Ϣ
Information and Control
1999ꡡ285ڡVol.28No.51999



Ŵ㷨֪ʶȡ
еӦо

־ żͻ ĺ

ժҪ ŻǶȳһֻŴ㷨Ļѧϰ÷ܹѧϰ ʵܽõ֪ʶԸŻģͣһ͵Ŵ㷨Ż ÷Ӧתе֪ʶȡ̣ʵ˵÷ǱȽЧģ
ؼʣѧϰ, ֪ʶȡ, Ŵ㷨, min-max׼
ͼţTP13ױʶ룺B

KNOWLEGE ACQUISITION BASED ON GENETIC ALGORITHMS
AND ITS APPLICATION TO FAULT DIAGNOSIS
PEN Zhi-gang ZHANG Ji-hui XU Xin-he
(Control & Simulation Center, Northeasterh University, Shenyang 110006)
AbstractA novel machine learning algorithm is proposed base d on optimization model, which can obtain knowledge from learning instances. It is used to the knowledge acquisition for fault diagnosis. Simulation results sho w its validity.
Key wordsmachine learning, knowledge acquisition, genetic algo rithms, min-max criterion

1
֪ʶȡǹרϵͳġƿ⣬ר֪ʶĺûֱӰϵͳܣΪ һƿ⣬˸ֲͬķѧϰΪһЧֶ ѧϰķܶ࣬ɡȡ͡ʾ[1,10]Уʾѧϰ ѧϰܹӴɢʵ̺ɵйɳһĹǽ ֪ʶȡֶΣʾѧϰķ϶࣬ϢغϢƵľ ѧϰϵͳID3[4]εCID3[5]ȣĴ Ƕ Ŵ㷨һ͵֪ʶȡӦڻеרϵͳ֪ʶ ȡʵ÷ǱȽЧģ
2һ͵Ļѧϰ㷨
ڹרϵͳ֪ʶòʽʾΪ
If A1 & A2 & ... & An Then R(1)
УAi(i=1,2,,n)ʾǰֵRʾۣСԡӦϵͳ ֢״򡱱ʾ֢״µĹ𣮻ѧϰĿͨʵ ʽĹר֪ʶʾѧϰһֻĻѧϰ[4] ӷɢʵ̺ɵйɳһĹΪ˻ȡRб׼򣬿 ʾΪѧϰѵʾΪ䷴ӣѧϰĿǴӹռ ѰһŹʹ̶ȵƣ䷴Ӳƣ ϵͳ֢״ȡֵ(m)ÿһ(1)Ĺǰһ ʾΪ
A=(A1,A2,,Am)
УAi(i=1,2,,m)ֻȡֵAi=0ʾ֢״ûг֣Ai=1ʾ֢״ R1(PEi(i=1,2,,lʾ)s(NEi(i=1,2,,s )ʾ)ӦǰmάʾΪ
PEi=(Ai1,Ai2,,Aim) i=1,2,,l

ڹռе
If A1 & A2 &  & Am Then R䡡(2)
ùѧϰʵƳ̶ȿӦļнǵֵʾΪ
(3)
(4)
ѧϰĿԹһŻ⣬ڹռѰ(2)Ĺ,ʹ
r(R,PEi)(5)

r(R,PEi)(6)
ֱﵽСֵ
Ż⣬ɿķǲ㷨ŹǰĿ ࣬ռ佫÷ǳ޴ǲеģŴ㷨ǽһ ЧŴ㷨MichiganѧHollandڴ[2] K.DeJongJ.GrefensetteD.GoldbergL.Davis˵оŻӦơѧϰ˹õ˹㷺Ӧ[3]ӦŴ㷨Ļڰ:룻ʼȾɫ壻ӦֵȷѡŴӣȷŴ㷨ṹѡŴ㷨ΪŻ(5,6)öƱ뷽ʽȾɫ峤ȵϵͳ֢״ĳһλȡ0ʾ֢״ûг֣ĳһλȡ1ʾ֢״Ѿ֣ŻĿ꣬Żָ꺯
(7)
Ҫָǣ[8,9]ҲƵķûпǵʽ(3,4)ĸΪ0ңĵĿۺ(7)Ҳ[8]зͬĿۺ(7)˸еmin-max׼׼ȷطӳܵĺû
3ѧϰʵ
Ϊ˵Чԣ[8,9]вõתе֪ʶȡΪʵ飮תеĳ[8]Ĥ񵴡ʼƽ⡢ҶƬṸתײתתƵȣʵУԻһʵӦ֢״ǲ[8]иʵ(1ʾ)ÿһӦһʵ еԪֵӦĳһ֢״ڴ˹ǵĹ֢״2ʾ 


1 ʵ

ϴŹʵ
1
Ĥ(1,4,5)(15,17)(4,5)(17,28)(5,17,24)(4,5,15)(4,17,24)(5,17,24)
(4,5,15)(4,17,24)(5,17,29)(4,5,17)
2
ʼƽ(2,6,11,18)(6,11,18)(11,14,18,25)(2,11,14,18)(6,11,14,18)(11,14,18,25)
(3,11,18)(2,11,18,25)(2,11,18)(11,18,25)
3
ҶƬṸ(6,11,34,35)(6,15,19,34,37)(6,11,35)(6,11,15,36)(15,19,35,36)
(6,11,34,37)(6,11,15,35,36)
4
תײ(3,10,16,31,33)(10,16,20,22)(10,16,20)(3,10,31,33)(10,20,22)(10,16,20)
(10,16,20,22)(10,15,20)(10,20,22)
5

ת(12,15,23,32)(12,19,23,32)(6,12,23,32)(3,15,19,25)(6,15,23)(12,19,23)
(19,25,32)(3,12,23)(3,6,15,19)(6,12,19,25)(12,18,23)(15,19,32)
(3,15,19,25)(13,23,32)
6
ת(8,13,23,27)(3,13,23,26)(13,26,27)(3,8,23,27)(8,15,23)(8,13,26,27)
(13,23,26)(15,23,27)(8,13,27)(15,26,27)
7
ת᲻(13,23,25)(7,13,23)(7,13,23,25)(7,13,23)(3,7,15,23)(8,23,25)(3,13,23,25)
(3,7,13,23)(7,13,23)(13,23,25)
8
ɶ(16,20,38,39)(10,20,38,40)(10,16,20)(10,16,30,39)(10,16,20,39)
(10,16,38,40)(16,38,39)(20,39,40)(10,16,20,40)


2 ѡ

֢״֢״
1񶯲ηֽز21Ƶ仯
2Ĺ켣Բ22ƵҺС
3Ĺ켣㽺λ23Ƶ
4Ĺ켣824ʱƵɷֻ
5Ƶ0.3-0.5ƵƵͻ25ʱƵ߱仯
6ƵԵĸƵ26ٽתʱ1x2xȶ
7ƵԵĶƵ271/21/3ٽתʱλԱ仯
8ƵԵƵ28нײ
9ƵгָߴƵ29ת¶в
10ƵгɵƵƵĹƵɷ30תλ
11Ƶл2x3x31תӲ
12ҪĹƵ⻹2x3x32תƫ
13Ҫ2x⻹һ1x3x33¸׵²
14仯34ˮӲ>5g/l
1535ˮ>50g/l
16ҺС仯36ˮPH<7
17Ƶ37ˮ>1/cm
18Ƶ仯38񶯼񶯲
19Ƶ39ֱ񶯺ˮƽ񶯷ֵ
20ƵҺС仯40ֱ񶯺ˮƽ񶯸Ƶ


ӦñĻѧϰ㷨Եõ3ʾ֪ʶѧϰУȡȺ ģΪ50pc(0.6,0.9)pm(0.05,0.2)ҲŸ屣 ԣֲֵ㣬üŴ㷨ҵȫŵ㣬Ĳһ ͵Ŵ㷨Ѱȫŵ㣮Ŵ㷨ˡ桢̭ ȻѡԭӦȵĸһĸʾ͸ߣںѰУܽϲ ĸ壬ҲܳΪܺܺõĸ壬ˣTing kuo[6]һ µŴ㷨ѡƣ֮Ϊɢѡ(disruptive selection)ƣڸѡ ʽУӦȶ巽ʽΪ
(8)
Уf(.)ʾĿ꺯ʾƽֵѡʽھж༫ֵҺֵ仯ϴĿ꺯Ѱ⣬нϸߵЧʣĲѡʽŴ㷨öȺ޸
(9)
Уk(1,2]ΪƲ޸ĺ󣬾ͲĳһöΪ0ڸӣMatlab̽˴ʵ飮ڲͬĽʺͱʣʵ鶼õ˽ϺõĽΪ˵һ͵ʵУĿ꺯ĵͽͼ1-2ʾ(Ϊӣ1͹2ѧϰ̣ʡ) 

ͼ11ѧϰ̡ͼ22ѧϰ

(3)е֪ʶд
if A1 & A2 &  & An Then R
ʽﲻдЩʵԿĻѧϰ㷨ǱȽЧ õ֪ʶܹȽ׼ȷطӳʵ
3 ѧϰ

֢״
1(0001100000000000100000010000000000000000)Ĥ
2(0000010000100100010000001000000000000000)ʼƽ
3(0000010000100010000000000000000001100000)ҶƬṸ
4(0000000001000001000101000000000000000000)תײ
5(0000000000010000001000001000000010000000)ת
6(0000000100001010000000101000000000000000)ת
7(0000000000001000000000101000000000000000)ת᲻
8(0000000001000001000100000000000000000000)ɶ


4
Ŵ㷨һ͵Ļѧϰ㷨תе֪ʶ ȡķʵ㷨֮Чģ㷨һصǼ УСʵӦã

߼飺־գУ1972ʿоΪܵȣϵͳ森
żͻ, , 28, ʿо. оΪɢ¼̬ϵͳȡܼ.
ĺ, , 57, , ʿʦ. оΪɢ¼̬ϵͳ桢ƻȵ.
ߵλѧƷ  110006
ο
1Sang chan park, Melinda T. Expanation -based Learning for Intellig ent Process Planning, IEEE Trans. SMC, 1993, 23(6), 15971615
2 Holland J H. Adaptation in Natural and Artificial Systems, 1st ed., 1975, 2n d ed. Cambridge, MA: MIT Press, 1992
3 Goldberg D E. Genetic Algorithms in Search, Optimization, Machine Learning, Reading MA: Addison Wesley, 1989
4 Quinlan. Decision trees and decision making, IEEE Tran. SMC, 1990, 20[STHZ](2):339343
5 Liu N, Cios K.J. A Machine Learning Method for Generation of Neural Network Architecture: a Continuous Algorithm, IEEE Trans on neural networks,1992, 3(2):281290
6 Ting Huo and Shu-yuen Hwang, A Genetic Algorithm with Disruptive Selection, IEEE Trans. on SMC, 1996, 26(20):299307
7  ܹ.ѧϰܳϵͳеӦãߣ1997, 12(3):222228
8 ѩӱֵ.ģ˻㷨֪ʶȡоߣ 1997, 12(4):327331
9 ѩӱ,. ˻ݻ㷨֪ʶȡƵо Ӧã199815(1):9399
10 ʯһ. ˹ԭ廪ѧ磬1993:170267
ոڣ1998-04-08
