Neuro-Fuzzy Tuning of PID Controller for Control of Gas Turbine Power Plant

Transcription

1 NeuroFuzzy Tunng of PID Controller for Control of Gas Turbne Power Plant Km, Dong Hwa, Won Pyo Hong Dept. of I&C, Hanbat Natonal Unversty San Duck myongdong Yu songgu Daejon cty Seoul Korea,. Tel:887, Fax:88 Abstract The purpose of ntroducng Combned Cycle wth gas turbne n power plants s to reduce losses of energy. Ther man role les n the utlzaton of waste heat that may be found n exhaust gases from the gas turbne or at some other ponts of the process to produce addtonal electrcty. The effcency of the plant ncreases reachng over %, whle the tradtonal steam turbne plants s approxmately %% or so. Up to date, the PID controller has been used to operate under such systems, but snce the gan of PID controller manually has to be tuned by tral and error procedures. Gettng an optmal PID gans s very dffcult to tune manually wthout control desgn experence. In ths paper, we studed an acqurng of transfer functon from operatng data of Gunsan gas turbne n Korea and a new DOF PID controller tunng by N s desgned for the optmum control to Gunssan gas turbne's varables varety. Snce the shape of a membershp functon n the N vary on the characterstcs of plant. ANFIS based control method s effectve for plant that ther varables vary. Its results are compared to the conventonal PID, DOF PID controller and represents satsfactory response. We expect ths method wll be used for another process because t s studed on the real operatng data. Key Words : Intellgent control, Neuro control, Fuzzy control, DOF control, Gas turbne control. Introducton The studes on the control of gas turbne have been a subject of nterest for many years, snce the gasturbne engnes have been wdely adopted as peak load canddates for electrcal power generaton. Especally, the fully automatc startup functon and the fast runup characterstc of gas turbne systems have made them partcularly sutable for peakload loppng and standby power supply purposes []. The startup procedure for a modern gas turbne conssts of the stages such as, warmng up of man steam ppelne, warmng up of turbne parts, turbne runup, synchronzaton, and loadng. So, the varous studes about control n each step from startup to loadng need to have stablty and safety because control procedure s very complcated. Startup and shutdown procedures are the most challengng problems for control applcatons to develop new control algorthms. They requre the sequence of operatons to be successfully performed leadng a gas turbne and assocated power plant components through a sequence of safe states. At the same tme many varables must be montored and checked to ensure safety of operaton []. Moreover, mnmum tme and mnmum energy losses durng the starup or runup procedures would be desrable. Addtonal problems are nvolved when startngup or shuttng down certan components wthn a bgger power staton. Ths s especally dffcult n case of Combned Cycle gas turbnes because, n ths case, nterconnectons between components occur not only through the electrc network but also on the heat exchange sde,.e. flows of gases, and flows of steam []. An effectve control s requred to mantan system stablty followng a system dsturbance. Falure to do so wll cause an nevtable plant shutdown, from whch a loss of producton and consderable damage to the plant may result. There s an ncreasng demand for a more accurate gasturbne control technology than those prevously studed [], to enable the system response be stablzed and mprovements to the assocated control system made. In ths paper, frst of all, we desgned the new DOF PID controller to the operatng data based transfer functon of Gunsan gas turbne generaton plant n Korea and studed an optmum tunng of a new DOF PID controller desgned usng NF (Neural network Fuzzy System).. Gunsan gas turbne system. The structure and characterstcs of GunSan gas turbne

2 Generally, the model s composed of compressor, combustor, turbne, and fuel system wth an arrangement shown n Fg. [~]. The hghpressure compressor turbne powered by the exhaust gas from the combuston chamber drves the compressor. The combnaton acts as a generator for the lowpressure free power turbne. The power turbne drves a generator through a gearbox. The gas generator s derved from sngleshaft engne. Fg. (a) Gasturbne generatng system W ar W fuel W nj Par Gas turbne P ω g no T o W o Fg. (b) The control varables of gasturbne generatng system Where, W ar =Ar flow to the turbne W fuel =Fuel flow to the turbne = Steam (water) njecton flow W nj P ar Fuel Compressor Ar Combuster Gas turbne Hot gas =Enthalpy, densty, pressure and temperature of nlet ar, temperature of njecton lqud P =Mechancal power ω=rotatonal speed g no = N ox content n exhaust gas T o =Exhaust gas temperature W o =Exhaust gas flow h o =Exhaust gas enthalpy Generator Ar at atmospherc pressure enters the compressor nlet. After compresson of the ar to acheve the most favorable condtons for combuston, fuel gas s mxed wth the ar n the combuston chamber, combuston takes place and h o the hot exhaust gases are expanded through the turbne to produce mechancal power. So, f we want to model, a gas turbne generaton plant can be represented by a lnear model whch s devsed assumng that the engne can be represented as a collecton of multvarable functons, whch can be lnearzed by wrtng them n a total dfferental form. Ths approach method has been extensvely used n gasturbne control studes up to now []. But t s not easy for us to model and control. In a gasturbne engne, the fuel flow s a true ndependent varable wth respect to the engne. The engne speed s an ndependent varable wth respect to the thermodynamc cycle but a dependent varable n the nerta/speed relatonshp. The dependent varables chosen are the compressor dscharge pressure, the exhaust gas pressure, the exhaust gas temperature and the exhaust gas power [, ]. Control system of Gunsan gas turbne plant Fg. represents a smplfed block dagram for a sngleshaft Gunsan gas turbne generaton plant. The control system ncludes speed control, temperature control, acceleraton control, and upper fuel and lower fuel lmts. Speed control s sutable for ether droop or sochronous control and operates on the speed error formed between a reference made up of one per unt speed plus the setpont, compared wth actual system or rotor speed. A droop governor s a straght proportonal speed controller n whch the output s proportonal to the speed error [,]. Temperature control s the normal means of lmtng gas turbne output at a predetermned frng temperature, ndependent of varaton n ambent temperature or fuel characterstcs. Acceleraton and deceleraton fuel schedules are basc control requrements for mantanng the gasturbne engne to operate wthn ts safe margn durng steady state or transent condtons. Acceleraton control s used prmarly gas turbne startup to lmt the rate of rotor acceleraton pror to reachng governor speed, thus ameloratng the thermal stresses encountered durng startup. Ths control serves a secondary functon durng normal operaton, n that t acts to reduce fuel flow and lmt the tendency to overspeed n the event that the turbne generator separates from the system [8]. These three control functons, speed governng under part load condtons, temperature control actng as an upper lmt, and acceleraton control to prevent overspeedng, are all nputs to a low value selector. The

3 output of the low value selector, whch s called VCE, s the lowest of the three nputs, whchever requres the least fuel. Transfer from one control to another s bumpless and wthout any tme lags. The output of the low value selector s compared wth maxmum and mnmum lmts. The mnmum lmt s the more mportant. Ths s because the mnmum lmt s chosen to mantan adequate fuel flow to nsure that flame s mantaned wth the gas turbne combuston system [8]. Gas turbne fuel systems are desgned to provde energy nput to the gas turbne n proportonal to the product of the command sgnal (VCE) tmes the unt speed. Ths s analogous to the actual mode of operaton of the fuel system. snce lqud fuel pumps are drven at a speed proportonal to turbne rotor speed. Gas fuel control s accomplshed n two stages wth the output pressure of the frst stage beng proportonal to rotor speed [~8]. respect to fuel flow and turbne speed over the 9~7% desgn ratng. The exhaust temperature equatons are somewhat less accurate at part load; however, snce temperature control s only actve at the desgn pont, the mpact of the part load naccuracy s neglgble to the overall smulaton. An deal acceleraton control allows the engne to accelerate at a reasonably fast rate wthout the engne beng drven nto the surge regon or overheatng, the engne components. Followng load rejecton, a rapd reducton of fuel flow s requred to lmt the maxmum speed rse. There s a mnmum level to whch the fuel flow can be reduced wthout causng a flameout problem. A deceleraton fuel schedule s requred to mnmze the turbne speed rse on load rejecton, but not the flamngout of the engne [~8]. Startup Control Speed Control Temp Control LVG Fuel flow VCE G( G( G( G( VCE=fuel flow sgnal, G(= transfer functon of fuel system, G(=transfer functon of between fuel flow and turbne speed, G(=transfer functon of between fuel flow and turbne exhaust gas temperature, G(=transfer functon of between fuel flow and generator output Fg. Control system of Gunsan gasturbne generatng system Speed Temp Load The fuel gas control system conssts of two valves n seres, the frst of whch controls the pressure between the two valves as a functon of speed. The second valve has a lnear characterstcs versus lft range and s aerodynamcally desgned so that sonc veloctes are attaned at the controllng area wth flangetoflange valve pressure ratos as low as.. If valve poston s mantaned proportonal to the VCE sgnal, the actual result s a flow rate of fuel gas, whch s proportonal to the produce of gas turbne speed [7]. Both the torque and exhaust temperature characterstcs of the sngleshaft gas turbne are essentally lnear wth. PID controller n gas turbne The threemode PID controller s wdely used n plants due to ease of control algorthms and tunng n the face of plant uncertantes. However, the response of those depends on the gan P, I, and D [7]. Nevertheless, the lnear PID algorthm mght be dffcult to deal wth processes or plant wth complex dynamcs, such as those wth large dead tme, nverse response and hghly nonlnear characterstcs. Up to date, many sophstcated tunng algorthms have been used to mprove the PID controller work under such dffcult condtons, snce the control performance of the system depends on the parameter gans. However, most control engneers can tune manually PID gans by tral and error procedures n many cases. So, PID gans are very dffcult to tune manually wthout control desgn experence [~8].. A feedforward DOF controller for gas turbne. Fller parameter separated type DOF PID controller The parameter of target flter s desgned nto separate two part α, The transfer functon of between process value PV( and settlng value SV(, process value PV( and dsturbance D( s gven as the followng equatons: G ( = PVSV = G DVSV ( = PV ( SV ( K p ( / T α{ a /( T s b)} ( K p s / T ) Td sg( DV ( SV ( ()

4 = G( ( K p s / T ) Td sg( In equaton (), the process value PV( depends on the two degree parameter flter parameter, a, and b.. But numerator s the same as that of conventonal PID controller T f () ths paper, we used N (Neural Fuzzy System) for tunng [, 9, ]. We used NeuralFuzzy based on the archtectures and learnng rules of adaptve networks have been studed n reference []. In N, fthen rules structure for tunng s appled by a frstorder Sugeno fuzzy model as the follows: SV Flter α D γ PI β PV Rule : If x s Al and U s Bl, then fl=plxqyrl, Rule : If x s A and w s Bl, then f=pqyr. Every node, that s, Layer Layer of Fg. have meanng as descrbed n reference []. Fg. s an N archtecture that s functonally equvalent to a threenput frst order Sugeno fuzzy model wth seven rules, where each nput s assumed to have three assocated MFs..8z. z.7z. l Here, we denote the output of the th node n layer l as L, and each node can be descrbed as the followng: R(k) Inlet gude vane sgnal.7z.7 z.7z. Fg. DOF PID controller wth N for Gunsan gas turbne DOF PID N M(k) plant Fg. The structure of a DOF PID controller for Gunsan gas turbne The proportonal gan s also affected by the flter parameter a and two degree parameter. The dsturbance s controlled by gans K p, T, Td. The structure of new DOF PID controller tunng by ANFIS suggested n ths paper s represented as Fg. and Fg. [].. Tunng of DOF PID controller by N Up to now, ultmate method, Z&N method has been used for tunng of DOF PID controller. Instead of that n D(k) C(k) Fg. The structure of N for a new DOF PID controller of Gunsan gas turbne system ) Layer All nodes n ths layer are adaptve nodes wth a node functon L, = h (x) A, for =,, () where x s the nput of process (gas turbne) to node and A s a lngustc label such as, BIG or ZERO assocated wth ths node to process. That s, L, s the membershp grade of a fuzzy set A ( A, A, B, B ). Here the membershp functon for A s gven as the followng bell functon: ha( x) b x c a = () where each parameter a, b, c decdes the shape of bell to exhbt varous forms of membershp functons for fuzzy set A. Parameters n ths layer are referred to as premse parameters as reference [] ) Layer

5 The functon of every node s a fxed node and output s the product of all the ncomng sgnals: L, = w = ha ( x) hb ( y), =,. () Fg. The structure of N for DOF PID controller of Gunsan gas ) Layer Nodes n ths layer are fxed node and the th node n ths layer calculates the rato of the th rule s frng strength to the sum all rules frng strengths: w L, = w =, =,. (7) w w ) Layer Every node of ths layer has an adaptve behavor wth L = w f = w ( p x q y r ), (8) where A A B, p q r ) ( B C C w s a normalzed frng strength from layer and s the parameter set of ths node. ) Layers As a fxed node, whch computes the overall output as the summaton of all ncomng process sgnals: Overall output = L, w f = w f = w (9) We can make an adaptve rules to process because the assgnment of node functons and the network confguraton are arbtrary, as long as each layer perform meanngful and modular functonaltes [].. Smulaton and dscuss. PID and conventonal DOF PID controller n Gunsan gas turbne system To compare the characterstcs of these controllers such as, PID, the conventonal DOF PID, new DOF PID, we adapted these controller to transfer functon based on P I D the operatng data of Gunsan turbne. Fg. 78 llustrate the response n case of applcaton PID controller to gas turbne control system. In Fg. 7, n case of fuel loop feedback control s havng a stable shape aganst 's dsturbance and stable but n case of temperature feedback loop, do not follow well n runnng part. So, f we use PID controller n startup and runnng of gas turbne, we fgure out that we can not control satsfactorly by PID controller. Fg. 8 s the response of applyng PID controller n case of P=., I=., D=. to a temperature loop of gas turbne control system. Fg. 8 s response of total loop (fuel flow sgnal, gas temperature sgnal, and gude vane sgnal) n case of P=, I=., D=. In Fg., s havng a stable shape aganst 's dsturbance but do not follow well n runnng part. So, f we use PID controller n startup and runnng of gas turbne, we fgure out that we can not control satsfactorly Fg. 9 show the results of the conventonal DOF PID controller n Gunsan gas turbne. In Fg. 9, between flow sgnal and flow rate s not matchng exactly from seconds pont. Fg. 9 represents the response of total loop feedback of DOF PID controller. There are many dsturbances n flow sgnal but sgnal s very smooth. In Fg. 9, s not controlled by.. A new DOF PID controller tunng by ANF n Gunsan gas turbne Fg. llustrate the response n case of applcaton a new DOF PID controller to gas turbne control system. From fgure, even f parameter values P, I, D,,,, change, the response of fuel flow s very stable n spte of the dsturbance of fuel sgnal. So, we can see a new DOF PID controller that we proposed n ths paper s havng a good characterstc aganst the PID and the conventonal DOF PID controller.. Concluson Up to date, the PID controller has been used to operate under such dffcult condtons n ths system, but snce the gan of PID controller manually has to be tuned by tral and error procedures. Gettng an optmal PID gans s very dffcult to tune manually wthout control desgn experence. In ths paper a new DOF PID controller method has been suggested usng N tunng. Tunng method used n ths system s N and ths

6 method s a good response wthout pror knowledge of the process. Also, Resultng by ths method s more good response than the conventonal PID or DOF PID controller. Ths control method s very useful to apply process control system. P=. I =. D=. Alpha=. Beta = Gamma= DOF PID References. H. Cohen, Gas turbne theory, Catalogngnpublshng data, PP. 9~9, 99.., Gas turbne, Corona publshng Co., 98.. W. W. Hung, Dynamc smulaton of gasturbne generatng unt, IEE processng, vol. 8, no., ~, 99.. DOF PID controller, SICE, vol., no. 9, PP. 889~89, MARK I Card applcaton data, General electrc, Gas turbne operatng manual, KEPCO., K. S. Ahluwala, Dynamc modelng of a combned cycle plant, Tarns. Of the ASME, Vol., Aprl, PP. ~7, Turbomachnery developments n steam and gas turbnes, ASME, Km, Dong Hwa, A applcaton of multvarable DOF PID controller wth Neural network Tunng method to the Heat exchange, FUZZIEEE99, Oct. ~, 99, Seoul.. J. S. R. Chang, NeuroFuzzy soft computng, Prentce Hall, P =. II =. D=. Alpha= Beta= gamma= DOF PID 8 P=. I =. D=. Alpha=.7 Beta= Gamma=. Flter / S New DOF 8 P=. I =. D= IGVO 8 GT Fg. A varyng characterstcs of each parameter Startup process of Gunsan gas turbne. New DOF & New DOF Cascade P =. DOF PID II =. D=. (Cascade) (Cascade) Alpha= Beta= gamma= 8 8 Fg. 7 A varyng characterstcs of each parameter n startup process of PID controller.