Optimization and Simulation of Boiler Water Level Control Based on the Fuzzy Neural Network

Transcription

1 2nd Internatonal Conference on Automaton, Mechancal and Electrcal Engneerng (AMEE 27) ptmzaton and Smulaton of Boler Water Level Control Based on the Fuzzy Neural Network Zhmn Yu*, Long Cheng2, Zhenyong Hao3 and Renzhong Wang4 Tann martme college. 335.Chna Tann Martme Safety Admnstraton.335.Chna 3 Chna lfeld Servces Lmted 335.Chna 4 School of ocean Yan ta Unversty.264.Chna *Correspondng author 2 Abstract [bectve] To adust the parameter PID of boler water level control.analyss of the problems estng n the Marne boler water level control.[method] Usng fuzzy neural network, Fuzzy neural network s the combnaton of neural network and fuzzy logc. the tradtonal PID control and fuzzy neural network control modelng, smulaton comparson.[result] The results show that the fuzzy neural network control convergence s good, the method s effectve.[concluson] The concluson s of reference sgnfcance to mprove boler water level control. Keywords-boler water level control; fuzzy logc;fuzzy neural network g; PID I. INTRDUCTIN The boler can be used to produce steam heatng shp fuel ol system. Some shp can use steam to drve a turbne power generaton, can use steam to drve a tanker cargo ol pump and ballast pump. Boler s an mportant equpment of shp. The generaton of steam to automatcally adust the water level, water level of boler automatc control s to mantan the normal level. It s through the control of feed water pump startup (feed valve openng) to control the boler feed water, feed water and evaporaton balance to ensure the normal operaton of boler. If the steam boler water level s too hgh humdty.mpact drve equpment.the evaporaton capacty of the boler equpment and perpheral equpment load demand at any tme, when steam need large, steam drum water level changes quckly, f the crew do not pay attenton to. It wll be easy to burn caused by false level even dry pot, or drum flled wth water. Water level drectly affects the separaton speed and the qualty of the producton of steam, boler water level s one of the mportant parameters to ensure the shp's safe. At present, the boler water level control often use PID controller. The PID parameter proporton wth ntegral and dfferental tme settng affect the level of control performance.parameter settng for the crew s dffcult.proportonal and ntegral dfferental tme adust wll take great effort, adustment effect s not deal. Therefore, the best boler water level control can realze ntellgent, PID parameters accordng to the outsde changes and automatc settng. Boler feed water system of boler feed pump, the pump pressure s constant, the water level sensor sensng water level change sgnal to the controller. The controller s computng. utput sgnal control the feed valve openng. By changng the feed valve regulate the boler water level. II. PID FUZZY CNTRLLER B ASED N NEURAL NETWRK Fuzzy controller s the bggest advantage s not dependent on the controlled obect accurate mathematcal model, and prone to prevent oscllaton stablty wth super se, control attenuaton and control overshoot.the fuzzy controller s suted to solve the problems ofthe marne boler water level control.in the dgtal control system, PID control algorthm s usually wrtten n ncremental form, mathematcal formula such as () u(k) Kp[e(k) e(k )] Ke(k) Kd[e(k) 2e(k ) e(k 2)]() Formula (), Kp s proportonal coeffcent, K s ntegral coeffcent, Kd s dfferental coeffcent. In tradtonal PID control, n order to get good control effect. Adust the approprate proporton of Kp,K and then to elmnate statc error ntegral tme must be approprate. Small strong dynamc process caused by vbraton of the system s not stable. The approprate dfferental tme Kd s consderng water mutaton regulaton n advance, Kp,K,Kd, and the sze of the three nterrelated, nteracton and regulaton methods eperenced method.ths method requres adustng personnel have rch practcal eperence, have attenuaton curve method s to utlze dampng ratos to control dynamc process.in partcular the boler feed water system certanly est the optmal Kp,K,Kd values, because wth a partcular learnng algorthm of neural network can take advantage of the neural network to fnd some knd of nonlnear combnaton of the optmal control law. To mplement the boler water level control tasks.marne boler water level control of the neural network fuzzy control system s controlled by the nput lnk.boler water level and the operaton ncludng tradtonal PID controller of fuzzy quanttatve lnk.the outsde changes smulated learnng algorthm and water level on the fuzzy quanttatve.the actual water level change learnng algorthm to gan recognton network NN,NN2 provde the necessary nformaton. Identfy network manly n order to establsh the dentfcaton model of boler water level control system, a neural network on-lne modfcaton. The controlled formaton of the closed loop control system. Fuzzy quanttatve module, equvalent to NN2 preprocessng of the neural network nput.the network NN2 output characterstc of the boler control varable Kp (rato), K Copyrght 27, the Authors. Publshed by Atlants Press. Ths s an open access artcle under the CC BY-NC lcense ( 47

2 (ntegral coeffcent), Kd, (dfferental coeffcent). as shown n fgure of fuzzy control system based on neural network structure. Implementaton method of PID algorthm. uses the fve-step method specfc as follows. Step: y (k) s obtaned by samplng and r (k), and then calculate e(k) = r(k) - y(k) Step 2: After the normalzaton processng wll e (k) fuzzy quantfcaton.accordng to the actual water boler and boler water level devaton gven to the sze.. y ( I ) ( k ) y (k ) u (k ((ni ) m ) ( k ) m) n n n m (5) Formula as an obect of the neural network nput layer n (5)after a delay of output y (k - I) and u (k - I).The tme sequence {y (k)} and {u (k)} as the model feature The hdden layer neural network such as formula(6) y n m I ( h ) V ( I ),,,R Algor {E } Fuzzy K e Kd boler PID y yˆ ( k ) such as formula (2), namely devaton wth the gven value than the absolute value of less than.3 and greater than.3 s less than., s greater than. s less than.3, s greater than.3 s less than.5, s greater than.5 shown n fve fles. Archve process s devaton wth the gven value than the scope of the absolute value of also to fuzzy quanttatve e (k). E.5 3sgn ( e ).3 2 sgn ( e ). sgn ( e ).3 sgn ( e ).3 sgn ( ) (2) f [ y (k ),, y(k (6) W (k ) (7) JI [ y (k ) yˆ ( k )] 2 [ y(k ) yˆ (k )] ( h ) (k ) W (k ) [y(k ) yˆ(k )]f (I (k))w (k)(i ) (k) (8) V (k) (9) In order to accelerate the convergence study, take the mamum effcency. In order to avod the oscllaton of the learnng can be adusted by the gradent of nerta coeffcent, the formula (8), (9) the value of alpha, beta, n the selecton on [,], alpha vector s, rate of beta s nerta. Step 4: The establshment of the neural network NN2, usable G (* * *) to descrbe the output of the PID, such as the formula () (3) Step 3: Identfy network NN created.set accordng to the actual of boler water level and water level s (4) output nonlnear functon and learnng algorthm such as formula. y (k ) 2 p mnmum.avalable functon for correspondng correcton formula :(8) and (9) V (k ) {W} n the formula.7 s output layer weght functon, weght functon s a lnear functon, through the neural network learnng algorthm to modfy the (n + m) (threshold).the W (k ) Formula (2) sgn (*) symbol functon, functon are the three condtons such as type of (3) ( k )] FIGURE I. THE FUZZY CNTRL SYSTEM BASED N NEURAL NETWRK STRUCTURE f [I R R. 4 sgn ( e ) (k ) {V} n equaton (6) s hdden layer weghts; V (n + m) s a threshold value; F () s the actvaton functons.superscrpt (I) s nput layer, superscrpt, respectvely s the hdden layer. The neural network output layer such as (7) N NN2 K r Algor (h ) n), u (k ), u(k m)] (4) the formula(4) (f * * * *) s a nonlnear functon, y s the nput sgnal, u s the output sgnal, n s the order tme {y (k)}, m s the order of {u (k)}. Neural network layer such as formula such as(5) u (k ) G[u ( k ), K P, K, K d, e( k ), e( k ), e( k 2)] () In the formula () G (*) s a nonlnear functon, and the nonlnear functon, y(k).usng neural network NN2 through tranng and learnng to fnd an optmal control. The nput of the network NN2 s after dealng wth the blur of system state varables, and neural network NN2 nput and output of each layer neurons usng three layer BP network as shown n fgure 2.The network has the nput node S, S for the number of nput varable, dependng on the complety of the controlled obect.the nput node to smulate the shakng, work status and shp consume a large amount of steam and produce the swtch state suchas soda bolng boler. Hdden layer nodes for H, H s 3 the three output node for the proporton of the PID controller 48

3 coeffcent dfferental coeffcent and ntegral coeffcent.the hyperbolc tangent functon as formula () KP KI KD Wk V {E} FIGURE II. NN2 NEURAL NETWRK STRUCTURE S I (2) (k) V() (k) (2) (2) (k) f [I (k)],,, H H(2) () In the formula (), f(***)as the hyperbolc tangent type actvaton functon. Superscrpt () s the nput layer.superscrpt (2) s hdden layer.superscrpt (3) the output layer, {V} s hdden layer weghts, Vs s the threshold. Step 5: The output of the tradtonal PID controller to adust the openng of the boler feed water valve,at the same tme sgnal nput dentfcaton network NN, classcs, cycle study and calculaton.the fnal output optmal Kp (rato), K (ntegral coeffcent), Kd, (dfferental coeffcent). As shown Fgure3 Neural network desgn parameter smulnk settngs, and Fgure4 Neural network smulnk desgn program. FIGURE III. NEURAL NETWRK SIMULINK DESIGN PARAMETER SETTINGS neural network fuzzy PID control system based on v fles as shown n fgure.3.valdaton s more superor to tradtonal PID control.fuzzy control accordng to water level fuzzy control prncple dagram, detaled smulaton desgn the followng steps: Frst step :Smulnk boler feed water system, through the step nput on behalf of the boler water level change.the actual devaton quanttatve factor s.6.the dfferental quantzaton factor s.3 quanttatve factor s derved from the neural network output.after the operaton to the nput of the fuzzy controller.the fuzzy controller output, consderng the outsde change mpact factor s.2, as shown n fgure 4.The actual water level of the transfer functon can be epressed as G (s) = H (s)/w (s) = K/(Ts2 + s) [].The formula of the H for the current water level, W for the current water flow, for feed water flow under the dsturbance of the coeffcent, K T boler feed water system for nertal tme constant. K values range from.2to.6.database, and provde data for reasonng machne. Rule base s used to store the fuzzy control method and the reasonng of the reasonng machne.provdes control law s based on the shpyard debuggng epert knowledge and shppng lne chef engneer and engneers such as long-term accumulaton of eperence.the advantages of the reasonng machne language representaton s a knd of ntutve reasonng. Boler water level fuzzy controller, the system nput varables to set the water level and the actual water level devaton.the output of the openng of the regulatng valve control sgnal, correspondng lngustc varables respectvely e, EC and u, fuzzy rules can be smple language, for eample (f e () set the water level and the actual water level devaton s bg then u (output of the openng of the regulatng valve control sgnal) s negatve).meanng the rules s hgh water levels rse faster, level control valve sgnal that rapdly fallng water levels). The fuzzy nference machne s based on the nput amount and completed database, rule base, fuzzy nference, and solve the fuzzy relaton equatons such as (2) U (E EC ) R (2) In the formula(2), symbol "."on behalf of the quantty of fuzzy synthetc operaton. Symbol "" on behalf of the drect product of fuzzy quantty calculatons; "R" represent the fuzzy relatonshp between symbols. FIGURE IV. NEURAL NETWRK SIMULINK PRGRAM III. SIMULATIN IMPLEMENTATIN F F UZZY CNTRLLER Smulaton mplementaton, based on the fuzzy controller s combned wth the cycle of neural network learnng features.automatc ntellgent adustment Kp (rato), K ( ntegral coeffcent), Kd, (dfferental coeffcent). Through the smulaton model, Use Smulnk smulaton and the desgn of the fuzzy controller, called the desgned fuzzy controller fle gshu. Fs.The desgn of the For accurate control of boler water level fuzzy quantty must be converted nto precse volume.it can only accept a control volume of the obect, and ther nterfaces are the fuzzy decson by fuzzy nference machne output nto a precse amount. Usng Matlab fuzzy logc edtor, fuzzy logc edtor s a graphcal system desgn tool.matlab fuzzy control toolbo s the desgn of fuzzy controller s very convenent, only need to set the correspondng parameter.quckly get fuzzy control, convenent and modfy the parameters. The fuzzy controller usng Matlab toolbo n Matlab command wndow, enter the fuzzy, clck enter the keyboard as shown n fgure 5 49

4 (a) (b) FIGURE VII. (A) BASED N THE DYNAMIC RESPNSE F THE NEURAL NETWRK FUZZY PID ADAPTIVE CNTRL ;(B)THE DYNAMIC RESPNSE F THE TRADITINAL PID CNTRL FIGURE V. THE ESTABLISHMENT F THE FUZZY CNTRLLER Determne the descrpton of language nput and output varables of the fuzzy subset of PB (board) - take + 6 nearby.pm (mddle) - take near + 4.PS s (small) - take near + 2. (zero) - take near zero. NS (negatve) - take a near 2. NM (negatve) - take a near 4.NB (negatve) - take near 6.Add the membershp functon E theory of doman n the range of [3 of 3]. So you need to add 7 membershp functon, and set up the theory of nput and output varables of a doman. E settng error, error of EC change, U control doman.select the correspondng membershp functon fuzzy lngustc varables. Usng the toolbo Member Functon Edt, open Member Functon Edt wndow. As shown n fgure 6, accordng to the fuzzy control rules establsh logcal relatonshp wll buld good FIS edtor s saved as a fle.(gshu.fs.) Boler feed water system smulaton model based on fuzzy algorthm and the tradtonal PID controller and fuzzy control system smulaton. Boler water levelsystem step response output n fgure.6 and fgure.7.we can drve the PID dynamc devaton, oscllaton frequency, low boler water level control precson, poor ablty to adapt to changes n the outsde world. Improvng desgn of fuzzy PID neural network greatly reduces the boler water level dynamc devaton, effectvely restran oscllaton n the process of boler control,avod the boler water level adustment n advance, for boler water level control s more precse, the statc devaton decreases. Improve the boler water level accordng to the change of adaptve performance s greatly mproved.n short.the neural network fuzzy PID easy stablty wth super prevent oscllaton attenuaton, control, control overshoot. IV. CNCLUSIN Usng Matlab toolbo Member Functon fuzzy controller desgn. Edt realze smulaton of boler feed water control based on fuzzy controller. Neural network output optmal Kp (rato), K (ntegral coeffcent), Kd, (dfferental coeffcent).the ntellgence of parameter modfcaton, n the smulaton can modfy the parameters of the obect. It can modfy the membershp functons and control rules and quanttatve theory doman, nput and output of language varable. FIGURE VI. FUZZY CNTRLLER LGIC RELATIN. The thrd step n Smulnk, the fuzzy control module to nvoke the desgned fuzzy controller fle gshu. Fs, and before the smulaton gshu. Fs read n work space, the smulaton result s when the control obect nput and system step response curve. Based on the dynamc response of the neural network fuzzy PID adaptve control s shown n fgure 7(a). The dynamc response of the tradtonal PID control as shown n fgure 7(b) Boler water supply system based on fuzzy algorthm s compared wth the tradtonal PID control smulaton analyss. Fuzzy algorthm of boler feed water system stablty wth super more easly prevent oscllaton attenuaton, control, control overshoot. From the pont of the smulaton results. The boler feed water system based on fuzzy algorthm s better than tradtonal control performance of PID control. The boler feed water system control based on fuzzy algorthm has very good ablty to adapt to envronmental changes and self-learnng ablty of automatc ntellgent modfyng PID proporton, ntegral and dfferental tme.when a shp s the outsde changes, has the very good control performance. 5

5 REFERENCES [] Guo Lne. Large ol tankers for aulary boler system modelng and smulaton study [D]; Dalan martme unversty; 23, [2] Xao-dan lu. Based on RBF fuzzy neural network of Marne boler steam drum water level control research [D]. Dalan martme unversty; 28, [3] L Sun. No meda research and realzaton of the hardware-n-the-loop smulaton system [D]. Tann unversty of scence and technology; 29 [4] FeJngZhou.Based on generalzed predctve control of Marne boler steam drum water level control research [D]. Harbn engneerng unversty; 28 [5] Sun Bng. Smulaton of a Marne aulary boler fuel ol system based on VB [D]. Dalan martme unversty; 28 [6] Qu Dabao. Large ol tanker aulary boler system smulator development [D]. Dalan martme unversty; 22 [7] WANG Yang.Mult-model Predctve Control of Drum Water Level Based on RBF Neural Network Dynamc Compensaton [J]. ShpScenceandTechnology, 27, (2): [8] YU Zhmn.ptmal and Control Smulaton of Shp Boler Based on Fuzzy Neural Network [J]. Journal of Tann Vocatonal Insttutes, 27, (4): [9] CHEN Wen-ng, XUE Sh-long, SUN Le, ZHANG Ya-mng.Modelng and montorng of water level control system based on PLC boler boler [J]. Journal of Shangha Dan Unversty, 27, (2): 98-2 [] WANG Zha-a, WANG Ln-n, FANG Yong. Detecton and control of lqud level n drum boler [J]. PetroChna, 27, (5): [] ZHNG Q-sheng.NAKAKITA NS778AN Type Water Level Automatc Control Falure Falure and Its Elmnaton Method [J]. PearlRverWaterTransport, 27, ():