Self-tuning PID-type Fuzzy Adaptive Control for CRAC in Datacenters

Transcription

1 Self-tuning PID-type Fuzzy Adptive Control for CRAC in Dtcenters Junwen Deng 1, Liu Yng 1, Xinrong Cheng 2, nd Wu Liu 1 1 College of Engineering, Chin Agriculturl University, Beijing , Chin 2 College of Infortion nd Electricl Engineering, Chin Agriculturl University, Beijing , Chin { , }@qq.co, yngliu@cu.edu.cn Abstrct. In order to eliinte the current negtive condition of Autotic Coputer Roo Air-Conditioning (CRAC) syste, self-tuning Fuzzy Logic Control (FLC) ws designed nd pplied to fn speed in CRAC syste. In this pper, we derive therodynic odel of dtcenter suitble for pplying dptive self-tuning PID-type fuzzy dptive control theory. It cobines the clssic PID control strtegy nd fuzzy dptive control theory. The clssic PID control uses the error nd rte of chnge of error s its inputs to control the teperture utoticlly, nd the fuzzy logic controller is used in the selftuning PID-type fuzzy control to tune the preters of PID controller on-line by fuzzy control rules. Siultion nd testing results show tht the proposed self-tuning FLC ethod cn chieve less stedy-stte error nd short settling tie in teperture control of dtcenter. Keywords: Air-Conditioning (CRAC) syste, self-tuning Fuzzy control, dptive control. 1 Introduction Significnt teperture pressure hs been plced on trditionl dtcenters therl ngeent by incresing server power dissiption. Autotic Coputer Roo Air- Conditioning (CRAC) syste iproveent chllenges hve been stedily incresing over the pst few yers due to n increse in energy consuption nd het genertion in dtcenters. CRAC fn speeds hve significnt effect on the recircultion flows in the dtcenter, which re verified to be jor cuse of inefficiency [1-3]. The energy needed for djusting fn speed is uch lower thn the energy needed for djusting the duty cycle of ir copressor. For this reson, flexible design of CRAC fn speed optiiztion syste supplying dend is substntilly to reduce the energy consuption. So fr, clssic control techniques, such s Proportionl-Integrl- Derivtion (PID) controllers re still widely used in prctice s they cn be esily ipleented, re low cost nd relible in hrsh field condition. K p, K i nd K d preters of PID were obtined to iniize the syste error in ny studies. D. Li nd Y. Chen (Eds.): CCTA 2013, Prt I, IFIP AICT 419, pp , IFIP Interntionl Federtion for Infortion Processing 2014

2 216 J. Deng et l. Recently, soe coplex control strtegies bsed on clssicl control concepts hve been proposed in ttepts to iprove the syste perfornces. Slsbury described feed forwrd control schee bsed on siplified physicl odel s suppleent of the conventionl PI feedbck control [4]. Kshr et l. proposed robust PID control schee to del with the odel uncertinty cused by the chnging chrcteristics of the plnt [5]. Bi nd Ci et l. developed n dvnced uto-tuning PID controller for both teperture nd pressure control [6]. Clssic control schees coonly use the first order or second order plus tie dely odels to represent process dynics. The perfornces of these control schees re liited when pplied to CRAC process due to their inherent nonlinerity nd tie vrying nture. To overcoe these drwbcks, pplictions of intelligent control to CRAC syste hve drwn soe interests. Soe uthors proposed fuzzy PID ethods to iprove perfornce of the control syste. In Shyeghi H, Shynfr HA, nd Jlili s pervious ppers [7] nd [8], the uthors proposed n iproved control strtegy bsed on fuzzy theory nd genetic lgorith (GA) technique hve been proposed. Fuzzy logic (FLC) of heting, ventilting nd irconditioning syste ws studied by Shn AJ, Krishnn N nd Schi RR [7, 9]. The obtined results were copring with those of PID control nd these studies indicted tht FLC hd better result. However, the stedy-stte error ws not totlly eliinted. Existing work on therl ngeent in dt centers, however, not tke into ccount unexpected therl nolies nd neglect the cost of control ctions tken by the ir conditioning syste. This cn led to overheting of soe spots. Obtining better perfornces of the siulted systes bece esier [10]. To eliinte this negtive condition, self-tuning FLC ws designed nd pplied to fn speed in CRAC syste. 2 Relted Work In this pper, we derive therodynic odel of dt center suitble for pplying dptive self-tuning PID-type fuzzy dptive control theory. We propose novel control strtegy bsed on this odel. It cobines the clssic PID control strtegy nd fuzzy dptive control theory. The clssic PID control uses the error nd rte of chnge of error s its inputs to control the teperture utoticlly nd the fuzzy logic controller is used in the self-tuning PID-type fuzzy controller to tune the preters of PID controller on-line by fuzzy control rules. It ws shown tht the usge of odeling nd siultion ethods for nlyzing, testing nd developing of CRAC systes decreses the design cost s well s the durtion of the design process. Siultion nd testing results show tht the proposed self-tuning FLC ethod cn chieve good perfornce in teperture control of dtcenter.

3 Self-tuning PID-type Fuzzy Adptive Control for CRAC in Dtcenters The Therl Model of the Dtcenter Fig.1 shows the lyout of dtcenter nd loction of blde. Dtcenter design plys n iportnt role in the efficient therl ngeent. Chilled wter CRAC units supply rised floor plenu underneth the rcks with cold ir. Perforted tiles re locted ner the rcks to trnsfer the cool supply ir to the front of the rcks. Fig. 1. The Lyout of dtcenter nd loction of blde As the CRAC syste hs fetured such s considerble interference, non-linerity, lrge tie dely, uncertinty nd non-predictive, it is difficult to obtin the ccurte theticl odel of therodynic in dtcenter by considering ll preters. For this reson, we consider the following ssuptions: According to the lw of the energy conservtion, the theticl eqution for dtcenter therl is s follows: ρ q (1) d c v = cρ f( td t) As μs( ts td ) + qn + q = c = μ A ( t td ) (2) As ost equipent in dtcenter re de of irons, they hve lrge coefficient of het conductive, so we ssue teperture vriety rte hs not uch difference between equipent teperture vriety rte. d = d n (3) If the fn otor het dissiption is ignored, the eqution (1) cn be trnsferred to trnsfer function for s follows: c v ρ s c n s c ρ f t A d = c ρ f t A t s s d If c is defined s CRAC hybrid equivlent therl cpcity s follows: c = c n c v ρ (5) s t (4)

4 218 J. Deng et l. And R is defined s CRAC entire therl resistnce nd R s is defined s Surrounding wlls, floor nd roof therl resistnce s follows: After stndrd rrngeent the result is s follows: 1 RS = (6) A μ S S k ( s) Ksts ( s) t ( s) = + (7) d Ts + 1 Ts + 1 The tie constnt nd other preters re s follows: T = C * R (8) K d = K * R (9) R K s = (10) Rs As the ir het trnsfer dely ffects the dtcenter teperture, consider dely tie is, the equivlent trnsfer to be s follows: k t ( s) Ts + 1 K t ( s) Ts + 1 d τs s s τs d ( s) = e + e (11) t Setting the rel-tie teperture in the dtcenter is 27, nd the expected teperture in the dtcenter is 20.The ir flux of the dtcenter ws deterined by the fn speed t end, Fn speed control is the core technology in Vrible Air Volue (VAV) ir condition syste. It decides the libility nd energy sving level of the CRAC syste in the dtbse.the fn speed is controlled by the open volue of the ir condition. The trnsfer function of open volue to teperture cn be expressed s n object of second-order inertil nd net dely: 12s 10e G( S) = (1 + 20s)(1 + 30s) (12) The dtbse CRAR syste Mthetic Model is showed in figure 2. In the odel, U(t) is the output of the controller nd y(t) is the ctul teperture of the dtbse roo. Fig. 2. CRAR Syste Mthetic Model

5 Self-tuning PID-type Fuzzy Adptive Control for CRAC in Dtcenters 219 Tble 1. Nottion 4 Self-tuning PID-type Adptive Control The control syste is optiized through the whole working rnge of the process nd ensures xiu of control loop qulity by very short response tie of ny lterntion in the process nd t negligible overshooting of the process vlue during the control phse. The fuzzy controller showed enorous dvntges in processed with intensive nonlinerity. An increentl fuzzy logic controller is used in the self- tuning PID-type fuzzy controller to tune the preters of PID controller on-line by fuzzy control rules. The controller uses the error nd rte of chnge of error s its inputs nd cn eet the desire of self-tuning preters bsed on tie-vrying e nd e, nd the output of the fuzzy controller is u. u( k) = K k e( k) + k e( i) + k [ e( k) e( k 1)] p i d (14)

6 220 J. Deng et l. The dtbse CRAR syste Mthetic Model is showed in Fig.2. In the odel, U(t) is the output of the controller nd y(t) is the ctul teperture of the dtbse roo. Where k p is the controller gin. k pt K d = T i (15) T is the sple tie; Ti is the integrl tie preter, T d is the derivtive tie preter [11]. The sub-block of lnguge ws given in the following nner: Positive Big (PB), Positive Middle (PM), Positive Sll (PS), Zero (ZO), Negtive Sll (NS), Negtive Middle (NM), Negtive Big (NB).The output of the fuzzy systes is u. The fuzzy vribles re defined for the rule bse s: (e,e ) = {the error, the vrition of error}, {NB, NS, Z, PS, PB}. k p, k d, k i ={the control preters, Z, PS,PM, PB, PVB (Positive Very Big), [0,1], l}. Fig.3 shows the rule bse fuzzy PID-type control nd the rule Ki fuzzy. Fig. 3. () Shows the rule bse fuzzy (b)shows the rule Ki fuzzy A ebership function is curve tht defines how ech point in the input spce is pped to ebership vlue (on degree of ebership) between 0 nd 1. In our study we will eploy the trpezoidl ebership functions for ech fuzzy linguistic vlue of error e nd the chnge rte of error e showed in Fig.4. By pplying the self-tuning PID-type fuzzy dptive controller, the control of the se CRAC syste hs been relized in order to copre of the control perfornce. The clssicl PID nd FLC theories re cobined in this study. The K p, K i, K d vlues of PID preters hve dptively been deterined by using the dynic FLC for ech tie-step. In this cse, FLC hs two inputs (e, e ) nd three outputs (K p, K i, K d ). The physicl doin of the inputs (e, e ) is -1, -0.8, -0.6, -0.4, -0.2, 0, 0.2, 0.4, 0.6, 0.8,1 nd tht of the outputs (K p, K i, K d ) is 0, 0.1, 0.2,0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, selected gin bsed on tril nd error. Fig.5 shows the ebership functions of output u.

7 Self-tuning PID-type Fuzzy Adptive Control for CRAC in Dtcenters 221 Fig. 4. The ebership functions for the input e nd e Fig. 5. The ebership functions for the output u 5 The Nuericl Siultion of the CRAC Syste By using MATLAB/SIMULINK pckge progr, the control, odel nd nuericl siultion of CRAC fn syste hve been relized. The control digr of CRAC fn syste hs been obtined by considering the input nd output vlues of the device nd the desired dtcenter conditions. The obtined results hve been presented in grphicl for. Every PID control syste hs its preter rnge. It will result in unrelibility nd out of control if the preters re out of the rnge. In the CRAC control syste, the PID preters re tuned by fuzzy inference. According to siultion, the tuning rnge of preter K p is between [25.4, 40.5], the tuning rnge of preter K i is between [0, 0.5] nd the tuning grnge of preter K d is between [15.2, 20.6].If the syste is overshoot, first decrese the preter K p, then decrese the preter K i nd in the end decrese the preter K d. If the syste hs long settling tie, first decrese the preter K i, then decrese the preter K p nd in the end decrese the preter K d. If the syste hs short settling tie nd High frequency fluctutions, first decrese the preter K d, then decrese the preter K p nd in the end decrese the preter Ki. The considered CRAC syste controlled by the fuzzy- PID type nd self-tuning PID-type fuzzy dptive controller is shown in the block digr Fig.6. T ref is the desired reference teperture nd T is the rel teperture in dtcenter. The PID preters re tuned by dptive fuzzy controller. Fig.6 shows the digr of self-tuning fuzzy-pid control siultion syste of CRAC fn. PID control odel, fuzzy PID odel [12] nd self-tuning fuzzy-pid control odel re respectively

8 222 J. Deng et l. estblished in the control link of the roo teperture control nd re contrsted through the MATLAB siultion [13]. Trnsferred discrete PID controller odule fro the sub-librry of Siulink [14], fter repeted experients, took the rtio coefficient K p =171, integrl tie constnt K i =0. 001, differentil tie constnt K d =10 in PID control syste [15]. The design ide of fuzzy-pid control is to find out the fuzzy reltionship between 3 preters of fuzzy- PID control nd devition e, chnge rte of devition e, continuous detection e nd clcultion the e during the opertion nd online odified control preters in ccordnce to the control principle, to eet the different requireents in different e nd e, so tht the controlled object hs good dynic nd sttic properties. We choose the set point to rect bsed on the teperture in which rnges fro 64.4F (18 ) to 80.6F (27 ). Fig. 6. The control digr of CRAC fn syste Fig. 7. The digr of self-tuninsiultion fuzzy-pid control, fuzzy-pid control nd PID control Fig.8 shows the siultion results: (1) PID control. More sensitive to interference [17], control is not stisfctory, with lrge overshoot. The siultion found tht there will be instbility if the interference is too lrge. (2) Generl fuzzy-pid control. Fig.7 shows the teperture control of dtcenter when the clssicl PID, fuzzy- PD type nd the self-tuning PID-type fuzzy dptive controller re pplied [16]. To siulte nd copre fcilitte the control odel, set the supposed teperture for 25.5, siulte tie for the 1500s.

9 Self-tuning PID-type Fuzzy Adptive Control for CRAC in Dtcenters 223 Subjective the control rules, there is no self-lerning function. The siultions found tht the doin of fuzzy controller hd been chnged, required different preters nd bigger devitions in different given tepertures. (3) Self tuning Fuzzy-PID control. Self-tuning Fuzzy-PID control cn chieve better control [18], shorter response tie, sll overshoot, nd with good response perfornce nd robustness, nd cn chieve better control. The perfornce of the self-tuning PID-type fuzzy dptive controller is the best ong the others in ters of both the stedy-stte error nd the settling tie. There is no stedy-stte error [19] nd the syste reches the desired reference teperture [20] while iniizing settling tie. Although the fuzzy-pd type controller hs no stedy-stte error, it hs longer settling tie. The clssicl PID controller hs the worst perfornces. Fig. 8. The teperture control of dtcenter Fig.9 shows K p -K i -K d vrition of dtcenter when the self-tuning PID-type fuzzy dptive controller is pplied. Fig. 9. Tuning trjectories of PID controller gins

10 224 J. Deng et l. 6 Conclusion Over the course of this study three types of CRAC fn speed controller were relized (nd tested in the dt center) by using MATLAB/SIMULINK pck-ge progr nely, the PID controller, the fuzzy-pd controller nd self-tuning fuzzy PID controller. The result shows tht the self-tuning fuzzy PID controller is the best ong the others, in ters of both the stedy-stte error nd the settling tie. References 1. Bsh, C.E., Ptel, C.D., Shr, R.K.: Efficient Therl Mngeent of Dt Centers Iedite nd Long-Ter Reserch Needs. Intl. J. HVAC&R. Res. (2003) 2. Boucher, T.D., Auslnder, D.M.: Vibility of Dynic Cooling Control in Dt Center Environent (2004) 3. Lee, E.K., Kulkrni, I., Popili, D., Prshr, M.: Proctive Therl Mngeent in Green Dtcenter. The Journl of Super-coputing (2010) 4. Slsbury, T.I.: A teperture controller for VAV ir-hnding units bsed on siplified physicl odels. HVAC nd R. Res. (1998) 5. Kshr, M., Mtsub, T., Bouchernd, T.D., Auslnder, D.M.: Design nd tuning of robust PID controller for HVAC syste (1999) 6. Bi, Q., Ci, W.-J., Wng, Q.-G.: Advnced controller uto-tuning nd its ppliction in HVAC syste (2000) 7. Shyeghi, H., Shynfr, H.A., Jlili, A.: Multi stge fuzzy PID power syste utotic genertion controller in the deregulted environent. Energy Convers Mnge (2006) 8. Shyeghi, H., Jlili, A., Shynfr, H.A.: Robust odified GA bsed ulti-stge fuzzy LFC. J Energy Convers Mnge (2007) 9. Schi, R.R.: Therl profile of high-density dt center-ethodology to therlly chrcterize dt center. Trns A Soc Het RefrigAirCondEng, ASHRAE (2004) 10. Schi, R.R., Krki, K.C., Ptnkr, S.V.: Rised floor coputer dt center: perforted tile flow rtes for vrious tile lyouts Therl profile of high-density dt centerethodology to therlly chrcterize dt center. In: Proc. of Intersociety Conference on Therl Phenoen in Electronic Systes, ITHERM (2004) 11. To, Y.Y.H., New-style, G.Y.X.: PID control nd ppliction (2001) 12. Lei, L., Wng, H., Yu, Y.: Adptive Fuzzy PID Control Method Bsed on Identifiction Structure. Interntionl Journl of Syste nd Control (2006) 13. Yng, Y.: Appliction of MATLAB in PID Control Theory s teching Refor. Journl of Chngshu Institute of Technology (2009) 14. Wng, S., Jing, W.: PID Tuning Bsed on MATLAB/Siulink. Industry Control nd Applictions (2009) 15. Lu, R.: Mtter-eleent Modeling of Prllel Structure nd Appliction bout Extension PID Control Syete. Jourl of Systes Science & Coplexity (2006) 16. Wng, L., Du, W.: Fuzzy self-tuning PID control of the opertion tepertures in twostged ebrne seprtion process. Journl of Ntrul Gs Cheistry (2008)

11 Self-tuning PID-type Fuzzy Adptive Control for CRAC in Dtcenters Zi, B., Dun, B., Qiu, Y.: Fuzzy-PID control bse on disturbnce observer nd its ppliction.systes Engineering nd Electronics (2006) 18. M, Y., Liu, Y., Wng, C.: Design of Preters Self-tuning Fuzzy PID Control for DC Motor. In: The 2nd Interntionl Conference on Industril Mechtronics nd Autotion (ICIMA 2010) (2010) 19. Yu, Q.: Bsed on MATLAB s Anlysis of Stedy-Stte Error in Control Syste. Journl of Chngshu College (2004) 20. Liu, Y.: Study on boiler teperture syste PID controlbsed on RBF neurl network. Journl of Boji University of ArtsndSciences (Nturl Science) (2011)