CN103679273A - Uncertainty inference method based on attaching cloud theory - Google Patents

Abstract

The invention discloses an uncertainty inference method based on an attaching cloud theory, which mainly solves an inference problem when uncertain information exists in a context. The uncertainty inference method based on the attaching cloud theory comprises the steps of: firstly, describing a given fixed concept by using digital characteristics; understanding the concept as cloud droplets without information of degree of certainty; then establishing a cloud rule controller; regarding the cloud droplets without the degree of certainty as an input of the controller; outputting a cloud droplet vector with the degree of certainty through the controller; finally making all the output cloud droplets accurate to obtain an output result. The uncertainty inference method based on the attaching cloud theory can guarantee inference accuracy and system reliability.

Description

A kind of uncertain inference method based on being subordinate to cloud theory

Technical field

The present invention relates to general fit calculation field, particularly, is a kind of for uncertain Context Reasoning algorithm, mainly solves uncertain contextual reasoning problems in general fit calculation.

Background technology

Context-aware computing (context-awarecomputing) refer to software can be according to the occasion of using, approach personnel and object set, and the time of adjusting and the software of ability.In recent years, researchist has carried out the research of a large amount of relevant context-aware computings, the important general fit calculation in expansion field and the wearable computing of mobile computing develop rapidly in a lot of fields, study widely and apply, particularly large enterprises count complexity technical equipment maintenance and assembling application program reality and product present good situation.

The uncertain reason producing mainly contains: first, knowledge-based inference is more difficult.This is that description due to contextual information model exists isomerism, and for same context, machinery and equipment may have different understanding, and this just causes sharing and reusing very difficult of knowledge.Secondly, also there is limitation in hardware device and Internet Transmission, thereby the contextual information of equipment collection has uncertainty, has noise or has imperfection.Finally, high-rise context is that the low layer context that has uncertain information from these is deduced out, and this process has uncertainty, so high-rise context also has uncertainty.

The firm academician of Objective Concept Li De who is subordinate to cloud proposes.Be defined as: for ordinary set U={u}, claim set for domain.In U, there is fuzzy set

for arbitrary element u, all there is a random number that has steady tendency

be called u couple degree of membership.If in domain U, element is simply orderly, so degree of membership

distribution on U, is called and is subordinate to cloud; If the element in domain U is not simple orderly, but according to certain rule f, U can be mapped to U ' on another orderly domain, in U ', have and only have a u ' corresponding with u, claim that U ' is basic underlying variables, the distribution of degree of membership on U ' is called and is subordinate to cloud.

Summary of the invention

Technical matters: the object of this invention is to provide a kind of uncertain inference method based on being subordinate to cloud theory, mainly solve in general fit calculation context exist uncertain, the problem of reasoning how.The present invention can guarantee the validity and reliability of reasoning.

Technical scheme: the present invention is the uncertain inference method based on being subordinate to cloud theory, is being subordinate on the basis of cloud and Membership Cloud Generators, generates cloud rule controller, realizes reasoning process.

One, Membership Cloud Generators

Membership Cloud Generators (MembershipCloudsGenerator is called for short MCG) is divided into forward and reverse two kinds, and structural drawing is as follows.Wherein, forward MCG is according to known three numerical characteristic E that are subordinate to cloud _x, E _nand D, produce and meet the known two-dimentional water dust Drop (u, μ) that cloud distributes that is subordinate to; Reverse MCG is just in time contrary, is that the known water dust that is subordinate to some in cloud distributes, and determines three numerical characteristic E _x, E _nand D.

Forward MCG under paper, at known MCG～N ³

situation under, generate the water dust that is subordinate to cloud meet this distribution.

(1) generate normal random number x, expectation value is E _x, bandwidth is E _n;

$x~N(E_x,E_n^2)$

(2) calculate normal state and be subordinate to the expectation curve of cloud at the degree of certainty at x place

$\stackrel{\‾}{{\mathrm{\μ}}_x}=\mathrm{\μ}\left(x\right)=e^{\frac{-{(u-E_x)}^2}{2E_n^2}}$

(3) calculation level

the variance D at place _x;

(a) calculate respectively along expectation curve MA, MB by the left and right bandwidth bl, the br that fall half normal state rule variation.

$\mathrm{bl}=\frac{1}{3}\&CenterDot;\underset{E_x}{\overset{E_x+\sqrt{\ln 8}{\mathrm{<figure-callout\; id="1"\; label="E\; n"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">E</figure-callout>}}_n}{\&Integral;}}\sqrt{1+{\left[{\mathrm{\&mu;}}^{\&prime;}\left(x\right)\right]}^2}\mathrm{dx}$

$\mathrm{br}=\frac{1}{3}\&CenterDot;\underset{E_x+\sqrt{\ln 8}{E}_n}{\overset{E_x+3{\mathrm{<figure-callout\; id="1"\; label="E\; n"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">E</figure-callout>}}_n}{\&Integral;}}\sqrt{1+{\left[{\mathrm{\&mu;}}^{\&prime;}\left(x\right)\right]}^2}\mathrm{dx}$

(b) calculation level

the variance D at place _x:

$D_x=D_{\max }\&CenterDot;e^{\frac{-{\left(\mathrm{\&Delta;L}\right)}^2}{2{\mathrm{bl}}^2}}$

$D_x=D_{\max }\&CenterDot;e^{\frac{-{\left(\mathrm{\&Delta;L}\right)}^2}{2{\mathrm{br}}^2}}$

$\mathrm{\&Delta;L}=\underset{E_x+\sqrt{\ln 8}{E}_n}{\overset{x}{\&Integral;}}\sqrt{1+{\left[{\mathrm{\&mu;}}^{\&prime;}\left(x\right)\right]}^2}\mathrm{dx}$

(4) producing expectation value is

variance is D _xnormal random number μ:

$\mathrm{\&mu;}=e^{\frac{-{(x-\stackrel{\&OverBar;}{{\mathrm{\&mu;}}_x})}^2}{2D_x}}$

And then obtain water dust Drop (x, μ).

(5) repeating step (1)～(4), until produce N water dust.

If domain corresponding to concept is n-dimensional space, algorithm can be opened up extensively, obtains n dimension Normal Cloud.

Introduce a kind of reverse MCG algorithm below.

Algorithm flow is as follows:

Input: water dust x _i(i=1,2 ... N).

Output: (E _x, E _n, D)

(1) according to x _i, the average of calculating sampling sample

with variance S:

Mean value formula: $\stackrel{\&OverBar;}{X}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{x}_i$

Formula of variance: $S^2=\frac{1}{N-1}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{(x_i-\stackrel{\&OverBar;}{X})}^2$

(2)

draw expectation value

(3) basis formula below, calculates entropy

(4) basis formula below, calculates variance D:

Two, cloud rule controller

Qualitative rule be divided into wall scroll part single gauge, two condition single gauge with many conditions single gauge etc., this is generally several former pieces and a consequent, to consist of due to rule, it is one or more that former piece can have, consequent but only has one.Equally, in Rule Generator, former piece cloud can be one dimension or multidimensional, and consequent cloud can only be one dimension.One dimension former piece cloud generator and one dimension consequent cloud generator combine, and form controller of wall scroll part single gauge, and algorithm is as follows.

Input: former piece MCG _anumerical characteristic (Ex _a, En _a, D _a), x _awith consequent MCG _bnumerical characteristic (Ex _b, En _b, D _b).

Output: the x corresponding with degree of certainty μ _b.

(1) become normal random number En _a', with En _afor expectation, D _afor mean square deviation;

(2) calculate μ: $\mathrm{\&mu;}=e^{\frac{-{(x_A-{\mathrm{Ex}}_A)}^2}{2{\left({\mathrm{En}}_A^{\&prime;}\right)}^2}}$

(3) production state random number En _b', with En _bfor expectation, D _bfor mean square deviation;

(4) if former piece x _a≤ Ex _a, consequent

(5) if former piece x _a> Ex _a, consequent

From then on can find out, this algorithm has uncertainty.For former piece cloud generator, by x _athe μ obtaining has uncertainty, for consequent cloud generator, and the x being obtained by μ _bthere is equally uncertainty.Therefore, regular controller in uncertain information inference process with probabilistic transmission.

The present invention is based on and be subordinate to four steps that the uncertain inference method of cloud comprises:

1) qualitative rule of cloud rule controller is comprised of several former pieces and a consequent, given formalized description: the former piece of each rule is

consequent is B ^t; Num=1 wherein, 2,3 ..., t=1 ..., M, M is M bar IF-THEN rule in qualitative rule storehouse, every IF-THEN rule R ^tbe described as:

and B ^tbe all sizing concept, use respectively 3-dimensional digital feature

with

describe, wherein represent former piece

under each normal random number;

represent

mathematical expectation;

represent

mean square deviation; represent respectively consequent B ^tunder normal random number, mathematical expectation and mean square deviation;

2) then, for given all definite input vector

i=1 wherein, 2 ... N, is understood as and there is no the i of a degree of certainty information water dust, goes to step 3); If existing degree of certainty information, output is with the water dust of degree of certainty information

to activate the water dust quantitative values obtaining after t rule, μ _tbe its corresponding degree of certainty, go to step 4);

3) i water dust quantitative values structure cloud rule controller: using step 2) obtaining, as the input of controller, by controller, exported water dust

4) to all output water dusts

carry out precision, become a concrete output valve x _b, be final output valve.

Described to all output water dusts

carry out precision, its method is exactly according to the virtual cloud of water dust structure obtaining, and then utilizes reverse MCG algorithm to calculate the numerical characteristic of this virtual cloud, and expectation is exactly the last Output rusults of virtual cloud.

Described to all output water dusts

carry out precision, another kind of method is method of weighted mean, and method of weighted mean is exactly by the water dust obtaining by regular controller

weighted mean, the degree of certainty that weight is each water dust, the later final Output rusults of weighted mean is:

$x_B=\frac{\underset{t=1}{\overset{M}{\mathrm{\&Sigma;}}}{x}_B^t{\mathrm{\&mu;}}_t}{\underset{t=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_t}.$

Beneficial effect: use this scheme to have the following advantages:

1. the method can correctly be carried out uncertain inference, and result is accurate.Example below can illustrate, with the rotating speed that algorithm herein draws, is 40r/min, with fuzzy reasoning, calculates, and the rotating speed obtaining is 39r/min.

2. improve the reliability of reasoning: in reasoning algorithm below, do not adopt accurate subordinate function, but adopt three numerical characteristics that are subordinate to cloud to describe known sizing concept, every corresponding regular controller of qualitative rule, first activate former piece cloud generator, by consequent cloud, obtain Output rusults subsequently, finally accurately turn to Output rusults.In this process, the process that the Output rusults of former piece cloud and consequent cloud draws all has randomness, therefore, for identical input value, adopt this inference method, the result that reasoning each time draws is all likely different, but last the reasoning results is to keep stable tendency on the whole.

3. the performance of optimization system: the present invention utilizes qualitative language to carry out the uncertain information of formalized description, and on this basis context is carried out to reasoning, judgement and decision-making, effectively the performance of optimization system.

Accompanying drawing explanation

Fig. 1 is Membership Cloud Generators, is divided into forward and reverse two kinds.(a) forward MCG; (b) reverse MCG.

Fig. 2 is controller of wall scroll part single gauge.

Fig. 3 is controllers of two condition single gauges.

Fig. 4 is the uncertain inference process flow diagram based on being subordinate to cloud theory.

Embodiment

Suppose to have in qualitative rule storehouse M bar IF-THEN rule, being described as of every IF-THEN rule: R ⁱ:

former piece

with consequent B ⁱbe the qualitativing concept of natural language description, the uncertain inference process based on being subordinate to cloud is as follows.

One, two condition single gauges controller

Input: former piece two-dimensional digital feature (Ex _a1, Ex _a2), (En _a1, En _a2), (D _a1, D _a2), (x _a1, x _a2) and consequent one dimension numerical characteristic (Ex _b, En _b, D _b)

Output: the x corresponding with degree of certainty μ _b.

(1) become normal random number En _a1', with En _a1for expectation, D _a1for mean square deviation;

(2) become normal random number En _a2', with En _a2for expectation, D _a2for mean square deviation;

(3) calculate μ: $\mathrm{\&mu;}=e^{\frac{-{(x_{A1}-{\mathrm{Ex}}_{A1})}^2}{2{\left({\mathrm{<figure-callout\; id="1"\; label="En\; A"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">En</figure-callout>}}_A^1\right)}^2}-\frac{{(x_{A2}-{\mathrm{Ex}}_{A2})}^2}{2{\left({\mathrm{En}}_{A2}^{\&prime;}\right)}^2}}$

(4) production state random number En _b', with En _bfor expectation, D _bfor mean square deviation;

(5) if x _a1≤ Ex _a1, x _a2≤ Ex _a2,

(6) if x _a1> Ex _a1, x _a2> Ex _a2,

(7) if x _a1≤ Ex _a1, x _a2> Ex _a2,

${\mathrm{\&mu;}}_1=e^{\frac{-{(x_{A1}-{\mathrm{Ex}}_{A1})}^2}{2{\left({\mathrm{<figure-callout\; id="1"\; label="En\; A"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">En</figure-callout>}}_A^1\right)}^2},}$ ${\mathrm{<figure-callout\; id="1"\; label="x\; B"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">x</figure-callout>}}_B={\mathrm{Ex}}_B-{\mathrm{En}}_B^{\&prime;}\&times;\sqrt{-2\ln \mathrm{\&mu;}}$

${\mathrm{\&mu;}}_2=e^{\frac{-{(x_{A2}-{\mathrm{Ex}}_{A2})}^2}{2{\left({\mathrm{En}}_{A2}^{\&prime;}\right)}^2},}$ $x_{B2}={\mathrm{Ex}}_B+{\mathrm{En}}_B^{\&prime;}\&times;\sqrt{-2\ln \mathrm{\&mu;}}$

$x_B=\frac{{\mathrm{<figure-callout\; id="1"\; label="X\; B"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">X</figure-callout>}}_B{\mathrm{\&mu;}}_1+X_{B2}{\mathrm{\&mu;}}_2}{{\mathrm{\&mu;}}_1+{\mathrm{\&mu;}}_2};$

(8) if x _a1> Ex _a1, x _a2≤ Ex _a2,

${\mathrm{\&mu;}}_1=e^{\frac{-{(x_{A1}-{\mathrm{Ex}}_{A1})}^2}{2{\left({\mathrm{<figure-callout\; id="1"\; label="En\; A"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">En</figure-callout>}}_A^1\right)}^2},}$ ${\mathrm{<figure-callout\; id="1"\; label="x\; B"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">x</figure-callout>}}_B={\mathrm{Ex}}_B+{\mathrm{En}}_B^{\&prime;}\&times;\sqrt{-2\ln \mathrm{\&mu;}}$

${\mathrm{\&mu;}}_2=e^{\frac{-{(x_{A2}-{\mathrm{Ex}}_{A2})}^2}{2{\left({\mathrm{En}}_{A2}^{\&prime;}\right)}^2},}$ $x_{B2}={\mathrm{Ex}}_B-{\mathrm{En}}_B^{\&prime;}\&times;\sqrt{-2\ln \mathrm{\&mu;}}$

$x_B=\frac{{\mathrm{<figure-callout\; id="1"\; label="X\; B"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">X</figure-callout>}}_B{\mathrm{\&mu;}}_1+X_{B2}{\mathrm{\&mu;}}_2}{{\mathrm{\&mu;}}_1+{\mathrm{\&mu;}}_2}.$

On the basis of this algorithm, can expand to easily controller of many conditions single gauge.

Two, the uncertain inference based on being subordinate to cloud

Reasoning process flow diagram is as Fig. 4, and concrete steps are as follows:

(1) qualitative rule of cloud rule controller is comprised of several former pieces and a consequent, given formalized description: the former piece of each rule is consequent is B ^t; Num=1 wherein, 2,3 ..., t=1 ..., M, M is M bar IF-THEN rule in qualitative rule storehouse, every IF-THEN rule R ^tbe described as:

and B ^tbe all sizing concept, use respectively 3-dimensional digital feature

with

describe, wherein represent former piece

under each normal random number;

represent

mathematical expectation;

represent

mean square deviation;

represent respectively consequent B ^tunder normal random number, mathematical expectation and mean square deviation.

(2) then, for given all definite input vector

i=1 wherein, 2 ... N, is understood as and there is no the i of a degree of certainty information water dust, goes to step 3); If existing degree of certainty information, output is with the water dust of degree of certainty information

to activate the water dust quantitative values obtaining after t rule, μ _tbe its corresponding degree of certainty, go to step 4).

(3) i water dust quantitative values structure cloud rule controller: using step 2) obtaining, as the input of controller, by controller, exported water dust

(4) to all output water dusts

carry out precision, become a concrete output valve x _b, be final output valve.The method of precision has a lot, and a kind of method is exactly according to the virtual cloud of water dust structure obtaining, and then utilizes reverse MCG algorithm to calculate the numerical characteristic of this virtual cloud, and expectation is exactly the last Output rusults of virtual cloud.Another kind method method of weighted mean is simple, intuitive comparatively speaking, accurately practical.Method of weighted mean is exactly by the water dust obtaining by regular controller

weighted mean, the degree of certainty that weight is each water dust, the later final Output rusults of weighted mean is:

$x_B=\frac{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{x}_B^i{\mathrm{\&mu;}}_i}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i}$

For convenience of description, our supposition has following application example:

Suppose the revolution speed control system of probe temperature and pressure in sensor network system.In system, affected by temperature and pressure very large for a certain part running speed, utilizes two dimension input of uncertain inference method design herein, the control system of one dimension output.Relation before temperature, pressure and rotating speed three IF-THEN rule descriptions below:

(1) if temperature is low, pressure is little, and rotating speed is slow

(2) if temperature is moderate, pressure is moderate, and rotating speed is moderate

(3) if temperature is high, pressure is large, and rotating speed is fast

Control system requires temperature and the pressure for accurate input, adopts the uncertain inference based on being subordinate to cloud, obtains the determined value of rotating speed, and process is as follows:

First, utilize three numerical characteristic E that are subordinate to cloud model _x, E _n, D, is converted into quantificational expression by Qualitative Knowledge." pressure is low " and " pressure is high " belongs to one-sided uncertain concept, is subordinate to cloud and left half normal state is respectively subordinate to cloud and represents by right half normal state, and " pressure is moderate " is subordinate to cloud by symmetrical normal state and represents.In rule base, the cloud method for expressing of three qualitativing concepts relevant to pressure P is as follows.

$P_{\mathrm{mall}}=\left\{\begin{array}{cc}1& x\&Element;\left[\mathrm{20,50}\right]\\ C\left(\mathrm{50,10,0.2}\right)& \end{array}\right.$

P _medium＝C(80,10,0.2)

$P_{\mathrm{pwoerful}}=\left\{\begin{array}{cc}1& x\&Element;\left[\mathrm{110,140}\right]\\ C\left(\mathrm{110,10,0.2}\right)& \end{array}\right.$

In like manner, in rule base, the cloud method for expressing of three qualitativing concepts relevant to temperature T is as follows:

$T_{\mathrm{cold}}=\left\{\begin{array}{cc}1& x\&Element;[-\mathrm{20,10}]\\ C(10,\frac{20}{3},0.1)& \end{array}\right.$

$T_{\mathrm{medium}}=C(30,\frac{20}{3},0.1)$

$T_{\mathrm{hot}}=\left\{\begin{array}{cc}1& x\&Element;\left[\mathrm{50,80}\right]\\ C(50,\frac{20}{3},0.1)& \end{array}\right.$

In rule base, the cloud method for expressing of three qualitativing concepts relevant to rotating speed S is as follows:

$S_{\mathrm{slow}}=\left\{\begin{array}{cc}1& x\&Element;\left[\mathrm{0,30}\right]\\ C(30,\frac{20}{3},0.1)& \end{array}\right.$

$S_{\mathrm{medium}}=C(50,\frac{20}{3},0.1)$

$S_{\mathrm{fast}}=\left\{\begin{array}{cc}1& x\&Element;\left[\mathrm{70,100}\right]\\ C(70,\frac{20}{3},0.1)& \end{array}\right.$

Work as T=15, during P=70, ask current motor rotation speed.

According to known conditions, the vector (x of input _a1, x _a2)=(15,70) be understood to that two without the water dust quantitative values of degree of certainty, using this group vector value respectively as three pairs of inputs of controller of condition single gauge by system three rules structures, utilize six pairs of condition single gauges of algorithm controller algorithm calculate respectively corresponding x _a1, x _a2degree of certainty.μ > 0 represents that this rule is activated; μ=0 represents that this rule is not activated.According to the cloud method for expressing of T and P, this group vector value of known input has activated front two rules.

Article one, regular certain specific algorithm being activated:

(1) input:

Ex _A1＝10,

D _A1＝0.1，Ex _A2＝50,En _A2＝10,D _A2＝0.2，

x _A1＝15,x _A2＝70，Ex _B＝30,

D _B＝0.1；

(2) generate normal random number En _a1'=6.65, En _a2'=9.79;

(3) calculate corresponding degree of certainty μ ₁:

${\mathrm{\&mu;}}_1=e^{\frac{-{(x_{A1}-{\mathrm{Ex}}_{A1})}^2}{2{\left({\mathrm{<figure-callout\; id="1"\; label="En\; A"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">En</figure-callout>}}_A^1\right)}^2}-\frac{{(x_{A2}-{\mathrm{Ex}}_{A2})}^2}{2{\left({\mathrm{En}}_{A2}^{\&prime;}\right)}^2}=0.094}$

(4) generate normal random number En _b'=6.5

(5) x _a1> Ex _a1, x _a2> Ex _a2so,

In like manner, certain specific algorithm being activated of second rule is as follows:

(1) input:

Ex _A1＝30, D _A1＝0.1，Ex _A2＝80,En _A2＝10,D _A2＝0.2，

x _A1＝15,x _A2＝70，Ex _B＝50,

D _B＝0.1；

(2) generate state random number En _a1'=6.65, En _a2'=10.12;

(3) calculate corresponding degree of certainty

${\mathrm{\&mu;}}_2=e^{\frac{-{(x_{A1}-{\mathrm{Ex}}_{A1})}^2}{2{\left({\mathrm{<figure-callout\; id="1"\; label="En\; A"\; filenames="HDA0000442841610000011.png"\; state="\{\{state\}\}">En</figure-callout>}}_A^1\right)}^2}-\frac{{(x_{A2}-{\mathrm{Ex}}_{A2})}^2}{2{\left({\mathrm{En}}_{A2}^{\&prime;}\right)}^2}=0.045}$

(4) generate normal random number En _b'=6.8

(5) x _a1≤ Ex _a1, x _a2≤ Ex _a2so,

Finally, adopt the later final output valve of method of weighted mean to be:

$x_B=\frac{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{x}_B^i{\mathrm{\&mu;}}_i}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_i}=\frac{0.094\&times;44+0.045\&times;33}{0.094+0.045}=40$

Be that rotating speed is 40r/min.

Claims (3)

1. the uncertain inference method based on being subordinate to cloud, is characterized in that four steps that the method comprises:

1) qualitative rule of cloud rule controller is comprised of several former pieces and a consequent, given formalized description: the former piece of each rule is

consequent is B ^t; Num=1 wherein, 2,3 ..., t=1 ..., M, M is M bar IF-THEN rule in qualitative rule storehouse, every IF-THEN rule R ^tbe described as:

and B ^tbe all sizing concept, use respectively 3-dimensional digital feature

with

describe, wherein

represent former piece

under each normal random number;

represent

mathematical expectation;

represent

mean square deviation;

represent respectively consequent B ^tunder normal random number, mathematical expectation and mean square deviation;

2) then, for given all definite input vector

i=1 wherein, 2 ... N, is understood as and there is no the i of a degree of certainty information water dust, goes to step 3); If existing degree of certainty information, output is with the water dust of degree of certainty information

to activate the water dust quantitative values obtaining after t rule, μ _tbe its corresponding degree of certainty, go to step 4);

3) i water dust quantitative values structure cloud rule controller: using step 2) obtaining, as the input of controller, by controller, exported water dust

4) to all output water dusts

carry out precision, become a concrete output valve x _b, be final output valve.

2. the uncertain inference method based on being subordinate to cloud according to claim 1, is characterized in that described to all output water dusts

carry out precision, its method is exactly according to the virtual cloud of water dust structure obtaining, and then utilizes reverse MCG algorithm to calculate the numerical characteristic of this virtual cloud, and expectation is exactly the last Output rusults of virtual cloud.

3. the uncertain inference method based on being subordinate to cloud according to claim 1, is characterized in that described to all output water dusts

carry out precision, another kind of method is method of weighted mean, and method of weighted mean is exactly by the water dust obtaining by regular controller

weighted mean, the degree of certainty that weight is each water dust, the later final Output rusults of weighted mean is:

$x_B=\frac{\underset{t=1}{\overset{M}{\mathrm{\&Sigma;}}}{x}_B^t{\mathrm{\&mu;}}_t}{\underset{t=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&mu;}}_t}.$

Images