AUTOMATION & CONTROL - Theory and Practice Part 8 pptx

AUTOMATION&CONTROL-TheoryandPractice166 Fi g T h la y ac c T h in t Pr o - s e it eq u Fi n fu n 3. 2 T h th e p o co n P D g . 5. Generic stru c h e procedure to c y er receives a va l c ordin g to equati h is activation co e t ermediate layer, o pa g ation ma y b e lection of an ap p is also necessar y u ation (5), and t h n ally, the output n ction that was u 2 Neural PDF h e principal reas o e anal y sis, desi g o ssible to acquire n trolled, this inf o D or PID and, in t h c ture for three la y c alculate the pro l ue X i , which is p on (3) where S is p j S e fficient S j p is p r in this case the s i h b e accomplished u p ropriate functio n y to define an e h is is obtained w h p k r t function of the u sed to connect t h o n that Artificial g n and impleme n any a priori kn o o rmation can be u h e case consider e y er ANN. pa g ation values p ropa g ated with called the activa t n n i p iij p WXW 1     r opa g ated b y an ig moid function, p j S p j e S h    1 1 )( u sin g alternative n depends on th e e xcitin g functio n h en the wei g hts V     l j p j p jk hV 1 output neuron ‘ h e input and hid d p k r p k e O    1 1 Neural Networ k n tation of contr o o wledge of the st u sed to improve t e d here, PDF. Th e for each la y er i s a weight W ij to t ion coefficient. p j,1  output functio n equation (4). p ran g e-limited f u e final applicatio n n , in order to a c V jk have been cal c   p kl V ,1 ‘ k’ is obtained u s d en la y ers, equati k s (ANNs) have e o l strate g ies is t t ructure of the m t he tunin g of t y p e re are man y con t s as follows. Th e the intermediat e n which represe n u nctio n s, such as n . As shown in fi g c cess the output c ulated. sing the same si on (6). e arned their posi t t heir flexibilit y . I m odel of a syste m ical controllers, s t ributions in the a e input e la y er, (3) n ts the (4) tanh -1 , g ure 5, la y er, (5) gmoid (6) t ion in If it is m to be s uch as a rea of ar t ad j G a T h G a h y fu z pr e tu r h y us e In di s (2 0 N e E q in t fu n th e Fi g t ificial neural ne j ust the parame t a rcez & Garcez, 1 9 h ere have been s e a rcez and Garce z y droelectric pow e z z y inference to e sented a sel f -le a r bine g overnor. R y bridized control l e d as g overnors o this work a ba s crete PDF re g ul a 0 00) with g reat s e ural-PDF schem e q uatio n s 7 and 8 t erconnection V j n ction for the er r e chan g e of si g n i g . 6. Neural PDF. tworks aimed a t t ers of discrete P 9 95). In this wor k e veral works w h z (1995) applie d e r plant. D j ukano v control a low h a r n in g control s ys R ecentl y , Shu-Qi l er based on g e n o f a h y droelectri c ck-propa g ation s a tor. This strate g s ucess in practic a e proposed. The r j tv 1 (  ji tw 1 (   8 are expressed and W ij . Equati o r or.  is include d i n the evolution o t definin g fast a n P ID control s y ste m k a similar strate g h ere ANN have d PI neural cont r v ic, et al. (1997) v h ead h y dropow e s tem usin g a PID n g et al. (2005) h n etic al g orithms a c power plant m o s trate gy has be e gy was used to a d a l implementati o r e g ulation can b e j signtv ()() 1   ji s ig n tw )() 1   j j e e h v tE      1 )(  to recursivel y o n 9 is used to d e d to calculate the o f the process. n d effective stra t m s (Narendra & gy is used to tun e been applied to r ol to a linear s v alidated an ada p e r s y stem. Yin-S o fuzz y NN and a h ave compared a a nd fuzz y NN w o del. e n used to ad j u s dj ust a PID cont r o ns. Fi g ure 6 sh o e calculated b y : j u y h e e 1 )    i j u y x e e n 2 )(    u y e e ad j ust the wei gh e velop the mini m g radient of the f t e g ies to calcula t Mukhopadh y a y e a discrete PDF. h y droelectric s y s imulator of a 2 0 p tive-network ba o n g , et al. (2000 ) a pplied it to a h yd a PID controller w w hen the controll e s t the paramete r r oller b y A g uado o ws the scheme hts for each ne u m ization of the t r f unction and to e t e and , 1996; y stems. 0 MW sed o n ) have d raulic w ith a e rs are r s of a Behar of the (7) (8) (9) u ronal r ansfer e xpress NeuralPDFControlStrategyforaHydroelectricStationSimulator 167 Fi g T h la y ac c T h in t Pr o - s e it eq u Fi n fu n 3. 2 T h th e p o co n P D g . 5. Generic stru c h e procedure to c y er receives a va l c ordin g to equati h is activation co e t ermediate la y er, o pa g ation ma y b e lection of an ap p is also necessar y u ation (5), and t h n all y , the outpu t n ction that was u 2 Neural PDF h e principal reas o e anal y sis, desi g o ssible to acquire n trolled, this inf o D or PID and, in t h c ture for three la y c alculate the pro l ue X i , which is p on (3) where S is p j S e fficient S j p is p r in this case the s i h b e accomplished u p ropriate functio n y to define an e h is is obtained w h p k r t function of the u sed to connect t h o n that Artificial g n and impleme n an y a priori kn o o rmation can be u h e case consider e y er ANN. pa g ation values p ropa g ated with called the activa t n n i p iij p WXW 1     r opa g ated b y an ig moid function, p j S p j e S h    1 1 )( u sin g alternative n depends on th e e xcitin g functio n h en the wei g hts V     l j p j p jk hV 1 output neuron ‘ h e input and hid d p k r p k e O    1 1 Neural Networ k n tation of contr o o wled g e of the s t u sed to improve t e d here, PDF. Th e for each la y er i s a weight W ij to t ion coefficient. p j,1  output functio n equation (4). p ran g e-limited f u e final applicatio n n , in order to a c V jk have been cal c   p kl V ,1 ‘ k’ is obtained u s d en la y ers, equati k s (ANNs) have e o l strate g ies is t t ructure of the m t he tunin g of t y p e re are man y con t s as follows. Th e the intermediat e n which represe n u nctio n s, such as n . As shown in fi g c cess the output c ulated. s in g the same si on (6). e arned their posi t t heir flexibilit y . I m odel of a s y ste m ical controllers, s t ributions in the a e input e la y er, (3) n ts the (4) tanh -1 , g ure 5, la y er, (5) g moid (6) t ion in If it is m to be s uch as a rea of ar t ad j G a T h G a h y fu z pr e tu r h y us e In di s (2 0 N e E q in t fu n th e Fi g t ificial neural ne j ust the parame t a rcez & Garcez, 1 9 h ere have been s e a rcez and Garce z y droelectric pow e z z y inference to e sented a sel f -le a r bine g overnor. R y bridized control l e d as g overnors o this work a ba s crete PDF re g ul a 0 00) with great s e ural-PDF schem e q uatio n s 7 and 8 t erconnection V j n ction for the er r e chan g e of si g n i g . 6. Neural PDF. tworks aimed a t t ers of discrete P 9 95). In this wor k e veral works w h z (1995) applie d e r plant. D j ukano v control a low h a r n in g control s ys R ecentl y , Shu-Qi l er based on g e n o f a h y droelectri c ck-propa g ation s a tor. This strate g s ucess in practic a e proposed. The r j tv 1 (  ji tw 1 (   8 are expressed and W ij . Equati o r or.  is include d i n the evolution o t definin g fast a n P ID control s y ste m k a similar strate g h ere ANN have d PI neural cont r v ic, et al. (1997) v h ead h y dropow e s tem usin g a PID n g et al. (2005) h n etic al g orithms a c power plant m o s trate gy has be e gy was used to a d a l implementati o r e g ulation can b e j signtv ()() 1   ji s ig n tw )() 1   j j e e h v tE      1 )(  to recursivel y o n 9 is used to d e d to calculate the o f the process. n d effective stra t m s (Narendra & gy is used to tun e been applied to r ol to a linear s v alidated an ada p e r s y stem. Yin-S o fuzz y NN and a h ave compared a a nd fuzz y NN w o del. e n used to ad j u s dj ust a PID cont r o ns. Figure 6 sh o e calculated b y : j u y h e e 1 )    i j u y x e e n 2 )(    u y e e ad j ust the wei gh e velop the mini m g radient of the f t e g ies to calcula t Mukhopadh y a y e a discrete PDF. h y droelectric s y s imulator of a 2 0 p tive-network ba o n g , et al. (2000 ) a pplied it to a h yd a PID controller w w hen the controll e s t the paramete r r oller b y A g uado o ws the scheme hts for each ne u m ization of the t r f unction and to e t e and , 1996; y stems. 0 MW sed o n ) have d raulic w ith a e rs are r s of a Behar of the (7) (8) (9) u ronal r ansfer e xpress AUTOMATION&CONTROL-TheoryandPractice168 4. 4. 1 Di n ar e th e sp e g o re f al s gr i Fi g T h re g in c si g th e re g va n re g fe e re a si g th e li n 4. 2 W i ar e th e Classic contr o 1 Dinorwig Gov e n orwi g has a di g e two control lo o e turbine’s g uid e e ed re g ulation d vernor. A PI co n f erence to the p o s o a derivative fe e i d frequency. g . 7. Scheme of t h h e g enerators m u g ulator y authori t c reasin g g enerati g nal to the g ove r e g overnor oper a g ulatio n (Manso o n e openin g and g ulation mode ( p e d-forward sign a a ction when bi g c g nal (control si gn e feed-forward s n ear relationship b 2 Anti-windup P I i th careful tunin g e sub j ect to const e se circumstanc e o llers for hydr o e rnor Configura t g ital Governor w o ps, for power a n e vane is ad j uste d d roop. The Dro o n fi g uration is use d o wer control loo p e d-forward loop h e Dinorwi g Go v u st maintain th ty . When the r e o n . On the othe r r nor valve will c l a tes with two d o r, 2000). The po w defines the oper a p art load respon s a l, directly sets t c han g es in the p o n al) is produced b s i g nal. The pow e b etween g uide v a I D g , PI control can raints and their b e s, the performa n o electric stati o t ion w hose g eneral co n n d frequenc y (M a d dependin g on o p g ain is used d for this control p , which is prop o that allows the s v ernor. e speed within e ference is raise d r hand, when th e l ose, decreasin g g roop settin g s; 1 % w er reference si g a tin g point for t h s e). Chan g in g th e t he guide vane o wer reference a p by addin g the ou t e r feedback loo p a ne openin g and offer g ood and r b ehaviour chan ge n ce of a linear o ns n fi g uration is sh o a nsoor, 2000). In the power devi a to chan g e the s . The frequenc y c o rtional to the fr e sy stem to respon d an operational d the g overnor e output si g nal i s g eneratio n (Wri g % for hi g h re g u l g nal sets the refer h e unit when it i s e power referen c position, in ord e p pear. The g uide v t put si g nals fro m p compensates t h power. r obust performa n e s when the cons controller, such o wn in Fi g ure 7. the power contr o a tion multiplied s peed reference c ontrol loop pro v e quenc y error. T h d to a rapidly-ch a band defined b valve will ope n s lowered the re f g ht, 1989). At Di n l ation and 4% f o ence position fo r s workin g in fre q c e, which also ac er to produce a v ane position re f m the P, I and D p h e s y stem for th e n ce. However, al l traints are activ a as PI, can dete r There o l loop b y the of the v ides a h ere is a n g in g by the n , thus f erence n orwi g o r low r g uide q uenc y ts as a rapid f erence p arts to e no n - l Plant a ted. In r iorate si g be c an ca u 20 0 be c th i o u s ys si g A t Fi g A t a n sa t be g a i li m is c tr a Fi g 5. T h g o w e g nificantl y (Pen g, c omes excessive l d it then “winds u sed b y the satu r 0 1). In other wo r c ause increasing i s behaviour per u tput of the pla n s tem back to its c g n for a lon g ti m t herton, 1995). g ure 8 shows a ge t herton, 1995). T h n e g ative value a n t uration is used t inte g rated is m o i n (K i ) are ad j us t m it and the dead z c ommonl y used. a ckin g anti-wind u g . 8. General sch e Simulink © Mo d h e Simulink © soft w vernors. This to o e re constructed , et. al., 1996). W h ly lar g e compar e up”. In addition , r ation effect (Pe n r ds, windup is p the control si g n sists the inte g ra t n t. As a conseq u c orrect stead y -st a m e. The result i s e neral PI control l h is controller has n d forces the out t o reduce the int e o dified b y the p r t ed in order to m z one depend on t In this work, th e u p structure will e me of PI anti-wi n d el and Progr a w are tool was us e o l has libraries o f usin g these sta n h en the plant has e d to a linear res p , a hi g her inte g ra t ng , et. al., 1996; B roduced when t h al can no longer t or value can be u ence, when re c a te value require s a lar g e overs h l er that includes a an internal feed b put of the s y ste m eg rator input. As r oportional g ain m aintain equival e t he constraints fi x e responses of th e be used as a basi n dup. a m e d to facilitate st u f specific functio n n dard Simulink © actuator saturat i p onse (an actua t t or output and a B ohn & Atherton , h e control si g nal accelerate the r e come ver y lar g e c overin g from s a s the control err o h oot and a long a tracking anti- w b ack path, which m to be in the li n can be seen fro m (K), therefore th e e nce with the cl a x ed b y the opera t e plant when it i s s of comparison. u dies of the pow e n s (blocks) and t h © functions. Us i i on the inte g rato r t or without satu r lon g er settlin g ti m , 1995; Goodwin , saturates the ac t e sponse of the p l , without affecti n a turation, bringi n o r to be of the o p settling time (B o w indup scheme ( B drives the inte g r n ear ran g e. The i n m Fi g ure 8 the si g e values of the i n a ssic PI. The sat u t or; a value of 0. 9 s g overned b y a P e r plant under di f h e power plant m i n g a dialo g b o r value r ation), m e are , et al., t uator, l ant. If ng the ng the p posite o hn & B ohn & ator to n ternal g nal to n te g ral u ration 9 5 p. u. P I with f ferent m odels o x, the NeuralPDFControlStrategyforaHydroelectricStationSimulator 169 4. 4. 1 Di n ar e th e sp e g o re f al s g r i Fi g T h re g in c si g th e re g va n re g fe e re a si g th e li n 4. 2 W i ar e th e Classic contr o 1 Dinorwig Gov e n orwi g has a di g e two control lo o e turbine’s g uid e e ed re g ulation d vernor. A PI co n f erence to the p o s o a derivative fe e i d frequenc y . g . 7. Scheme of t h h e g enerators m u g ulator y authori t c reasin g g enerati g nal to the g ove r e g overnor oper a g ulatio n (Manso o n e openin g and g ulation mode ( p e d-forward si g n a a ction when bi g c g nal (control si gn e feed-forward s n ear relationship b 2 Anti-windup P I i th careful tunin g e sub j ect to const e se circumstanc e o llers for hydr o e rnor Configura t g ital Governor w o ps, for power a n e vane is ad j uste d d roop. The Dro o n fi g uration is use d o wer control loo p e d-forward loop h e Dinorwi g Go v u st maintain th ty . When the r e o n . On the othe r r nor valve will c l a tes with two d o r, 2000). The po w defines the oper a p art load respon s a l, directl y sets t c han g es in the p o n al) is produced b s i g nal. The pow e b etween g uide v a I D g , PI control can raints and their b e s, the performa n o electric stati o t ion w hose g eneral co n n d frequenc y (M a d dependin g on o p g ain is used d for this control p , which is prop o that allows the s v ernor. e speed within e ference is raise d r hand, when th e l ose, decreasin g g roop settin g s; 1 % w er reference si g a tin g point for t h s e). Chan g in g th e t he g uide vane o wer reference a p by addin g the ou t e r feedback loo p a ne openin g and offer g ood and r b ehaviour chan ge n ce of a linear o ns n fi g uration is sh o a nsoor, 2000). In the power devi a to chan g e the s . The frequenc y c o rtional to the fr e sy stem to respon d an operational d the g overnor e output si g nal i s g eneratio n (Wri g % for hi g h re g u l g nal sets the refer h e unit when it i s e power referen c position, in ord e p pear. The g uide v t put si g nals fro m p compensates t h power. r obust performa n e s when the cons controller, such o wn in Fi g ure 7. the power contr o a tion multiplied s peed reference c ontrol loop pro v e quenc y error. T h d to a rapidly-ch a band defined b valve will ope n s lowered the re f g ht, 1989). At Di n l ation and 4% f o ence position fo r s workin g in fre q c e, which also ac e r to produce a v ane position re f m the P, I and D p h e s y stem for th e n ce. However, al l traints are activ a as PI, can dete r There o l loop b y the of the v ides a h ere is a n g in g by the n , thus f erence n orwi g o r low r g uide q uenc y ts as a rapid f erence p arts to e no n - l Plant a ted. In r iorate si g be c an ca u 20 0 be c th i o u s ys si g A t Fi g A t a n sa t be g a i li m is c tr a Fi g 5. T h g o w e g nificantl y (Pen g, c omes excessive l d it then “winds u sed b y the satu r 0 1). In other wo r c ause increasing i s behaviour per u tput of the pla n s tem back to its c g n for a lon g ti m t herton, 1995). g ure 8 shows a ge t herton, 1995). T h n egative value a n t uration is used t inte g rated is m o i n (K i ) are ad j us t m it and the dead z c ommonl y used. a ckin g anti-wind u g . 8. General sch e Simulink © Mo d h e Simulink © soft w vernors. This to o e re constructed , et. al., 1996). W h ly lar g e compar e up”. In addition , r ation effect (Pe n r ds, windup is p the control si g n sists the inte g ra t n t. As a conseq u c orrect stead y -st a m e. The result i s e neral PI control l h is controller has n d forces the out t o reduce the int e o dified b y the p r t ed in order to m z one depend on t In this work, th e u p structure will e me of PI anti-wi n d el and Progr a w are tool was us e o l has libraries o f usin g these sta n h en the plant has e d to a linear res p , a hi g her inte g ra t ng , et. al., 1996; B roduced when t h al can no longer t or value can be u ence, when re c a te value require s a lar g e overs h l er that includes a an internal feed b put of the syste m eg rator input. As r oportional g ain m aintain equival e t he constraints fi x e responses of th e be used as a basi n dup. a m e d to facilitate st u f specific functio n n dard Simulink © actuator saturat i p onse (an actua t t or output and a B ohn & Atherton , h e control si g nal accelerate the r e come ver y lar g e c overin g from s a s the control err o h oot and a long a tracking anti- w b ack path, which m to be in the li n can be seen fro m (K), therefore th e e nce with the cl a x ed b y the opera t e plant when it i s s of comparison. u dies of the pow e n s (blocks) and t h © functions. Us i i on the inte g rato r t or without satu r lon g er settlin g ti m , 1995; Goodwin , saturates the ac t e sponse of the p l , without affecti n a turation, bringi n o r to be of the o p settling time (B o w indup scheme ( B drives the inte g r n ear range. The i n m Fi g ure 8 the si g e values of the i n a ssic PI. The sat u t or; a value of 0. 9 s g overned b y a P e r plant under di f h e power plant m i n g a dialo g b o r value r ation), m e are , et al., t uator, l ant. If ng the ng the p posite o hn & B ohn & ator to n ternal g nal to n te g ral u ration 9 5 p. u. P I with f ferent m odels o x, the AUTOMATION&CONTROL-TheoryandPractice170 pa m o s ys Si m co m tu r fo r in s n o w i m o bl o C u fu n (le th e m o vi e pa o u al g ch a ha ch a Fi g rameters of a s p o dels ma y be ch a s tem and linear o m ulink © power m binin g the f r bine/ g enerator a r this stud y ; the y s tance, there are o nlinear no n -elas t i thout rate limit a o del can be ad j u s o ck has the optio n u stomised Simul i n ctions can be i n arnin g paramete r e plant and its o u o del to be chan ge e wed and assess e rameters. The c u u tput deviatio n fr g orithm takes ar o a n g e) to find th e ve been reache d a n g e. g . 9. Simulink © p o p ecific block can a n g ed. These m o o r nonlinear beh a plant model. T h f our sub-s y ste m a nd sensor filter s y can be select e three models a t ic and nonlinear a tion and satura t s ted to represen t n of classical and i nk © S-functions n corporated wit h r s) and sample ti u tput is the con t e d easil y or the c e d. The neural al u rrent criterion o om the set-point ; o und 10 iteratio n e “best” ran g e o d the parameter s o wer plant mode l be ad j usted. For o dels can represe n a viour ma y be sel h e full h y droel e m s: Guide va n s . Each block is p e d to represent a a vailable to sim u elastic. The g uid e t ion. The sensor t different condi t advanced contr o were develope d h in Simulink © m o me. Its inputs ar e t rol si g nal. The v c ontrol al g orith m g orithm calculat e o f optimalit y is q ; however this cr i n s (the exact val u o f parameter val u s sta y constant u l . example, the o p n t the power pl a e cted. Fi g ure 9 s h e ctric station m o n e d y namics, p art of the Simu l a diversit y of m o u late the h y drau e vane d y namics filters block is a ions of the natio o llers. d for the neural P o dels. The neur a e the reference a n v ersatilit y of Sim u m to be modifie d e s the optimal v a q uadratic error, i terion can be ch a u e depending o n u es (trainin g ti m u ntil the set-poi n p eratin g point of a nt as a SISO or M h ows a schemati c o del is construc t h y draulic subs y l ink © librar y dev e o des of operati o lic subs y stem - can be selected w a fixed block. T h nal g rid. The g o v P DF al g orithms. a l PDF block acc n d the output si gn u link © allows th e d and new result s a lues of the cont r where the error a n g ed if necessa r n the ma g nitude m e). When these r n t or the plant linear M IMO c of the t ed by y stem, e loped o n. For linear, w ith or h e g rid v ernor These c epts η n als of e plant s to be r ol law is the ry . The of the r an g es model 6. A s pr o s ys te s Fo an co n pa ba pl a sh o co n co n o p dr i Fi g ca s Simulation re s s discussed previ o vide timel y an d s tem. The actual s tin g , it can be sp e r all simulations , d 50 Hz, and ass u n nected to the n rameters fixed a sis of compariso n a nt under anti- w o ws the small s n nected. In bot h n troller, being r e p erational cases. T i vin g the process g . 10. Small-step s e o f one unit in o s ults ousl y , the role o f d accurate sup p form of the pow e e cified in terms o , the model is e x u mes a Grid s y s t n onlinear model t K=0.1 and T i = 0 n . Figure 10 sho w w indup PI and n e s tep responses ( 0 h cases, the h yd e spectivel y 10% a T he undershoot . response of the o peratio n . f a h y droelectric p l y of its dema n e r demand is rel a o f step, ramp and x pressed in the p t em with infinite of the h y droele c 0 .12 (as currentl y w s the small step r e ural PDF contro l 0 .05 p.u.) of the d roelectric plant a nd 30% faster i n is also reduced i hydro plant wit h station in frequ e n ded power con t a ted to Grid freq u random input si p er-unit s y stem, n busbars. The ne u c tric power plan t y implemented i n r esponses (0.05 p l lers for one uni t power station w performs better n the one unit o p i n both cases w h h neural PDF an d e nc y control mo d t ribution to the u enc y variation b g nals. n ormalized to 3 0 u ral PDF controll t . A PI controll e n practice) is us e .u.) of the h y dro e t operational. Fi g w hen all six un with the neur a p erational and si x h en a PDF contr o d PI controllers f d e is to power b ut, for 0 0 MW er was e r with e d as a e lectric g ure 11 its are a l PDF x units o ller is f or the NeuralPDFControlStrategyforaHydroelectricStationSimulator 171 pa m o s ys Si m co m tu r fo r in s n o w i m o bl o C u fu n (le th e m o vi e pa o u al g ch a ha ch a Fi g rameters of a s p o dels ma y be ch a s tem and linear o m ulink © power m binin g the f r bine/ g enerator a r this stud y ; the y s tance, there are o nlinear no n -elas t i thout rate limit a o del can be ad j u s o ck has the optio n u stomised Simul i n ctions can be i n arnin g paramete r e plant and its o u o del to be chan ge e wed and assess e rameters. The c u u tput deviatio n fr g orithm takes ar o a n g e) to find th e ve been reache d a n g e. g . 9. Simulink © p o p ecific block can a n g ed. These m o o r nonlinear beh a plant model. T h f our sub-s y ste m a nd sensor filter s y can be select e three models a t ic and nonlinear a tion and satura t s ted to represen t n of classical and i nk © S-functions n corporated wit h r s) and sample ti u tput is the con t e d easil y or the c e d. The neural al u rrent criterion o om the set-point ; o und 10 iteratio n e “best” ran g e o d the parameter s o wer plant mode l be ad j usted. For o dels can represe n a viour ma y be sel h e full h y droel e m s: Guide va n s . Each block is p e d to represent a a vailable to sim u elastic. The g uid e t ion. The sensor t different condi t advanced contr o were develope d h i n Simulink © m o me. Its inputs ar e t rol si g nal. The v c ontrol al g orith m g orithm calculat e o f optimalit y is q ; however this cr i n s (the exact val u o f parameter val u s sta y constant u l . example, the o p n t the power pl a e cted. Fi g ure 9 s h e ctric station m o n e d y namics, p art of the Simu l a diversit y of m o u late the h y drau e vane d y namics filters block is a ions of the natio o llers. d for the neural P o dels. The neur a e the reference a n v ersatilit y of Sim u m to be modifie d e s the optimal v a q uadratic error, i terion can be ch a u e depending o n u es (trainin g ti m u ntil the set-poi n p eratin g point of a nt as a SISO or M h ows a schemati c o del is construc t h y draulic subs y l ink © librar y dev e o des of operati o lic subs y stem - can be selected w a fixed block. T h nal g rid. The g o v P DF al g orithms. a l PDF block ac c n d the output si gn u link © allows th e d and new result s a lues of the cont r where the error a n g ed if necessa r n the ma g nitude m e). When these r n t or the plant linear M IMO c of the t ed by y stem, e loped o n. For linear, w ith or h e g rid v ernor These c epts η n als of e plant s to be r ol law is the ry . The of the r an g es model 6. A s pr o s ys te s Fo an co n pa ba pl a sh o co n co n o p dr i Fi g ca s Simulation re s s discussed previ o vide timel y an d s tem. The actual s tin g , it can be sp e r all simulations , d 50 Hz, and ass u n nected to the n rameters fixed a sis of compariso n a nt under anti- w o ws the small s n nected. In bot h n troller, being r e p erational cases. T i vin g the process g . 10. Small-step s e o f one unit in o s ults ousl y , the role o f d accurate sup p form of the pow e e cified in terms o , the model is e x u mes a Grid s y s t n onlinear model t K=0.1 and T i = 0 n . Figure 10 sho w w indup PI and n e s tep responses ( 0 h cases, the h yd e spectivel y 10% a T he undershoot . response of the o peratio n . f a h y droelectric p l y of its dema n e r demand is rel a o f step, ramp and x pressed in the p t em with infinite of the h y droele c 0 .12 (as currentl y w s the small step r e ural PDF contro l 0 .05 p.u.) of the d roelectric plant a nd 30% faster i n is also reduced i hydro plant wit h station in frequ e n ded power con t a ted to Grid freq u random input si p er-unit s y stem, n busbars. The ne u c tric power plan t y implemented i n r esponses (0.05 p l lers for one uni t power station w performs better n the one unit o p i n both cases w h h neural PDF an d e nc y control mo d t ribution to the u enc y variation b g nals. n ormalized to 3 0 u ral PDF controll t . A PI controll e n practice) is us e .u.) of the h y dro e t operational. Fi g w hen all six un with the neur a p erational and si x h en a PDF contr o d PI controllers f d e is to power b ut, for 0 0 MW er was e r with e d as a e lectric g ure 11 its are a l PDF x units o ller is f or the AUTOMATION&CONTROL-TheoryandPractice172 Fi g ca s Fi g w i re s pe fa s P D T o t= 3 th e an g . 11. Small-step s e o f six units in o g ure 12 shows t h i ndup PI and ne u s ponses (0.35 p. u rformance is be t s ter in, respectiv e D F controller red u o evaluate the cr 3 00 to units 2-6 a e neural PDF res p d a hi g her unde r response of the o peratio n . h e lar g e ramp re s u ral PDF controll e u .) of the power s t ter usin g the n e e ly, the one unit u ces the undersh o oss couplin g int e a nd the perturba p onse has a hi gh r shoot. hydro plant wit h s ponses (0.35 p. u e rs for one unit o s tation when six e ural PDF contr o operational and o ot. e raction a 0.8 p. tion of unit 1 o b h er overshoot, th e h neural PDF an d u .) of the h y dro e o perational. Fi g u r units are gener a o ller, the respon s d six units opera t u. step was ap p b served. Fi g ure 1 4 e PI response ha s d PI controllers f e lectric plant wit h r e 13 shows lar ge a tin g . In both cas s e being 15% a n t ional cases. Aga p lied simultaneo u 4 shows that, al t s a lon g er settli n f or the h anti- e ramp es, the n d 13% in, the u sl y at t hou g h ng time Fi g o n Fi g si x g . 12. The lar g e r a n e unit in operati o g . 13. The lar g e r a x units in operati o a mp response of on . a mp response of on . the hydro plant w the hydro plant w w ith neural PDF w ith neural PDF and PI controlle r and PI controlle r r s with r s with NeuralPDFControlStrategyforaHydroelectricStationSimulator 173 Fi g ca s Fi g w i re s pe fa s P D T o t= 3 th e an g . 11. Small-step s e o f six units in o g ure 12 shows t h i ndup PI and ne u s ponses (0.35 p. u rformance is be t s ter i n , respectiv e D F controller red u o evaluate the cr 3 00 to units 2-6 a e neural PDF res p d a hi g her unde r response of the o peratio n . h e lar g e ramp re s u ral PDF controll e u .) of the power s t ter usin g the n e e l y , the one unit u ces the undersh o oss couplin g int e a nd the perturba p onse has a hi gh r shoot. hydro plant wit h s ponses (0.35 p. u e rs for one unit o s tation when six e ural PDF contr o operational an d o ot. e raction a 0.8 p. tion of unit 1 o b h er overshoot, th e h neural PDF an d u .) of the h y dro e o perational. Fi g u r units are gener a o ller, the respon s d six units opera t u. step was ap p b served. Fi g ure 1 4 e PI response ha s d PI controllers f e lectric plant wit h r e 13 shows lar ge a tin g . In both cas s e being 15% a n t ional cases. A g a p lied simultaneo u 4 shows that, al t s a lon g er settli n f or the h anti- e ramp es, the n d 13% in, the u sl y at t hou g h ng time Fi g o n Fi g si x g . 12. The lar g e r a n e unit in operati o g . 13. The lar g e r a x units in operati o a mp response of on . a mp response of on . the hydro plant w the hydro plant w w ith neural PDF w ith neural PDF and PI controlle r and PI controlle r r s with r s with AUTOMATION&CONTROL-TheoryandPractice174 Fi g 7. In th e pl a o p in c n o th e an T h st o st e ta k be e in fu t 8. T h g . 14. The cross c o Conclusions this chapter a s o e performance o a tform has been i p en architecture m c remental impro o nlinear model o f e Dinorwi g pow e d electrical subs y h e results have s h o ra g e station to i e p response of t h k en into account e n included in t h this application t ure work. Acknowledg m h e authors wish t o o uplin g respons e o ftware tool that f different contr i mportant to gra d m akes possible t h vement of the c f pumped stora ge e r plant. The mo d y stems and conta i h own how the n i mprove its d y n a h e s y stem with n to represent clo s h e nonlinear mo d and encourage u m ents o thank First H yd e of the h y dro pla models a h y dr o ollers has been d ually increasin g h e rapid inclusio n c ontrol approac h e stations has be e d el includes rep r in s the principal f n eural PDF can b a mic response. I n eural PDF is im p s el y the real pla n d el. These are pro m u s to address th e d ro Compan y for n t with PI and n e o power plant an d described. The m g the complexity n of other contro h es and models e n discussed. Thi s r esentation of th e f eatures of the p l b e applied to a h n particular, it h a p roved. Multiva r n t. The coupling b m isin g results fo r e issue of robust n their assistance. e ural PDF contro l d allows compar i m odular nature o of the simulatio n l methods and a l alread y include d s model was ap p e g uide vane, h yd l ant’s d y namics. hy droelectric pu a s been shown t h r iable effects hav e b etween pensto c r the use of neur a n ess of the resp o l lers. i son of o f this n s. The l so the d . The p lied to d raulic mped- h at the e been c ks has a l PDF o nse in 9. References Aguado-Behar, A., “Topics on identification and adaptive control” (in Spanish), Book edited by ICIMAF, La Habana, Cuba. 2000. Bohn, C. and Atherton, D. P., "An analysis package comparing PID anti-windup strategies", IEEE Control Systems Magazine, vol. 15, p.p. 34-40. 1995. Djukanovic, M. B., Calovic, M. S., Vesovic, B. V., and Sobajic, D. J., “Neuro-fuzzy controller of low head hydropower plants using adaptive-network based fuzzy inference system”, IEEE Trans. on Energy Conversion , 12, pp. 375-381. 1997. Garcez, J. N., and Garcez, A. R., “A connectionist approach to hydroturbine speed control parameters tuning using artificial neural network”, Paper presented at 38th IEEE Midwest Symposium on Circuits and Systems, pp. 641-644. 1995. Goodwin, G. C., Graebe, S. F. and Salgado, M. E., "Control system design", Prentice Hall, USA. 2001. Gracios, C., Vargas, E. & Diaz-Sanchez A., “Describing an IMS by a FNRTPN definition: a VHDL Approach”, Elsevier Robotics and Computer-Integrated Manufacturing, 21, pp. 241–247. 2005. Kang, J. K., Lee, J. T., Kim, Y. M., Kwon, B. H., and Choi, K. S., “Speed controller design for induction motor drivers using a PDF control and load disturbance observer”, Paper presented at IEEE IECON, Kobe, Japan, pp. 799-803. 1991. Kundur, P., Power System Stability and Control, New York, NY: Mc Graw Hill. 1994. Mansoor, S. P., “Behaviour and Operation of Pumped Storage Hydro Plants”, Bangor, U.K.: PhD. Thesis University of Wales. 2000. Mansoor, S. P., Jones, D. I., Bradley, D. A., and Aris, F. C., “Stability of a pumped storage hydropower station connected to a power system”, Paper presented at IEEE Power Eng. Soc. Winter Meeting, New York, USA, pp. 646-650. 1999. Mansoor, S. P., Jones, D. I., Bradley, D. A., Aris, F. C., and Jones, G. R., “Reproducing oscillatory behaviour of a hydroelectric power station by computer simulation”, Control Engineering Practice, 8, pp. 1261-1272. 2000. Miller T., Sutton S.R. and Werbos P., Neural Networks for Control, Cambridge Massachusetts: The MIT Press. 1991. Minsky, M. L., and Papert, S. A., Perceptrons: Introduction to Computational Geometry. Cambridge, USA: MIT Press. 1988. Munakata, T., Fundamentals of the New Artificial Intelligence: Neural, Evolutionary, Fuzzy and More. London, UK: Springer-Verlag. 2008. Narendra, K. S., and Mukhopadhyay, S. “Adaptive control using neural networks and approximate models”, Paper presented at American Control Conference, Seattle, USA, pp. 355-359. 1996. Peng, Y., Vrancic, D. and Hanus, R., "Anti-windup, bumpless, and conditioned transfer techniques for PID controllers", IEEE Control Systems Magazine, vol. 16, p.p. 48-57. 1996. Rumelhart, D. E., McClelland, J. L., and Group, T. P., Parallel distributed processing: Explorations in the microstructure of cognition (Vol. 1). Cambridge, USA: MIT Press.1986. Shu-Qing, W., Zhao-Hui, L., Zhi-Huai, X., and Zi-Peng, Z. “Application of GA-FNN hybrid control system for hydroelectric generating units”, Paper presented at International Conference on Machine Learning and Cybernetics 2, pp. 840-845. 2005. NeuralPDFControlStrategyforaHydroelectricStationSimulator 175 Fi g 7. In th e pl a o p in c n o th e an T h st o st e ta k be e in fu t 8. T h g . 14. The cross c o Conclusions this chapter a s o e performance o a tform has been i p en architecture m c remental impro o nlinear model o f e Dinorwi g pow e d electrical subs y h e results have s h o ra g e station to i e p response of t h k en into account e n included in t h this application t ure work. Acknowledg m h e authors wish t o o uplin g respons e o ftware tool that f different contr i mportant to g ra d m akes possible t h vement of the c f pumped stora ge e r plant. The mo d y stems and conta i h own how the n i mprove its d y n a h e s y stem with n to represent clo s h e nonlinear mo d and encoura g e u m ents o thank First H yd e of the h y dro pla models a h y dr o ollers has been d uall y increasin g h e rapid inclusio n c ontrol approac h e stations has be e d el includes rep r in s the principal f n eural PDF can b a mic response. I n eural PDF is im p s el y the real pla n d el. These are pro m u s to address th e d ro Compan y for n t with PI and n e o power plant an d described. The m g the complexit y n of other contro h es and models e n discussed. Thi s r esentation of th e f eatures of the p l b e applied to a h n particular, it h a p roved. Multiva r n t. The coupling b m isin g results fo r e issue of robust n their assistance. e ural PDF contro l d allows compar i m odular nature o of the simulatio n l methods and a l alread y include d s model was ap p e g uide vane, h yd l ant’s d y namics. hy droelectric pu a s been shown t h r iable effects hav e b etween pensto c r the use of neur a n ess of the resp o l lers. i son of o f this n s. The l so the d . The p lied to d raulic mped- h at the e been c ks has a l PDF o nse in 9. References Aguado-Behar, A., “Topics on identification and adaptive control” (in Spanish), Book edited by ICIMAF, La Habana, Cuba. 2000. Bohn, C. and Atherton, D. P., "An analysis package comparing PID anti-windup strategies", IEEE Control Systems Magazine, vol. 15, p.p. 34-40. 1995. Djukanovic, M. B., Calovic, M. S., Vesovic, B. V., and Sobajic, D. J., “Neuro-fuzzy controller of low head hydropower plants using adaptive-network based fuzzy inference system”, IEEE Trans. on Energy Conversion , 12, pp. 375-381. 1997. Garcez, J. N., and Garcez, A. R., “A connectionist approach to hydroturbine speed control parameters tuning using artificial neural network”, Paper presented at 38th IEEE Midwest Symposium on Circuits and Systems, pp. 641-644. 1995. Goodwin, G. C., Graebe, S. F. and Salgado, M. E., "Control system design", Prentice Hall, USA. 2001. Gracios, C., Vargas, E. & Diaz-Sanchez A., “Describing an IMS by a FNRTPN definition: a VHDL Approach”, Elsevier Robotics and Computer-Integrated Manufacturing, 21, pp. 241–247. 2005. Kang, J. K., Lee, J. T., Kim, Y. M., Kwon, B. H., and Choi, K. S., “Speed controller design for induction motor drivers using a PDF control and load disturbance observer”, Paper presented at IEEE IECON, Kobe, Japan, pp. 799-803. 1991. Kundur, P., Power System Stability and Control, New York, NY: Mc Graw Hill. 1994. Mansoor, S. P., “Behaviour and Operation of Pumped Storage Hydro Plants”, Bangor, U.K.: PhD. Thesis University of Wales. 2000. Mansoor, S. P., Jones, D. I., Bradley, D. A., and Aris, F. C., “Stability of a pumped storage hydropower station connected to a power system”, Paper presented at IEEE Power Eng. Soc. Winter Meeting, New York, USA, pp. 646-650. 1999. Mansoor, S. P., Jones, D. I., Bradley, D. A., Aris, F. C., and Jones, G. R., “Reproducing oscillatory behaviour of a hydroelectric power station by computer simulation”, Control Engineering Practice, 8, pp. 1261-1272. 2000. Miller T., Sutton S.R. and Werbos P., Neural Networks for Control, Cambridge Massachusetts: The MIT Press. 1991. Minsky, M. L., and Papert, S. A., Perceptrons: Introduction to Computational Geometry. Cambridge, USA: MIT Press. 1988. Munakata, T., Fundamentals of the New Artificial Intelligence: Neural, Evolutionary, Fuzzy and More. London, UK: Springer-Verlag. 2008. Narendra, K. S., and Mukhopadhyay, S. “Adaptive control using neural networks and approximate models”, Paper presented at American Control Conference, Seattle, USA, pp. 355-359. 1996. Peng, Y., Vrancic, D. and Hanus, R., "Anti-windup, bumpless, and conditioned transfer techniques for PID controllers", IEEE Control Systems Magazine, vol. 16, p.p. 48-57. 1996. Rumelhart, D. E., McClelland, J. L., and Group, T. P., Parallel distributed processing: Explorations in the microstructure of cognition (Vol. 1). Cambridge, USA: MIT Press.1986. Shu-Qing, W., Zhao-Hui, L., Zhi-Huai, X., and Zi-Peng, Z. “Application of GA-FNN hybrid control system for hydroelectric generating units”, Paper presented at International Conference on Machine Learning and Cybernetics 2, pp. 840-845. 2005. [...]... process mechanism and main functions of CD, the partial directory can be described as shown in Figure 5 It shows information of CF (lines 1-9 ) and members of cluster (lines 2 0-2 2), the cluster directory also records meta-data about cluster such as cluster name (line 12), cluster description (lines 1 3-1 5), ontology used in cluster (lines 1 6-1 8) , and so on 190 AUTOMATION & CONTROL - Theory and Practice 1 . Conference on Machine Learning and Cybernetics 2, pp. 84 0 -8 45. 2005. AUTOMATION & CONTROL - Theory and Practice1 76 Werbos, P. J., Beyond regression: New Tools for Prediction and Analysis in the Behavioral. plant w w ith neural PDF w ith neural PDF and PI controlle r and PI controlle r r s with r s with AUTOMATION & CONTROL - Theory and Practice1 74 Fi g 7. In th e pl a o p in c n o th e an T h st o st e ta k be e in. Transactions on Energy Conversion, vol. 4, p.p. 45 3-4 58. 1 989 . Yin-Song, W., Guo-Cai, S., & Ong-Xiang, “The PID-type fuzzy neural network control and it's application in the hydraulic turbine