central X square distribution, noncentral f distribution, binomial distribution, negative binomial distribution, polynomial expression distribution, Pareto distribution, Bai Song distribution, Xue Shengshi t-distribution, Sa Lisi distribute and any associating of above distribution.

In one embodiment, stochastic variable for example, comprises that normal distribution, Bai Song distribute and Gaussian distribution by lack of balance probability distribution assignment.In another embodiment, the generation value at random of variable is by equiblibrium mass distribution assignment or relevant with it, and therefore, the stochastic variable of equiblibrium mass distribution can be at interval [V _Min, V _Max] (V _Min〉=0, V _Max≤ 1) supposes any random value with equal probabilities.

Persons skilled in the art as can be known, lack of balance distribution random value (or number) can produce (as the superimposed producer of linearity) by random number generator.The general formula of the superimposed producer of this linearity is V _i=(aV _I-1+ c) mod m, wherein a, c and m are predefined constants, and a is a multiplier value, and c is an increment, and m is a coefficient.Please refer to Park and Miller " random number generator: good difficulty is looked for " (" Random Number Generators:Good Ones are Hard to Find ", Comm.ACM 31:1192-1201,1988).Random number generator comprise " the transferring the register series random number generator fast " that Kirkpatrick and Stoll show (" A Very Fast Shift-Register Sequence Random Number Generator ", Journal ofComputational Physics 40:517-526,1981) described in the transfer register series.In addition, random number generator also comprises accurate number generator at random, please refer to Press and Teukolsky " quasi random number " (" Quasi Random Numbers ", Computers in Physics 3:76-79,1989).

Lack of balance distribution random value as normal distribution or Gaussian distribution random value, also can produce by the method that association area is known.Please refer to Robinstein " simulation and Monte Carlo method " (Rubinstein, " Simulation and the Monte Carlo Method "By John Wiley ﹠amp; Sons published 1981).One of method comprises the conversion function, as famous Bock Shi Mole conversion, in order to the equiblibrium mass distribution stochastic variable is converted to the stochastic variable (for example, Gauss or normal distribution) that new one group of lack of balance distributes, please refer to " note that departs from generation at random " (Box of Bock Shi Mole; Muller, " A Note on the Generation of Random Deviates ", Annals Math.Stat.29:610-611,1958).

In one embodiment, random sampling methods comprises Monte Carlo method or simulation.Here " Monte Carlo method " or " Monte Carlo simulation " speech refer to a kind of of stochastic technique or the random sampling methods of the problem probability approximation that is used for studying a question and achieve a solution.Especially, used here " Monte Carlo method " or " Monte Carlo simulation " speech is meant the process (as the occurrence value at random of any given variable) that produces random occurrence especially.This process is reached by computer algorithm usually, and this process repeats repeatedly, and analyzes and calculate all test-results in order to approximate answer to be provided.Monte Carlo simulation please refer to Mi Teluo pool rein in " Monte Carlo method " (the periodical 44:335-341 of U.S. statistical association, 1949) of this and Wu Lamu (Metropolis and Ulam, " The Monte Carlo Method ", Journal of American StatisticalAssociation 44:335-341 1949); " Monte Carlo method " of longevity Charles Bell (Sobol, " The Monte Carlo Method ",The University of Chicago Press, 1974); Solemn Buddhist nun's " Monte Carlo simulation " (Mooney, " Monte Carlo Simulation ",Sage University Paper, 1997).

Monte Carlo method is constantly developed in this area.For example, this method is applied to dart on the numerical value of estimating π by going up to throw at standard coordinate (by the circumscribed circumference of square) at first.By a large amount of tests, find that dartlike weapon hits circumference and foursquare quantity is proportional with circumferential area and area respectively, and have suitable tolerance range.Correspondingly, the ratio that dartlike weapon hits circumference and square number of times is similar to the mark of π value, please refer to Luo Si " probability Lesson One " (Ross, " A First Course in Probability" 2 ^NdEdition, Macmillan, 1976).

Another example, Monte Carlo simulation can be applicable to estimate following integral formula 6:

$N={\∫}_{i,\min }^{i,\max }V\left(x\right)\mathrm{dx}$ Formula 6

In this embodiment, function V (x) has a scope frame on every side, and the integration of V (x) can be regarded as in the scope frame part at V (x).If choosing at random and non-homogeneous of scope frame mid point, the probability that is arranged in V (x) is so then determined in the shared area portions of frame by V (x).So Monte Carlo simulation produces a large amount of random points (occurrence value at random) and calculates V (x) mid point in frame quantity is to obtain area.As a result of, the integration of formula 6 can be expressed as following formula 7:

$N\≈\frac{A}{B}C$ Formula 7

Wherein A is the quantity of V (x) mid point, and B is the quantity of being had a few that produces in the frame, and C is the area of scope frame.In addition, ratio A/B is relevant with V (x) relative area occupied ratio in the scope frame.

Another example of Monte Carlo simulation comprises the random variable values that generation is stressed by experimental knowledge.For example, about the experimental knowledge of component (or component ratio) require to give variable between different given zone with different probability density (continuous variable), or require to give variable in different values with different value probability (discrete variable).The example of another Monte Carlo simulation comprises the Markov chain computing.The markov computing is a random value sequence, and the probability that its each incident takes place depends on the value that results from previous moment.Please refer to " understand molecular simulation: from algorithm to using " (Frenkel ﹠amp of Frank and Smith; Smith, " Understanding Molecular Simulation:From Algorithm to Applications "Academic Press, 1996).

The present invention is about select the method for qualified samples from pseudo-sample on the other hand." qualified samples " speech here refers to that variable satisfies the pseudo-sample of one or more particular constraints by what method produced described in the present invention, and the pseudo-sample of non-qualified samples is called as failed test sample.In the one embodiment of the invention, pseudo-sample produces (as Monte Carlo simulation) by random sampling methods, promptly in a large amount of tests, at interval [V _Min, V _Max] (V _Min〉=0 and V _Max≤ 1) value that produces at random by uniform distribution in is given to the fractions variable, the pseudo-sample that is under no restraint with generation.Each pseudo-sample, is determined it and whether satisfies specific one or more constraints for example with a computerized algorithm by checking (investigation).Selecting the pseudo-sample that satisfies constraint stores as qualified samples.Simultaneously, all the value relevant with each qualified samples is recorded into a vector, and is mapped with the component ratio, so that synthesize in sample library and the design qualified samples, because in qualified samples, those are worth all just component ratios.

Another aspect of the present invention provides a kind of method for the sample that produces at sample library to determined number.This method may further comprise the steps:

(1) provides the fractions of forming sample;

(2) give each component one variable;

(3) be at least one constraint condition of specification of variables;

(4) provide needed sample size;

(5) produce pseudo-sample;

(6), determine that this puppet sample is a qualified samples if the variable of pseudo-sample satisfies this constraint condition;

(7) repeating step (5) and (6) reach desired number until qualified samples quantity.

The present invention has disclosed a kind of method of calculating the qualified samples ratio on the other hand, the " qualified samples ratio (R here _Qs) " speech is meant that in random sampling methods, variable satisfies the ratio of the pseudo-sample of one or more constraints.In one embodiment, qualified samples ratio (R _Qs) can be by qualified samples quantity (N in the random sampling methods _Qs) divided by pseudo-sample size (N _Ps) estimate (formula 8).

$R_{\mathrm{qs}}\≈\frac{\mathrm{Nqs}}{\mathrm{Nps}}$ Formula 8

Work as N _PsIncrease, counting accuracy diminishes, and rule change is followed following formula 9:

$\mathrm{Accuracy}~\±\sqrt{\frac{1}{N}}$ Formula 9

Wherein N is stochastic simulation (if you would a special Caro) experiment quantity.After having carried out a large amount of Monte Carlo simulation tests, along with 1/N constantly reduces, the change of qualified samples ratio reduces and tolerance range increases.In other words, after the test of carrying out sufficient amount, the ratio of qualified samples can reach quite high tolerance range and accuracy.For example, with regard to constraint condition $1-\mathrm{\Δ}\≤\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{V}_i\≤1+\mathrm{\Δ},$ Sample (the V of Monte Carlo simulation _i) many more, tolerance range is high more.

In one embodiment, produce at the random number generator (C++ compiler 7.1.3091 version, 2003) that adopts Microsoft in the Monte Carlo experiment operation of random digit function, observe and obtain a qualified samples (as N _QsThe tolerance range of calculating=1) reaches-100% to 100%; Work as N _QsReach at 10 o'clock, its tolerance range is between-30% to 30%; Work as N _QsReach 10 ², its tolerance range is between-10% to 10%; Work as N _QsReach 10 ³, its tolerance range is between-3% to 3%; Work as N _QsReach 10 ⁴, its tolerance range is between-1% to 1%.

On the other hand, the present invention has also disclosed the method for estimating the qualified samples optimal number." qualified samples optimal number " speech here refers to, satisfies specific one or more constraint and can show the sample number of the random sampling of sample space rightly.

In one embodiment of this invention, the qualified samples optimal number obtains by detecting all pseudo-samples that satisfy particular constraints by the possible pseudo-samples and the identification of discrete variable generation.For to discrete variable, by with interval [V _{I, min}, V _{I, max}] cut apart (evenly or anisotropically) and become M part or lattice to give particular variables V _iProduce one group of discrete value, thus produce one group should the interval definition value.If the interval evenly is divided into the M part, the discrete value of variable can be { V _iMiddle arbitrary value, wherein V _i=V _{I, min}+ l* (V _{I, max}-V _{I, min})/M _i, l ∈ 0,1,2 ... M}.M is positive integer and can is 1 to 1,000, any number between 000.In one embodiment, M ∈ 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16 17,18 ... 20..25...30...40..50..10 ².10 ³..10 ⁴..10 ⁵.10 ⁶.

If variable only is set at the discrete values in the interval, the sum of pseudo-sample (Z) can be produced by the quantity of the particular grid of variable.For example, comprising component C arbitrarily ₁, C ₂... C _nSample in, each component C _iHas variable V _i, wherein i ∈ 0,1,2...n}, each V _iBe separated into M _iPart or grid or point, so V _iBe taken at interval [V _Min, V _Max] (V _Min〉=0 and V _Max≤ 1) one group of discrete value.According to us the experience of the pairing variable of component is determined this separation, it can be uniform segmentation or segmentation heterogeneous.If do not consider or do not provide the constraint condition of variable, based on one group of { M _iPseudo-total number of samples amount (Z) represented the sample spot of n dimension sample space, Z can calculate by following formula 10:

$Z=\underset{i=1}{\overset{N}{\mathrm{\Π}}}{M}_i$ Formula 10

When given at least one when constraint, can detect all pseudo-samples, select variable and satisfy the pseudo-sample of this constraint and store into and constitute the qualified samples collection in the vector.After whole sample spaces (whole Z pseudo-sample) were detected, the quantity of qualified samples, that is the number of above-mentioned vector were exactly the optimal number corresponding to qualified samples in the sample space of given one group of discrete value.

If relevant component (or variable) quantity increases, and the number of fragments of each component also increases, and then the search fully by sample space can become quite heavy.For example, to being separated into the sample library that 100 lattice are constituted between each given zone by five components and every component variable, sample space is a quintuple space, and the sum of all pseudo-samples (Z) is 100 ⁵(or 10 ¹⁰, 10,000,000,000).When introducing one or more constraints, it is more complicated that calculating can become.Although can discern it and whether satisfy these one or more constraints by detecting each pseudo-sample, can adopt other method, promptly provide the approximate evaluation of optimal values by carrying out random sampling described here (as Monte Carlo simulation).

Therefore, in one embodiment, we produce pseudo-sample by stochastic simulation, detect this puppet sample and whether satisfy described constraint, obtain qualified samples and qualified samples ratio according to method of the present invention.In this stochastic simulation, randomized number can be based on that one group of discrete value produces, and wherein each discrete value is arranged in the interval that variable supposes and has certain probability.Randomized number also can be any value that particular probability distributes that has in corresponding a certain interval.Therefore, the optimal number of qualified samples depends on Z and R _QsProduct.Can expect that the optimal number of qualified samples has change, this depends on one group of parameter.The variable in the sample of comprising for example of this parameter is cut apart (M) number, the method that produces random digit, statistical distribution state, Monte Carlo simulation mode, Monte Carlo experiment number of times, variable bound, tolerance limits and the required accuracy or the precision of variable.

Be appreciated that the generation of the randomized number that random sampling (as Monte Carlo simulation) is relevant with probability distribution, the selection and the calculating of the qualified samples of doing according to method provided by the present invention are all carried out by computer system or server system usually.

Computer system among the present invention (as server system) is meant design and is disposed for carrying out the computer or the computer-readable media of part or all of method described in the invention.Here the computer of Cai Yonging (as server) can be the computer of any broad variety general applications, as PC, the webserver, workstation or other computer platform of development now or in the future.Known in the art, computer include especially as treater, operating system, computer memory, input unit and these parts of output equipment partly or entirely.Computer can further comprise as cache memory, data backup unit and some miscellaneous equipments.Persons skilled in the art are appreciated that these machine elements can have many other possible structures.

But the treater that is adopted can comprise one or more microprocessors territory programmed logic array (PLA) here, or one or more special unicircuit corresponding to the special type application.For example, but treater includes but not limited to the Pentium series processors of Intel company, the microprocessor of Sun Microsystems, the workstation system treater of Sun Microsystems, the personal desktop machine treater of Motorola Inc., the MIPs treater of MIPS Science and Technology Ltd., the highest series territory programmed logic array (PLA) and some other treater of Xilinx company.

Here the operating system that is adopted comprises machine code, by the execution of treater, can coordinate and object computer in the function of other parts, and help treater to carry out the function of the various computing machine program that may write with multiple programming language.The data stream of other parts, operating system also provides scheduling arrangement, input and output control, file data management, memory management and Communication Control and related service in supervisory computer, and all these is a prior art.Unix that typical operating system comprises Windows as Microsoft, provided by all multi-providers or (SuSE) Linux OS, other or the operating system of development in the future, and the combination of these operating systems.

Here the computer memory that is adopted dissimilar memory storage arbitrarily.For example comprise optical mediums such as magneticmedium storages such as the random access memory that is seen everywhere, permanent hard disk or tape, read-write compact disc, or other accessing storage devices.Memory storage can be any one device existing or that develop in the future, comprises compact disc driving mechanism, tape drive, removable hard disk drive or disc driver.The memory storage of these types generally is to be read or written to from computer program memory medium in this medium, as CD, tape, removable hard disk or floppy disk.All these computer program memory mediums can be considered to the product of computer program.The product of these computer programs is stored computer software programs and/or data usually.Computer software programs generally are stored in system memory and/or the memory storage.

Persons skilled in the art readily understand, thereby the computer software programs among the present invention can be by coming execution in loading system storer and/or the memory storage with certain input unit.On the other hand, this software program of all or part also can be present in read-only storage or the similar memory storage, and such device does not need this software program at first to be written into by input unit.Those possessing an ordinary skill in the pertinent arts are appreciated that this software program or its some part can be loaded into system memory or cache memory or the combination of the two by treater by existing manner, are beneficial to carry out and carry out random sampling.

In one embodiment of this invention, software is stored in the computer server, and this computer server is connected with use terminal, input unit or output equipment by data line, radiolink or network system.The present technique field is generally known, and network system is included in the hardware and software that is electrically connected in computer or the device.For example network system can comprise the equipment on the medium basis of internet, 10/1000 Ethernet, the 802.11x of electric electronic engineering association, electric electronic engineering association 1394, xDSL, bluetooth, local area network, WLAN (wireless local area network), GSP, CDMA, 3G, PACS or any other ANSI Recognized Standards.

General description is carried out to the present invention in the front, next exemplifies some certain embodiments and further describes, to help understanding the present invention.

Embodiment 1

How present embodiment shows from the pseudo-sample of being made up of two components (cerium and iron) that the Monte Carlo simulation mode produces selects qualified samples.The variable V of cerium _CeValue between 0 to 1, the variable V of iron _FeIt equally also is value between 0 to 1.Monte Carlo simulation adopt equally distributed produce at random 0 to 1 between V _CeAnd V _FeValue is carried out, V in this simulation _CeGeneration value at random and V _FeGeneration at random value separate.And in this simulation, V _CeAnd V _FeThe pressure that there is no any relation or constraint limits.The result of Monte Carlo simulation has produced pseudo-sample group.The pseudo-samples set of whole formations of point (comprising hollow, grey and dark color) as shown in fig. 1.

We can utilize experimental knowledge to reduce the number of qualified samples, and this realizes by introducing constraint condition.First constraint condition is defined as 0.2＜V _Ce＜0.8 and 0.2＜V _Fe＜0.8.After the selection process had been considered first constraint, one group of pseudo-sample selecting was shown as stain or grey point as shown in Figure 1.Second constraint definition is 1-Δ＜V _Ce+ V _Fe＜1+ Δ when further contemplate this second constraint condition on the basis of first constraint condition, is selected simultaneously the one group of pseudo-sample that satisfies two constraint conditions and is shown as shown in Figure 1 stain.

Therefore, when designing the sample library of being made up of Ce and Fe, this Monte Carlo simulation has been introduced design by these two constraints with experience, and has obtained definite information in design sample storehouse.As, since satisfy relevant digital known of the pseudo-sample size of two constraints, the quantity of qualified samples can be known so.As shown in Table I, our component ratio of each qualified samples as can be known.Table I shows the pseudo-sample value that produces by Monte Carlo simulation, is the pseudo-sample value that satisfies first constraint with the italics display digit, and the numeral in the square frame is to satisfy the pseudo-sample value of first and second constraints.

If variable is cut apart by specific lattice point, the optimal number of sample can obtain according to method provided by the present invention.

V in the Table I Monte Carlo simulation _CeAnd V _FeValue

Fe Ce

0.040000 0.160000

0.120000 0.340000

0.440000 0.120000

0.160000 0.040000

0.460000 0.020000

0.460000 0.040000

0.460000 1.000000

0.560000 0.060000

0.860000 0.040000

0.960000 0.020000

0.000000 0.180000

0.000000 0.500000

0.000000 0.720000

0.600000 0.000000

0.760000 0.140000

1.000000 0.260000

1.000000 0.760000

0.180000 0.020000

0.280000 0.080000

0.820000 0.620000

0.820000 0.660000

0.840000 0.440000

0.960000 0.220000

0.820000 0.740000

0.820000 0.780000

0.860000 0.320000

0.040000 0.620000

0.060000 0.460000

0.120000 0.880000

0.180000 0.280000

0.320000 0.880000

0.420000 0.880000

0.940000 0.680000

0.020000 0.960000

0.020000 0.980000

0.520000 0.940000

0.720000 0.980000

0.820000 0.940000

0.840000 0.740000

0.880000 0.320000

0.880000 0.360000

0.980000 0.320000

0.980000 0.340000

0.540000 0.840000

0.180000 0.580000

0.240000 0.980000

0.340000 0.920000

0.880000 0.580000

0.980000 0.520000

0.360000 0.820000

0.560000 0.840000

0.860000 0.880000

0.560000 0.960000

0.860000 0.920000

0.880000 0.760000

0.580000 0.860000

0.680000 0.960000

0.240000 0.200000

0.440000 0.200000

0.540000 1.000000

0.740000 0.060000

0.100000 0.560000

0.100000 0.640000

0.200000 0.160000

0.500000 0.980000

0.700000 0.120000

0.700000 0.840000

0.800000 0.140000

0.800000 0.280000

0.900000 0.800000

0.060000 0.200000

0.360000 0.100000

0.360000 0.200000

0.860000 0.100000

0.860000 0.600000

0.960000 0.700000

0.380000 0.100000

0.480000 0.800000

0.120000 0.080000

0.520000 1.000000

0.720000 1.000000

0.920000 0.020000

0.020000 0.180000

0.620000 0.160000

Embodiment 2

How present embodiment shows from the pseudo-sample of being made up of four components (cerium, iron, tungsten and nickel) that the Monte Carlo simulation mode produces selects qualified samples.The variable V of cerium, iron, tungsten and nickel _Ce, V _Fe, V _W, V _NiValues between 0 to 1 all.We adopt equally distributed 0 to 1 the numerical value that produces at random, the separate and qualification that is under no restraint of the generation value at random in this simulation to each variable in the Monte Carlo simulation.The result of Monte Carlo simulation is the sample spot (pseudo-sample) of four-dimentional space, and this four-dimension sample spot projection in three-dimensional space as shown in Figure 2.

For selecting to have proposed first constraint condition here corresponding to real sample physically, it is defined as V _Ce+ V _Fe+ V _W+ V _Ni=1.When considering first constraint in the selection process, the one group of pseudo-sample of selecting (the first pseudo-sample) that satisfies this first constraint condition is presented among Fig. 3.

Certain experience of imagination makes us conclude that the component sum of cerium and iron always equals the component sum of tungsten and nickel, and we can introduce second constraint so, and it is defined as V _Ce+ V _Fe=V _W+ V _NiWhen further contemplate this second constraint condition on the basis of first constraint condition, the one group of pseudo-sample (the second pseudo-sample) that satisfies these two constraint conditions when selecting is shown as on the two dimensional surface that is dispersed in three confining spaces (as shown in Figure 4).This shown qualified samples point is to observe from a side of two dimensional surface among Fig. 4.

Therefore, this Monte Carlo simulation of considering two constraints provides about the definite information of design by four components sample library that form, that possess these two constraints.For example, the qualified samples ratio can calculate by removing the pseudo-sample size that satisfies two constraints with pseudo-sample all amts.Since it is digital known that the pseudo-sample size of satisfied two constraints is correlated with, the quantity of qualified samples can be known so.Write down the component ratio of variable in each qualified samples, as can be known the component ratio of this qualified samples.If variable is cut apart by specific lattice point, the optimal number of sample can obtain according to method provided by the present invention equally.

In above-mentioned Monte Carlo simulation, obtain whole 28561 pseudo-samples, wherein 460 are satisfied first constraint, and 47 are satisfied two constraints simultaneously, and the qualified samples ratio that therefore should the puppet sample satisfies two constraints is 0.0016456.

Embodiment 3

Present embodiment has illustrated that the permission user calculates and simulate (comprising Monte Carlo simulation) computer program with design qualified samples storehouse by graphic user interface input information and execution.

As shown in Figure 5, graphic user interface allows the user to select the required component of design sample.For example, a sample of being made up of component A, B and C, component A can be any one from the element set of being made up of vanadium (V), niobium (Nb) and molybdenum (Mo), the variable (V of component A _a) variation range (please refer to the scope of 0.00 shown in Fig. 5 to 1.00) between 0 to 1, this variable change scope is divided into 10 parts (as shown in Figure 5 10 sections).Consequently, component A is endowed the variable (V of value between 0 to 1 _a) (please refer to Fig. 6), similarly, B component and C also are endowed variables corresponding (V _bAnd V _c) (please refer to Fig. 6).

As a kind of mode of experimental knowledge being included in the sampling design, graphic user interface allows the user that the constraint condition of specifying between a variable or a plurality of variable is provided.Variable V _a, V _bAnd V _cGive tacit consent to or the first hiding constraint is V _a+ V _b+ V _c=1 ± Δ.Δ is error (or constraint tolerance), and is given as 0.01 (please refer to Fig. 6) in this example.The second required constraint is V _a: V _b=2: 1, and by graphic user interface input (please refer to Fig. 7).

Graphic user interface further allows the user to determine how to estimate the optimal number of qualified samples.As shown in Figure 8, graphic user interface provides six other calculating of different class of accuracies.In the accurate calculation therein, pseudo-sample produces according to the formula 10 of no any constraint shown in the present, should also only select to satisfy the part that retrains by the puppet sample by COMPUTER DETECTION then.In this example, 198 samples satisfy constraint.In addition, also obtained the component ratio of all 198 qualified samples.

When counting accuracy is considered to very low-100% between 100% the time, when it thinks low between-30% to 30%, tolerance range-10% to 10% is medium, and is high between-3% to 3%, then is very high tolerance range between-1.0% to 1.0%.

Embodiment 4

Present embodiment has shown the computer program that obtains the qualified samples point of specified quantity.This computer program allows the user to calculate and simulation (comprising Monte Carlo simulation) by graphic user interface input information and execution, draws all component ratios of each sample spot in these sample spot with this.

Fig. 9 display graphics user interface allows the user to import specified gross sample point 125, and wherein each sample spot all has 4 components.The component (please refer to Figure 10) that graphic user interface also allows the user to specify to want, and the constraint condition of defined variable (please refer to Figure 11).After definition constraint tolerance (please refer to Figure 12), begin to carry out simulation test, each pseudo-sample all checks whether satisfy set constraint therebetween.When the quantity of qualified samples reaches 125, stop simulation, obtain the component ratio (as shown in Table II) of four components (element Pd, Pt, Ce and V) in 125 sample spot with this.

Table II

Pd Pt Ce V

0.039474 0.105263 0.039474 0.815789

0.065789 0.118421 0.144737 0.671053

0.078947 0.171053 0.684211 0.065789

0.052632 0.144737 0.776316 0.026316

0.078947 0.184211 0.684211 0.052632

0.013158 0.118421 0.723684 0.144737

0.026316 0.157895 0.697368 0.118421

0.039474 0.105263 0.723684 0.131579

0.052632 0.105263 0.407895 0.434211

0.052632 0.131579 0.697368 0.118421

0.065789 0.105263 0.552632 0.276316

0.065789 0.118421 0.381579 0.434211

0.065789 0.144737 0.381579 0.407895

0.013158 0.184211 0.407895 0.394737

0.026316 0.144737 0.552632 0.276316

0.039474 0.184211 0.381579 0.394737

0.052632 0.105263 0.355263 0.486842

0.065789 0.144737 0.210526 0.578947

0.013158 0.157895 0.815789 0.013158

0.039474 0.157895 0.644737 0.157895

0.078947 0.105263 0.671053 0.144737

0.105263 0 0.644737 0.25

0.184211 0 0.065789 0.75

0.065789 0.131579 0.421053 0.381579

0.078947 0.157895 0.302632 0.460526

0.078947 0.184211 0.447368 0.289474

0.105263 0 0.75 0.144737

0.184211 0 0.315789 0.5

0.026316 0.144737 0.539474 0.289474

0 0 0.302632 0.697368

0.039474 0.157895 0.407895 0.394737

0.078947 0.144737 0.026316 0.75

0.078947 0.144737 0.276316 0.5

0.052632 0.144737 0.552632 0.25

0.013158 0.026316 0.776316 0.184211

0.013158 0.092105 0.618421 0.276316

0.013158 0.092105 0.710526 0.184211

0.092105 0.092105 0.447368 0.368421

0 0.171053 0.526316 0.302632

0.039474 0.118421 0.75 0.092105

0.039474 0.013158 0.513158 0.434211

0.065789 0 0.092105 0.842105

0.171053 0.013158 0.210526 0.605263

0.171053 0.039474 0.184211 0.605263

0.026316 0.013158 0.342105 0.618421

0.171053 0.026316 0.144737 0.657895

0.171053 0.092105 0.394737 0.342105

0.184211 0.013158 0.184211 0.618421

0.184211 0.092105 0.592105 0.131579

0.013158 0.052632 0.605263 0.328947

0.013158 0.078947 0 0.907895

0.039474 0 0.052632 0.907895

0.026316 0.092105 0.736842 0.144737

0.184211 0.052632 0.434211 0.328947

0.184211 0.092105 0.684211 0.039474

0.026316 0.013158 0.105263 0.855263

0.065789 0.026316 0.776316 0.131579

0.105263 0.039474 0.342105 0.513158

0.118421 0.131579 0.039474 0.710526

0.184211 0.013158 0.592105 0.210526

0.013158 0.013158 0.407895 0.565789

0.013158 0.039474 0.565789 0.381579

0.052632 0.026316 0.342105 0.578947

0.065789 0.092105 0.815789 0.026316

0.078947 0.013158 0.881579 0.026316

0.105263 0.052632 0.131579 0.710526

0.144737 0.026316 0.184211 0.644737

0.144737 0.092105 0.118421 0.644737

0.157895 0.184211 0.026316 0.631579

0.171053 0.092105 0.578947 0.157895

0.184211 0.026316 0.052632 0.736842

0.026316 0.039474 0.552632 0.381579

0.026316 0.092105 0.315789 0.565789

0.052632 0.026316 0.723684 0.197368

0.052632 0.052632 0.592105 0.302632

0.078947 0.092105 0.618421 0.210526

0.092105 0.052632 0.552632 0.302632

0.105263 0.026316 0.276316 0.592105

0.105263 0.052632 0.394737 0.447368

0.184211 0.078947 0.302632 0.434211

0.013158 0.078947 0.723684 0.184211

0.026316 0.065789 0.723684 0.184211

0.039474 0.039474 0.736842 0.184211

0.052632 0.065789 0.789474 0.092105

0.065789 0.039474 0.802632 0.092105

0.105263 0.026316 0.394737 0.473684

0.144737 0.092105 0.026316 0.736842

0.171053 0.039474 0.026316 0.763158

0 0.118421 0.355263 0.526316

0.092105 0.026316 0.236842 0.644737

0.184211 0.026316 0.302632 0.486842

0 0.105263 0.447368 0.447368

0.026316 0.065789 0.552632 0.355263

0.078947 0.026316 0.815789 0.078947

0.131579 0.052632 0.684211 0.131579

0.157895 0.039474 0.184211 0.618421

0.157895 0.052632 0.684211 0.105263

0 0.171053 0.789474 0.039474

0.013158 0.039474 0.144737 0.802632

0.052632 0.039474 0.473684 0.434211

0.052632 0.065789 0.723684 0.157895

0.065789 0.078947 0.789474 0.065789

0.092105 0.171053 0.736842 0

0.105263 0.065789 0.105263 0.723684

0.118421 0.013158 0.842105 0.026316

0.131579 0.078947 0.434211 0.355263

0.157895 0.026316 0.513158 0.302632

0.026316 0.078947 0.736842 0.157895

0.039474 0.052632 0.210526 0.697368

0.065789 0.013158 0.315789 0.605263

0.092105 0.078947 0.118421 0.710526

0.092105 0.078947 0.210526 0.618421

0.105263 0.039474 0.592105 0.263158

0.184211 0.013158 0.210526 0.592105

0 0.157895 0.223684 0.618421

0 0.184211 0.618421 0.197368

0.013158 0.078947 0.315789 0.592105

0.065789 0.078947 0.802632 0.052632

0.078947 0.026316 0.381579 0.513158

0.105263 0.065789 0.684211 0.144737

0.131579 0.013158 0.618421 0.236842

0.157895 0.092105 0.539474 0.210526

0.157895 0.144737 0.078947 0.618421

0.013158 0.118421 0.868421 0

0.105263 0.013158 0.223684 0.657895

Listed paper and patent all are to quote as a reference in the present patent application.Related description in the above-mentioned embodiment that exemplifies, for example and data only as demonstrating and the usefulness of illustration, and non-limiting scope of the present invention.Any insubstantial modifications processing of being done according to the present invention all falls in the claim scope of the present invention.Therefore, the spirit and scope of annex claims are not limited to the explanation version of the application to this invention.

Claims (20)

1, a kind of method of design of combined sample library may further comprise the steps:

(1) provides all component forming this sample;

(2) for each described component provides a variable, this variable is value in certain interval;

(3) be at least one at least one constraint condition of described specification of variables wherein;

(4) produce pseudo-sample;

(5) check this puppet sample, determine whether it is qualified samples;

(6) repeating step (4) and (5) are until determining at least one described qualified samples.

2, in the described method of claim 1, the interval of component variable is determined by experimental knowledge.

3, in the described method of claim 1, constraint condition is the relation between the above-mentioned variable of being determined by experimental knowledge.

4, the described method of claim 1 further comprises the step of determining the qualified samples ratio, and this qualified samples ratio draws divided by described pseudo-sample size with described qualified samples quantity.

5, in the described method of claim 1, pseudo-sample is that the method by random sampling produces.

6, in the described method of claim 5, random sampling is a kind of Monte Carlo simulation.

7, in the described method of claim 5, random sampling is to finish by the variate-value that utilization has certain probability distribution.

8, in the described method of claim 7, probability distribution is a uniform distribution.

9, in the described method of claim 7, probability distribution is a non-uniform Distribution.

10, in the described method of claim 9, non-uniform Distribution is to be selected from one or more as in next group: Bernoulli distributions, beta distributions, X square distribution, exponential distribution, F-distribution, gamma distribution, Gaussian distribution, normal distribution (for example lognormality, multivariate normal distribution and single argument normal distribution), non-central X square distribution, noncentral f distribution, binomial distribution, negative binomial distribute, polynomial expression distributes, Pareto distributes, Bai Song distribution, student's t-distribution and Sa Lisi distribution.

11, among the described side of claim 7, variate-value is produced at random by randomizer.

12, in the described method of claim 11, random number generator be selected from one group of linear congruence producer, sequencer and quasi random number word generator are deposited in transfer.

13, a kind of method of sample library of the qualified samples that specified quantity is provided may further comprise the steps:

(1) provides the quantity of the sample of required design;

(2) provide all component forming described sample;

(3) for each described component provides a variable, this variable value in certain interval wherein;

(4) be at least one at least one constraint condition of described specification of variables;

(5) produce pseudo-sample;

(6) check this puppet sample, determine whether it is qualified samples;

Repeating step (4) and (5), the qualified samples quantity of in reaching (1), being scheduled to.

14, in the described method of claim 13, the interval of component variable is determined by experimental knowledge.

15, in the described method of claim 13, constraint condition is the relation between the above-mentioned variable of being determined by experimental knowledge.

16, the method for the optimal number of sample in a kind of definite combined sample library may further comprise the steps:

(1) provides all component forming described sample;

(2) for each described component provides a variable, this variable value in certain interval wherein;

(3) be at least one at least one constraint condition of described specification of variables;

(4) produce pseudo-sample;

(5) from pseudo-sample, determine qualified samples;

(6) determine the qualified samples ratio, this qualified samples ratio draws divided by pseudo-sample size with qualified samples quantity;

(7) decision sample size;

(8) optimal number of experiment with computing, wherein, described optimal number is the product of described qualified samples ratio and described sample size.

17, in the described method of claim 16, the variable-value interval is determined by experimental knowledge.

18, in the described method of claim 16, constraint condition is the relation between the above-mentioned variable of being determined by experimental knowledge.

19, a kind of computer-readable medium of store instruction when this medium is performed in one or more treaters, can make described one or more treater carry out as the described method of one of claim 1-18.

20, a kind of system of composite design sample library, it comprises:

(1) a kind of method that all component forming described sample can be provided;

(2) a kind of can be for each described component provides a variable, and this variable one interval in the method for value;

(3) a kind of is the method for at least one described at least one constraint condition of specification of variables wherein;

(4) a kind of method that produces pseudo-sample;

(5) a kind of method of checking this puppet sample is to determine whether it is qualified samples; And

(6) a kind of method that can determine at least one described qualified samples.

Images