Fuzzy Control of a Supercritical Once-Through Boiler

Transcription

1 Fuzzy Contol of a Supectcal Once-hough Bole S. Al A. Moosavan Al Ghaffa 2 Al Chabakhsh 3 Depatment of Mechancal Engneeng. N. oos Unv. of echnology, ehan, Ian, P.O. Box , Fax: ( Emal: moosavan@kntu.ac. Abstact Inceasng the use of electcty and need fo moe and safe powe geneaton has motvated nvestgaton nto new contol methods esultng n bette pefomance. Bette system pefomance means ncease n powe geneaton effcency, also decease n the mantenance costs. o desgn sutable contolles, adequate nfomaton about the system dynamcs s equed. In ths pape, fo the supectcal once-though bole of Neka powe plant, a multlaye neuo fuzzy model s used to desgn P 2 ID contolles. Based on bole eal pefomance, these contolles ae desgned to contol supeheates and eheates tempeatue. Smulaton esults show vey good pefomance of these contolles n tems of moe accuate and less fluctuaton n the tempeatue of coespondng subsystems. eywods: Contol, Powe Plants, Fuzzy Logc. I. Intoducton. In powe geneaton, an essental equement s to acheve optmal opeaton n tems of vaant obectves, such as mnmzaton of load tackng eos, mnmzaton of fuel consumpton and heat ate, maxmzaton of duty lfe, and mnmzaton of pollutant emssons. Load tackng, voltage and fequency stablty have been the basc ssues of moe effectve contol desgn, []. Geneal dutes fo the powe plant contol systems ae expessed below. Capablty of geneatng a desed load fo a wde ange of opeaton commanded by the dspatchng cente; Havng a fast esponse to the set pont changes; Accuate pefomance even n the pesence of dstubances; Safe and smooth opeaton of powe plant; Reducton of themal stesses due to steam tempeatue and pessue fluctuatons, that n tun mnmzes opeatonal and mantenance costs; Reducton of the fuel consumpton, due to hghe effcency of combuston; Reducton of the envonmental pollutons; Reducton of the opeatos supevsoy effots; Fault dagnostcs capabltes to mnmze the mantenance costs. Contol of fossl fuel boles has been extensvely addessed, usng classc and moden contol appoaches, [2-4], lnea o non-lnea model of bole, [5-6]. Despte the sgnfcant developments n the contol theoes, PID contolles ae stll the mao contolles used n the ndusty. In the last thee decades, fuzzy logc contol has evolved as an altenatve o complementay to the conventonal contol stateges n vaous applcatons, [7-8]. he fuzzy logc contolles ae ecommended fo systems wth nonlneaty and uncetanty. Unlke conventonal contols, desgnng a fuzzy logc contolle does not eque pecse knowledge of the system model such as the poles and zeoes of the system tansfe functon, [9]. It has been shown a hybd scheme of fuzzy contol and PID contol has a bette pefomance compaed to the conventonal classc and fuzzy contol algothms, [0-2]. o mplement dffeent contol laws and stateges, havng models of system dynamcs s necessay. hese models fo steam powe plants can be obtaned based on fundamental laws of Physcs such as conveson of mass, - Assstant Pofesso. 2 - Assocate Pofesso. 3- Gaduate Student.

2 momentum and enegy, sem-empcal laws fo Heat tansfe and hemodynamcs, [3-4]. Such models can be also developed usng mathematcal dentfcaton pocedues based on measued data dung eal wokng state of the plant, [5-6]. he pocedue to detemne a test-data-based model fom nput-output bole data, nvolves fou followng steps, [7]: Expements and collect nput-output data Elect and defne a model stuctue Estmate the paametes of model Model valdaton In ths appoach In ths pape, two fuzzy P 2 ID contolles ae desgned and mplemented fo the supeheate and eheate tempeatue contol of Neka powe plant. hs new contolle has benefts of both fuzzy PD and PI contolles. o ths end, an Adaptve Neuo Fuzzy Inteface System (ANFIS s used to pepae the plant model based on eal data measued at the powe plant. hese fuzzy models ae developed to cope wth nonlneaty and uncetanty. Also, these nonlnea fuzzy models ae able to cove a wde ange of opeatonal condtons. Smulaton esults show the effect of usng these new fuzzy P 2 ID tempeatue contolles on the cycle pocesses. II. System Descpton A 440 MW unt (Unt #2 at Neka Powe Plant s studed fo dentfcaton, modelng, and new contolles desgn. Neka powe plant s located at 25 km fa fom Neka, a cty n Nothen Ian. he plant conssts fou fossl fueled geneatng unts wth the ated capacty of 440 MW. he steam geneato of each unt s a supectcal once though Benson bole, Fg.. A sngle funace wth 4 bottom bunes seves to heat all pats of bole. he hot wate s conveted to steam n evapoato, and gets supeheated by passng though supeheates. he evapoato les n the aea of the hghe flue gas tempeatue and s suppled as a spal belt. he evapoato outlet tempeatue s about 365 o C. Neka supeheate conssts of 4 sectons bult n the bole second pass. he heatng suface of supeheate # s located n the bole walls. he steam leaves supeheate # outlet heade, and goes towad supeheate #2. he outlet steam tempeatue fom these sectons should be constant at 407 o C. hese sectons have an emegency tempeatue contol that woks ust when the bole s ovehaulng. he steam leaves supeheate #2 towad supeheate #3. Fo apd steam tempeatue contol, fou spay attempeatos have been consdeed between the two sectons. In the de-supeheate mxng path, atomzed feedwate s nected by nozzles nto the steam path, theeby, educng the steam tempeatue. he spay flow s taken fom feed wate pumps and should not nomally exceed appoxmately 6% of the bole load at the tme. he mean wate spay flow fom each nozzle s 0 ton pe hou. In ths case, the outlet tempeatue should be kept constant at 470 o C. he steam tempeatue contol of supeheate #3 s shown n Fg. 2. Its contol system s a cascade contol method that can mpove system pefomance compaed to sngle closed-loop contol by lmtng the effect of the dstubances of the seconday vaable (nlet steam tempeatue on the pmay output (outlet steam tempeatue. In oute contol loop, the supeheate #3 outlet tempeatue s compaed wth ts set pont (470 o C to yeld an eo sgnal fo a PI contolle. hs contolle adusts the set pont fo nne contol loop to be compaed wth supeheate #3 nlet tempeatue. Fo a bette pefomance, the ate of fuel consumpton sgnal (as a pedcton sgnal and hgh pessue steam flow sgnal ae consdeed n the contol loop. hs sgnal goes to a P-Acton contolle to adust the wate spay valve stem poston. he supeheate #4 tempeatue contol s same as that of #3. he steam tempeatue at outlet heade of bole should be constant at 535 o C. At full load condton, the outlet supeheated steam pessue at bole outlet heade s 8. MPas. he supeheated steam fom man steam heade s fed towad the hgh pessue (HP tubne, and fom hgh pessue tubne s dschaged nto the cold eheat heade, Fg. 3. he steam tempeatue n cold eheat lne s 35 o C. he eheate conssts of two sectons. hee ae fou attempeatos between these sectons. he eheate outlet tempeatue s contolled n the same way to supeheate sectons wth small dffeences (.e. pedcton sgnals ae geneated by the ate of steam tempeatue and feed wate flow sgnals, Fg. 4. he spay flow s taken fom an low pessue (LP stage of the feed pump. he outlet eheated steam tempeatue afte RH-A should be constant at 452 o C and fo RH-B should be constant at 535 o C. Also, at the full load condton, the outlet eheated steam pessue s 4.35 MPas. he eheated steam s used to feed the ntemedate pessue (IP tubne. Exhaust steam fom IP tubne s fed nto the low pessue tubne. Also the extacton steam fom IP tubne s fed nto the feedwate tubo pump and feedwate stoage tank (de-aeato. Fou low pessue heates ae fed by the extacton steam fom LP tubne. he low qualty low pessue steam fom all pats s dschaged nto the man condense. hs condense s an open loop condense usng the sea wate to cool up the hot wate and steam. he condensate wate s pumped nto the feedwate stoage tank va a tan of LP heates. A feedwate tubo pump wth 3.2 MW powe (o two electcal pumps wth 9MW powe when the tubo pump fals pepaes compessed feedwate at 28 MPas Pessue fo bole. hen, the feedwate s fed nto the hgh pessue heates and ts tempeatue s nceased about 80 o C. he hot feedwate nlet nto the economze heade and the powe geneaton cycle s epeated. A concse descpton of bole HP sectons, and IP pats s summazed n ables I and II, espectvely.

3 III. Fuzzy Model Development. he fuzzy logc modelng s best suted fo systems wth nonlneaty and uncetanty such as steam powe plants, [8-9]. Most of pocesses n a steam bole have mult nput and mult output (e.g., tempeatue, pessue. he S fuzzy models can be expessed by a set of followng typcal ules, [] and [20]: IF x s LX and... and xn s ( LX n, HEN y = c0 + c x cn xn whee LX s a membeshp functon assocated wth the nput vaable x n. Neual netwok leanng technques facltate these fuzzy model paamete tunng. he neual netwoks to pepae these models ae the geneal-pupose adaptve neuo fuzzy nfeence system (ANFIS technque. In the ANFIS method, a fuzzy system wth n nputs and N ules s epesented by a fve-laye feed fowad netwok stuctue wth N neual pocessng unts n layes L, L 2, L 3, and L 4, and a sngle unt n L 5, [4], as follows. - Laye (L : Each neuon n laye L fuzzfes the ncomng nput sgnal usng Gaussan membeshp functons: ( x (2 µ = x α 2 ( β e - Laye 2 (L 2 : At ths laye by multplyng all ncomng values, each node calculates the degee of fulfllment of the coespondng ule. n = = τ µ ( x (3 - Laye 3 (L 3 : Each unt calculates a elatve degee of fulfllment of the coespondng ule by nomalzng ts degee of fulfllment wth espect to the degees of fulfllment of all the ules. τ τ (4 = n = τ - Laye 4 (L 4 : Each node calculates the consequent of the coespondng ule weghted by ts elatve degee of fulfllment. y = τ y = τ c + c x c x (5 ( 0 n n - Laye 5 (L 5 : he only neual unt n L 5 s connected to all unts n L 4. he node calculates the fnal output, y, of the fuzzy system by addng all the ncomng weghted consequents. y = N y = All nput-output patents can be defned as follows: (6 Y M = X M ( n+ NC( n+ N (7 Usually M > (n + N and thee ae moe pattens than paametes to calculate. In ths case, thee s no exact soluton fo equaton (7. o estmate C, the squaed eo 2 XC - Y s mnmzed. So, C s obtaned as below. Clse = ( X X XY (8 Also C can be computed ecusvely by Goodwn and Sn adaptng method. C+ = C + ψ + x+ ( y+ x+ C ψ x+ x+ ψ (9 ψ + = ψ, = 0,,2,..., M + x+ ψ x+ he paametes of membeshp functon ae detemned by backpopagaton. Fo any paametes of membeshp functons z, the change n z fo a sngle ule afte a patten has been popagated,.e. z, s obtaned as:. E z = σ (0 z whee σ s an abtay leanng ate facto and E s the usual eo gven by the sum of squaed dffeence between the taget output y* and actual output y: P * ( p 2 E( W = ( y p y ( 2 p= By applyng chan ule fo each paamete we have. E E µ z = σ y µ z (2 σ µ = y ( y * y τ ( τ µ z o mplement the above pocedue, each bole secton s assumed as a fuzzy system wth 3 nputs and one output. Fo nstance, fuel flow, the po secton output tempeatue and the steam flow ae consdeed as system nputs and the tempeatue of outlet s used as system output. hese fuzzy systems ae adapted wth hybd leanng algothm. Fst, the nput pattens ae popagated keepng the antecedent paametes constant, and then the optmal consequent paametes ae estmated ecusvely usng the least squae estmaton pocedue n (9. Next, the nput pattens ae popagated agan whle keepng the consequent paametes constant, then the antecedent paametes ae modfed by back popagaton usng (2. he fst step fo pepang the model s gatheng the bole subsystems data fom powe plant unts. he unt data ae ecoded n Unts Data Access System (UDAS at the powe plant contol oom that ae easly accessble. hese expemental eal data ae ecoded fo 60 hous wth a 5 second samplng tme. About 20,000 steps of the collected data ae used n the leanng pocedue. he fuzzy models ae adapted fo the full ange of vaaton. So, when the load s amped down fom 00 to 50 pecent, the models ae taned and when load s amped up fom 50 to

4 00 pecent, the models ae checked. Followng ths pocedue, the obtaned models fo dffeent bole sectons have been valdated based on compasons wth eal measued data, [2]. Next, two fuzzy P 2 ID contolles ae desgned and mplemented fo the supeheate and eheate tempeatue contol of Neka powe plant. IV. Contolle Desgn and Smulatons. he contol system at Neka powe plant s a conventonal system havng electonc P and PI cads. hs powe plant has been bult n 978, and snce then the contol system has lost ts dese effcency. o eplace ths old contol system wth new PC-based one, a new Fuzzy P 2 ID Contolle may be suggested as explaned n ths secton. he am of the new suggested contol system s to acheve smultaneously a fast se tme and a mnmum steady-state eo wth obustness wthn an acceptable toleance. It s known that the fuzzy PD contolle can handle the fome goal, whle the fuzzy PI contolle can deal wth the latte, [22-23]. he Fuzzy PI contolles used fo nonlnea systems s shown n Fg. 5. he output of the dgtal fuzzy PI contolle n the dscete tme-doman s as follows: upi ( n = upi ( n + u u ( PI PI n (3 upi ( n = Pev ( n + I ep ( n whee >0 s the samplng tme, P and I ae the popotonal and ntegal gans, espectvely, and uip s a fuzzy contol gan s detemned followng the desgn pocedue, and: upi ( n upi ( n upi ( n = (4 e( n e( n ev ( n = Hee, u s the ncemental contol output, e PI P the eo sgnal, and e I the ate of change of the eo sgnal. he Fuzzy PD contolles ae the most conventonal used fo nonlnea systems, Fg. 6. he output of the dgtal fuzzy PD contolle n the dscete tme-doman s as follows: upd ( n = upd ( n + u u ( PD PD n (5 upd ( n = dep ( n + pe ( n he poposed Fuzzy P 2 ID contolle s an ntegated scheme that combnes the fuzzy PI and fuzzy PD contol actons, whee the advantages of both contol actons ae utlzed to compensate and complement each ndvdual shotcomng. he stuctue of ths new contolle s shown n Fg. 7. heefoe, t s obtaned: u( n = upd ( n + upi ( n (6 he fuzzy membeshp functons and ules ae the same fo both Fuzzy PI and PD Contolles. Fo each nput, thee ae thee membeshp functons, whle fo the output thee ae seven ones. In the FPI, the fome nput s e and the p latte s Pe. In the FPD, the fome nput s v de and the p latte s e. he contolles nputs, e P v, e and e p ae nomalzed to le n the nteval (0,, and output s nomalzed to le n the nteval (0,20, Fg. 8. he dynamc behavo of the bole subsystems vaes wth the load condton. So, fo mpovng the contol pefomance, dung a ange of opeaton, the contolle gans ae changed accodng to load vaatons. o ths am a S fuzzy system s used to change the gans. he fuzzy P 2 ID contolle gans fo dffeent loads ae pesented n able III. hese sets of gans wee obtaned by fne-tunng of seveal tals to get the best possble esults fo a fa compason. hese contolles ae appled to contol the supeheate and eheate tempeatue, whle othe subsystems utlze the old contolles. he effect of these Fuzzy P 2 ID contolles s obseved on all pats tempeatue. he mpovement n steam tempeatue and pessue fluctuatons nceases the geneated powe [2]. Some sample esults at steady state phase and 97% load (427 MW ae shown n fgues 9 to 3. As t s seen, wth the new contolles the tempeatue fluctuatons n bole pats ae deceasng and the wate spay flow used to de-supeheat the hot steam n the supeheate and eheate s clealy less than the conventonal P and PI contolles, fgues 9 and 0. Also, the mplementaton of new contolles affects the outlet steam pessue, and nceases the geneated powe whle the fuel consumpton s educed, fgues to 3. V. Conclusons. In ths pape, two fuzzy P 2 ID contolles wee desgned and mplemented fo the supeheate and eheate tempeatue contol of Neka powe plant. hs new contolle has benefts of both fuzzy PD and PI contolles. An Adaptve Neuo Fuzzy Inteface System (ANFIS was used to pepae the plant model based on eal data measued at the powe plant. hese fuzzy models have been developed to cope wth nonlneaty and uncetanty. Also, these nonlnea fuzzy models ae able to cove a wde ange of opeatonal condtons. he pupose of the contolle desgn s to have hghe and safe powe geneaton. he benefts of new contolle ae faste se tme and lowe steady-state eo wth consdeable obustness wthn an acceptable toleance. Smulaton esults wee dscussed,, and evealed that usng these new fuzzy P 2 ID tempeatue contolles has sgnfcant effect on the cycle pocesses. Refeences [] Ramez, R.G., Oveall Intellgent Hybd Contol System fo Fossl Fuel Powe Plant, PhD hess, he Pennsylvana State Unvesty, he Gaduate School, Depatment of Electcal Engneeng, [2] Gey, H.C., he evoluton of coodnated contol, Instument Socety of Ameca. Pape #88-047, pp

5 [3] Lands, R. and E. Wulfsohn he contol phlosophy fo a unt contol system fo coodnated opeaton of a bole and tubne Elekton. Febuay 988. [4] Ghezelayagh, H. and Lee,.Y., Intellgent Pedctve Contol of a Powe Plant wth Evolutonay Pogammng Optmze and Neuo- Fuzzy Identfe IEEE, /02, [5] Ben-Abdennou, A. and Lee,.L., A Decentalzed Contolle Desgn fo Powe Plant Usng Robust Local Contolles And Functonal Mappng, IEEE ansacton on Enegy Conveson, Vol., No. 2, June 996. [6] Rovank, J.A. and Cols, R., Dynamc Matx Based Contol of Fossl Powe Plant, IEEE tansacton on Enegy Conveson, Vol. 6, No.2 June 99. [7] Puessmann, D., ause, B. and Von Altock, C, Fuzzy Logc Supevsoy Contol fo Coal Powe Plant, IEEE, /97, 997. [8] Lee,.B., Applcaton of a Fuzzy Logc Based Self unng PID Contolle o he Contol of Dum Wate Level of a Bole fo a Powe Geneatng Plant IEE 2nd Intenatonal Confeence n Advances n Powe System Contol, Opeaton and Management, Decembe 993. [9] E. Mamdan, Advances n the Lngustc Synthess of Fuzzy Contolles, Intenatonal Jounal of Man-Machne Studes, 8( , 976. [0] Ramez, R.G. and Lee..Y., Wde Range Opeaton of a Powe Unt va Feed Fowad Fuzzy Contol, IEEE ansactons on Enegy Conveson, Vol. 5, No. 4, Decembe [] Behesht, M..H. and Rezaee, M.M., A New Hybd Maste Contolle, Poceedng of he Amecan Contol Confeence, Anchoage, A May 8-0, [2] Chen, G., Conventonal and fuzzy PID contolle: An ovevew, Intenatonal Jounal of Intellgent Contol System, 996. [3] Weng, C.., Ray, A. and Da, X., Modelng of Powe Plant Dynamcs and Uncetantes fo Robust Contol Synthess, Applcaton of Mathematcal Modelng, Vol. 20, Elseve Scence Inc, July 996. [4] Changlang, L., Jzhen, L. and Yuguang, N. and Wepng, L. Nonlnea Bole Model of 300MW Powe Unt fo System Dynamc Pefomance Studes, IEEE, /0, 200. [5] Ghezelayagh, H. and Lee,.Y. Applcaton of Neuo Fuzzy Identfe n the Pedctve Contol of Powe Plant 5th ennal Wold Congess, Bacelona, Span, [6] Chen, F.C. and hall H.., Adaptve contol of nonlnea systems usng neual netwoks, Int.J. Contol, 55( , 992. [7] Lung, L., System Identfcaton heoy fo the use, Pentce Hall, Uppe Saddle Rve, NJ, 987. [8] Chang, X. and L, W., A C-Mean Clusteng Based Fuzzy Modelng Method, IEEE, /00, [9] Uk Youl Huh and Jn Hawn m MIMO Fuzzy Model fo Bole-ubne System IEEE, /96, 996. [20] Roge, J.S., Jang, C.. Sun and Mtutan, E., Neuo-Fuzzy and Soft Computng, Pentce Hall, Uppe Saddle Rve, NJ 07458, 997. [2] Chabakhsh, A., Neuo-Fuzzy Contol n a Powe Plant Subsystem, MSc. hess,. N. oos Unvesty of echnology, Mechancal Engneeng Depatment, Sept [22] Sooaksa, P., Pattaade,., and Chenb, G. Desgn and Implementaton of Fuzzy P 2 ID Contolle fo Handleba Contol of a Bcycle Robot, Cty Unvesty of Hong ong, owloon, Hong ong SAR, P. R. Chna. [23] Ms, D., Malk, H.A. and Chen, G., Desgn and analyss a fuzzy popotonal-ntegal-devatve contolle, Fuzzy Set and System 79, 996. able I: Descpton of steam geneato (HP pats M.C.R. Unt Fuel Gas Fuel Ol Electcal Rate MW Steam ate t/h Acceptance P. MPas 2 2 Feedwate emp. o C Inlet P. MPas Outlet P. MPas Pessue Loss MPas empeatue o C able II: Descpton of steam geneato (IP pats M.C.R. Unt Fuel Gas Fuel Ol Steam Flow ate t/h Acceptance P. MPas Reheate Inlet em. o C Reheate Outlet em. o C Inlet Pessue MPas Outlet Pessue MPas Pessue Loss MPas Fuel Flow g/s Cold A em. o C A em. Befoe AH o C Excess Com. A Exhaust Gas em. o C Effcency %

6 Load (MW able III: Fuzzy P 2 ID Contolle Gan p d u PI upd Fg. 2: Supeheate empeatue Contol Fg. 3: Subsystems of Powe Unt Fg. 4: Reheate empeatue Contol Fg. : Bole Coss Secton Fg. 5: Fuzzy PI Contolle

7 Fg. 6: Fuzzy PD Contolle Fg. 0: Reheate-B outlet empeatue Fg. 7: Fuzzy P 2 ID Contolle Fg. : Bole outlet Pessue Fg. 8: Fuzzy Inteface Suface Fg. 2: Geneated Powe Fg. 3: Fuel consumpton Fg. 9: Supeheate #4 outlet empeatue