Ϣ
Information and Control
1999ꡡ285ڡVol.28No.51999



ʽ˹
ķ̱ʶ
  

ժҪҪо˹ڱʶʵʹҵϵ ͳеӦãڱʶĶһ̬ϵͳض˹ڱʶҵ ̵Ķ̬˷ԷΪԲͬ·ص ֱʶͬµķ̣ڿϵͳȶǰ£ ѵ˹ʶģ͵ķ 
ؼʣ̣ϵͳʶBP磬̬ϵͳ
ͼţTP29ױʶ룺B

IDENTIFICATION OF GENERATING PROCESS BASED ON COMBINED
ERROR-FEEDBACK ARTIFICIAL NEURAL NETWORK
KANG Lei XU Nan-rong
(Institute of System Engineering, Southeast University, Nanjing 2 10096)
Abstract: This paper mainly researched the application of artificial neural netw ork in real industrial systems. Because the object is a dynamic system, so we di scussed the dynamic characteristics of artificial neural network used in identif ying industrial process specially. With the background of power generating proce ss, considering the characteristic of different bear, the paper proposed a novel method to identify the model that used sub-network to identify the mathematics model of generating process in different loads. Considering the stability of th e system, this paper introduced a new method to train neural network models with error feedback.
Key wordspower generating process, system identification, BP n eural network, dynamic system

1ǰ
ôͳıʶзԵĹҵ̵ıʶѣNarendra[1]Ķһ㶯̬ϵͳбʶ緽ΪԹҵ̵ıʶṩ˼·˹籾ķԣϵͳʶдͳûеŵ㣮ԻΪ˼ģڹҵ̱ʶеʵӦãȽ˸ģ͵ıʶЧ
糧ÿ춼ϸذ涨ļƻ磬ܹ֤ϵͳУͨ糧ÿյϵͳվ涨һԤߣ糧밴ոɷʵʷУڵǧ򻯣վʱҪ糧ı䷢ֳԱͨ¯һϵ豸趨ֵɣʹ糧繦ʵʵ߾ڷ繦ʵԤߣֵ趨 ֵĲȫһ־Ƚϴֲڣʵ涨Ҫ 
ֻĸɿƷʽ£ڸУҪĵǹ¯úɽƽʱʵʱ繦ʻȽ׵أȽϴʱú뷢繦ʵĹϵԵһһӦģʱȻռ򵥵Աϵúʹ繦ֳ߲ܼ߳ʱָķ繦Ԥߣ糧ʵʷ繦վָķ繦ƫ涨޶ʱ糧ܵһĴ 
Ϊ˸õɷҪüʵʱһڱ䶯£ȷƻ鸺ɱ䶯ʵʱΪֳָԱοʹֳʵʷ繦ʾܺõʡĵҪ󣮱ʶ糧̵ѧģͣǸ鸺ɱ䶯ĻҪо缼ʵʷ ̱ʶ
2ʶģ͵Ļṹ
糧̼ϵͳļ򵥿ͼʾ

ͼ1 糧̼
뷢繦ʹϵΪеǹ¯úΪʶ֮Ĺϵǲ˹磬ͨеʦķѵ˹ʶģ,ʹģܽϺõطӦʵʷ̣һʵʱ̬ϵͳѵʱҪõǰݣһҪùȥݣһҪж̬ԣַͨһṹʱӵԪһԪصĴͻϼʱ亯Ԫ ʱǰNarendraĶԶ̬ϵͳʶĲṹ-ṹ[1]WilliamsZipserʵʱѭreal-time recurrent network[2] Hopfield Wan FIR multilayer perceptronģ[2]
Ĳ˹ṹʱϵͳķʹʶģ;ҪĶ̬ԣø˹ʶģͶԷ̽бʶʱϵͳṹͼ 2ͼ2Уuʾ¯úypʾʵ繦ʣymAĹƷ繦ʣeypym֮ùʽʾƷ繦ym
ym(n)=f(yp(n-1),yp(n-2),,yp(n-p), u(n-1),u(n-2),,u(n-q))(1) 
ʵӦУΪ˹ıʶȼʶЧʣģ黯 ˼룬һʽ磮ڱУ˹Ҫ̬ Ժ⣬Ҫɵȷʵӳ䣮ģ黯˼ʵ ʾǽ϶ı䶯Ƚϴķ繦ʵʱ߷Ϊ򼸸֣Ӷ ͸߱仯ķȣͬڽһӵֽΪ򼸸ϼ򵥵ֱ ӶԴ˹ıʶȼʶЧʣ
Ϊһ߱ʶȣ ͼ2еĴʶģһЩƣ ɵõͼʾϵͳṹ

ͼ2 дṹʶģ

ʱʶĹƷ繦Ϊ
ym(n)=f(yp(n-1),yp(n-2),yp(n-p), 
u(n-1),u(n-2),u(n-q)e(n-1))(2)
ͼ3ʶģ͵Ҫŵ뷴ʱñʶʵģ͵ԤֵʶBIBOȶԾ˱ʶϵͳBIBOȶԣұģдڷ·ñ׼BP㷨ѵͼ3ģдһ·ѲϸNarendraĴģͣe=ym-ypʶϵͳ޷ȷBIBOȶԣǵymypǳӽ£ͼ3еģͼ2еģAʶϵͳһܹ֤BIBOȶԣʱĴڿԽĸţ

ͼ3 ʶģ
3о
31 ģͱʶеģ黯˼
ڻ糧ʵУҪϵؽһһʱ䣬ȻһˮƽڸߡͲͬĸɹ£¯ϵͳǲͬģԣڲͬɹ£ʶӦͬһ㿼ǣǽһʽ磬Ҫɣ磨ߣ磨ͣ 磨ߣ磨ͣǻͼ2Ĵģ͵磮ֳͨ飬 ǽʽ(1)еpȷΪ2qȷΪ3ӳٵλʱȡΪһӣ230MWΪ磬 ʦݼ800飩ֳɸ߸ݶΣ410飩͵͸ݶΣ390飩Щ ߸ݶѵ磨ߣ Щ͸ݶѵ磨ͣμͼ4ڱʶУʵ繦ʸ230MW230MWʱʽֱ 磨ߣ磨ͣıʶ

ͼ4 ʦݵĻ

ñ׼BP㷨磨ߣ磨ͣѵϺ󣬿Խʽ˹ıʶЧֱӲȫʦѵһıʶЧбȽϣΪͳ һṹ5331ʱĽǶʦݷѵ磨ߣ磨ͣȫʦѵһʱһģĸ ʽ5331磬68ȨؼƫֵΪ֤ʽ ıʶЧʣһƵʶһ5641磬 69ȨؼƫֵѵʱȻ5331ɵʽ磮
ͼ5ʾʽıʶЧԵȫʦѵĵ53 31ṹıʶЧҲȫʦѵĵ5641ṹıʶЧɴ˿Կڽ̱ʶʱʽ볣ȣнϸߵıʶȺͱʶЧʣ

ͼ5 ģ͵ıʶЧȽ

32 ʶЧ
ֳ֮ͨǽʽ(2)еpȷΪ2qȷΪ3ӳٵλʱȡΪһ ͼ6һͼ3ʾ6331ʶģͣ

ͼ6 6331ģ

ֱñ׼BP㷨ѵ˹ʶģͣBP㷨 ȶڽϸˮƽ6331ʶģͣ 仯ͼ7ʾ

ͼ7 仯

ݲȥͼ6еe(n-1)Ȩwe(1)we(2)we(3), ͵õһ5331ģͣȻһBIBOȶĴģͣ ѵ5331磬ֱe(n-1)Сʱٽ룮ע⣬Ȩwe(1 )we(2)we(3)ĳʼֵҪõýСԱͻȻϵͳĲ ȻѵǱʶģͱֳϺõȶԣʶҲΪ⣮
ǰһС۵ʽ £ø߸ݶκ͵͸ݶηֱѵʽֻ ǷֱһߵĲ֣ͬûбʵĲǵȨֵӦ뵥 ȨֵѵУȡѵһw e(1)we(2)we(3) ȨֵΪʱwe(1)we(2)w e(3)ʼֵBP㷨ֳϺõԣٶȴڵƾ趨w e(1)we(2)we(3)ʼֵ
ַıʶЧͼ8ʾ

ͼ8 ANNıʶЧ
4
ĸڱʶҵ̵˹Ƽ糧Ļ¯һҵϵͳ಻ͬĿƻ·и߶ȵķԣôͳĽģʶùҵϵͳʮѣòͬṹ˹ԱȽϼ򵥵ؽ÷Զ̬ϵͳıʶ⣮
߼飺ףУ1974ʿоΪ˹ڱʶͿеӦã
٣У1929ʿʦоΪԶƣϵͳ̣
ߵλϴѧϵͳо Ͼ 210096
ο
1Kumpati S. Narendra, Kannan Parthasarathy. Identification and Cont rol of Dynamically Systems using Neural Networks. IEEE Trans. on Neural Networks . 1990, 1(1):127
2Simon Haykin. Neural Networks. Macmillan College Publishing Company. 1994
3Thierry Catfolis. Generating Adaptive Models of Dynamic Systems with Recurren t Neural Network. The 1994 IEEE International Conference on Neural Network. 1994 , 5: 32393243
4Feedback-error Learning Scheme Using Recurrent Neural Networks for Nonlinear Dynamic Systems. 1994 IEEE International Conference on Neural Network. 1994, 1: 175180
5K Warwick, G W Irwin, K J Hunt. Neural Networks for Control and Systems. Peter Peregrines Ltd. 1992
ոڣ1998-07-08
