Load Frequency Control Using Intelligent Techniques

Transcription

1 Load Frequency Control Usng Intellgent echnques D. Ghanbar,. Mahmood hahd Abbaspour Dam & Hydro Power Plant Operaton & Generaton Co, Masedsoleyman, Iran Islamc Azad Unversty Izeh ranch, Izeh, Iran Abstract: In ths paper a bran emotonal learnng based on ntellgent controller (ELIC) and fuzzy logc wth Mamdan and ugeno type fuzzy nference has been suggested for two area hydro electrcal power system and compared wth PI controller for elmnaton of the problems load frequency control such as frequency devatons and power flow control n transent condton. he Proposed controllers are substtuted nstead of conventonal PI controller. Intellgent controllers have some parameters when tuned better response can be acheved. hese controllers are flexble and robust aganst change n system parameters, loads, generaton rate constrants and other nonlnear factors n a power system. hese controllers are used for twoarea hydro electrcal power system and smulaton results have confrmed the superorty of ntellgent controllers over conventonal controllers. he Applcaton and performance of these controllers have been verfed n lnear and nonlnear systems. he mulaton has been done n matlab smulnk software. Keywords: Load Frequency Control, ran Emotonal, Fuzzy Logc. I. INODUCION In recent years, the formaton of power system has become more complcated, and ts constructon s made of nonlnear elements. Many parts of ndustres depend on power qualty. nce power systems are constructed of nterconnected parts, f a fault occurs n a subsystem, the subsequent perturbaton can affect n the power system qualty. Any unbalancng between producton and demand can cause frequency devaton. he frequency of a system depends on balancng the actve power. nce the frequency s the common factor n the whole system, the change n the demand of actve power n a dstrct of power system can be affected n overall power system. At nterconnected system wth two or more ndependent control areas, besde frequency control, generaton control s crucal to kept nterchange power at a programmed level. Controllng producton and frequency s referred as "Load Frequency Control" (LFC) and s a very mportant factor for power system qualty. he dynamc operaton of many ndustres and factores s very affected by perturbaton of the changes of the acton pont and functon of the power system. he ntal am of LFC s to have a constant frequency durng changes of the load. In a multareas nterconnected power system, control areas are connected together through lnes and load ncreasng lead to frequency decrease. It s necessary to have a control strategy to acheve the followng goals []: Keep frequency at a nomnal value. Load changng n one area should be responded at the same area. Fnally LFC should be senstve to load frequency changng, so desgnng the controller for LFC s very mportant. PI controller s one of the conventonal controllers for LFC. hs controller has a constant gan that s desgned based on nomnal work pont, so the utlzaton of ths controller s very smple but frequency devatons are hgh. When system parameters and load suddenly change, ths controller cannot respond well and t has devatons n transent condtons. o overcome ths problem ntellgent methods and algorthms such as genetc algorthm[], partcle swarm optmzaton[3],artfcal neural network[4] and nonlnear controllers such as sldng mode control[5], adaptve controller[6] are proposed for LFC for twoarea system n recent years. Due to the complcated constructon of large power systems and the dffculty of smulatng and analyzng such systems, n ths paper a two area power system s consdered to study and smulate. II. WO AEA POWE YEM wo area power system structure depcted n Fg [7]. g p Governor urbne Power ystem Copyrght to DOI 0.748/ ACE K ACE K Area L ynchronzed g p Governor urbne Power ystem Area L Fg.. wo area power system block dagram ransfer functon and turbnegovernor model of a power system are shown n Fg. whle Kt, Kg, H,,, K, Δf

2 are turbne transfer functon gan, governor transfer functon gan, generator nerta constant, speed regulaton coeffcent, bas frequency coeffcent, controller gan and frequency error at th area, respectvely. he tate varables area s as follows: X A x ( t ) u ( t ) L d ( t ) Whle A s the state matrx of the system, s the nput of the system and L s the nput's dsturbance. In () the state vector(x) s equal to: x F, P, P, P, F, P, P g t te g t Whle Pte, Pt, P g, F are synchronzaton coeffcent, mechancal power changes, nput flud changes and frequency error of th area respectvely. u u, u and, Whle U s the output of controller n th area d P L P L whle P s load dsturbance n th area. In two area power system, area control error (ACE ) s a lnear combnaton of Δf and ΔP te () and should be equal to zero. For a two area power system, ACE n each area control s defned as equatons, 3: A C E P F A C E P F 3 o the output of controller n each area s as follows: o u t KI A C E d t 4 o u t KI A C E d t 5 And to determne the system response n steady state, must: P F 0 6 P F 0 7 hese equatons are true n the followng condton: F P 0 III. INELLIGEN YEM Conventonal control methods are based on dentfcaton and system modelng and conventonal controllers are desgned accordng to predefned obectves. Identfyng the complex system s dffcult and sometmes mpossble. On the other sde due to change n system parameters, t s crucal to redesgn the controllers whch s very dffcult and ntrcate. One of the new control methods n engneerng scences s the use of ntellgent controllers whch are ndependent to the system model. In ths method there s a strong tendency to get nspred by natural systems. In ths method there s a strong tendency to get nspred by natural systems. Desgnng ntellgent controllers has been the focus of research n recent years. Artfcal ntellgence s referred to systems whch can do some humanlke reactons such as: percevng complcated condtons, ntellectual process smulaton, human reasonng methods and respondng them well, learnng capablty and reasonng to solve problems. Artfcal ntellgence was mpossble before nventng computers and t s a part of computer scences. oday's artfcal ntellgence s one of the mportant topcs n engneerng, basc, medcal and management scences. ome branches of artfcal ntellgence are as follows: Computatonal Intellgence, evolutonary ntellgence, swarm ntellgence, artfcal neural network, fuzzy system and bran emotonal controller [8]. III A. Fuzzy Logc Fuzzy logc was frst ntroduced by Prof Lotfzadeh at 965 as fuzzy sets [9]. Fuzzy logc s a new method to smplfy the conventonal methods to desgn and modelng systems there s a need for complex and advanced mathematc Fuzzy logc algorthm usually ncludes a set of reasonng rules where can spot t as a nonmathematcal and adaptable algorthm based on verbal process n contrast to conventonal algorthms. oday fuzzy systems are used n many scences such as: control, sgnal processng, communcatons, ntegrated crcuts, medcne and busness [0]. An mportant applcaton of fuzzy logc s n control systems. he man structure of the fuzzy algorthm s shown as a block dagram n Fg.. Actual Input Fuzzfcaton Inference Engne Defuzzfcaton Fg.. Fuzzy logc algorthm structure Actual Output Where fuzzfcaton step s defned as fuzzy sets for nputoutput varables. o defne fuzzy sets, there should be the prmary knowledge of the varaton doman of each varable. At nference step fuzzy rules are made. hese rules determne the value of control sgnals wth regard to nputs. Fnally at defuzzfcaton stage the real value of outputs s determned wth regard to nference step level. III. Intellgent Control ased on ran Emotonal Learnng ran emotonal learnng algorthm s based on the exstence of antbotcs and s a member of ntellgent system famly whose applcaton s wdespread Nowadays ntellgence computatonal s successfully used for solvng complex problems. he man dfference between ntellgent system and classc system s learnng and educaton ssues. A common feature of the learnng progress s the capablty of adaptaton to parameters varatons and envronmental condtons. One of the necesstes of the learnng algorthm s crcumstance evolutonary. ran emotonal learnng s one of the newest of these methods and s based on bran emotonal evolutonary. hs progress was made n a part of the bran of mammals called "lmbc system"[]. Moren and alkenus offer a method nspred by amygdala computatonal and orbtofrontal cortex model n lmbc system [3] and wth respect to ths model they offer bran emotonal learnng based on ntellgent controller (ELIC) [4]. Proposed model s adapted for complex system [5]. ELIC can supersede conventonal controllers such as PID controller. mulaton results show Copyrght to DOI 0.748/.05.30

3 ELIC's responses are faster, wth lower overshoot or undershoot and are robust aganst the varatons of the system parameters and dsturbances compared wth conventonal PID controller. In real tme control systems due to smplfcaton, low computatons and fast educatng, emotonal learnng s a very powerful method [6]. ecently ELIC controller s proposed for systems such as: power systems [7], electrcal machnes control [8] and nonlnear systems [9].All of the studes show robustness and good operaton of ELIC. IIIC.lmbc system structure Lmbc system s a part of mammal bran and ts task s processng emotonal sgnals. It s composed of Amygdala, orbt frontal cortex, thalamus, hypothalamus, sensory cortex etc. Fg 3 shows lmbc system structure [0]. and orbtofrontal cortex have lattce structures where there s a node for senstve nputs n each. One node s thalamus's nput n amygdale and ts value s equal to maxmum value of senstve nputs. he output of nodes n amygdale and orbtofrontal cortex s computed based on the followng equatons respectvely: Fg.5.Amygdala computatonal model block dagram Fg.3. lmbc system structure he relatonshp between amygdala and other lmbc system element s shown n fg.4 [0]. A V 8 O W Where A, O are nod's output n amygdale lattce and orbtofrontal cortex respectvely. V and W are nod's weght and s sensory nputs. V and W changes are computed through followng equatons: V m a x 0, A W A O 9 0 Fnally the output of the model s computated as follows: E A O IV. LFC UING FUZZY LOGIC LFC block dagram wth fuzzy logc s shown n Fg.6. Area Fg.4. Amygdala relatonshps n lmbc system. III C. Amygdala computatonal model lock dagram of amygdale computatonal model s shown n fg. 5 [0]. ased on new theores, amygdala system and orbtofrontal complete the learnng progress n two steps. Actuaton nternal sgnals are frst evaluated and then used for renforcement coeffcents n affected actuaton response. hese sgnals actuate amygdala and a amygdala's response s used for learnng contnuously. esdes orbtofrontal cortex acts as a reformer for the napproprate amygdala's responses and reactons. In computatonal model amygdale ACE Fuzzy Controller ACE Fuzzy Controller L g p Governor urbne Power ystem ynchronzed g p Governor urbne Power ystem Area L Fg.6.LFC system usng Fuzzy Logc Copyrght to DOI 0.748/

4 he only dfference between fuzzy and conventonal control s that fuzzy controller s a substtute for ntegrator gan. IVA.Mamdan ypeinference ystem hs method was proposed frst by Mamdan n 975[]. hs method was appled based on the combnaton of verbal control rules wth human experences. In ths method all of the nput and output varables are defned as fuzzy membershp functon. In ths paper turbne tme constant (P), bas frequency coeffcent () and synchronzaton coeffcent () are the nputs of fuzzy controller and k s the output. Input and Output membershp functon s shown n Fg.7. V. LFC UING ELIC ELIC structure s shown n Fg 9. ELIC replaces conventonal PI controller specfcally. ran Emotonal Learnng ased Intellgent Controller (ELIC) Fg.7. Input Output Membershp Functons ACE eward gnal Functon W ensory Input ΔW β ( A O w ) ΔV α( MAX(o, w A )) A V, O W E A O U Plan Load Frequency Control(LFC) ΔF Input level control based on output s shown n Fg.8. Fg.9. ELIC' tructure ensory nput () s a functon based on ACE and s desgned as follows: d K A C E K A C E K A C E d t s s 3 s dt 3 esdes emotonal sgnal s desgned as follows: Fg.8. level control IV. ugeno ype Inference ystem ugeno Inference method was frst proposed by akag ugeno Kang n 985[].hs method s smlar to Mamdan nference method n many aspects such as: nput fuzzfcaton and utlzng logc operators. he only dfference s the output membershp functon of ugeno method s a lnear combnaton of nputs or constant. Fuzzy rules for Mamdan and ugeno nference system are shown n table. hese rules are desgned based on prmary knowledge of conventonal PI controller n LFC. able I. Fuzzy rules for Mamdan and ugeno nference system. w K A C E K A C E r r F K 3 r u 4 We can have proper response wth tunng K r, K r, K 3r and K s, K s, K 3s and select α and β adequately. VI. IMULAION EUL Comparson between ELIC and fuzzy logc wth conventonal PI controllers are shown n Fg9 to Fg6. he smulatons tme s 5 second and mulatons are done n matlab smulnk software wth contnuous ntegrator. Frequency devaton n area and area n normal condtons are shown n Fg9 and Fg0. e lne power devaton s shown n Fg. Frequency devaton n area and area and te lne power devaton n abnormal condton such as change n dsturbance load, bas coeffcent frequency, synchronze coeffcent, tme constant turbne n two areas are shown n Fg to Fg 4 respectvely. Fnally, devaton n power mechancal n two areas s shown n Fg 5 and Fg6. Copyrght to DOI 0.748/

5 Fg.9. Frequency devatons n area Fg3. Frequency devatons n area wth abnormal condtons Fg.0. Frequency devatons n area Fg4. e Lne power devatons wth abnormal condtons Fg. e Lne power devatons Fg5. Power devatons n Area Fg. Frequency devatons n area wth abnormal condtons Fg6. Power devatons n Area Copyrght to DOI 0.748/

6 VII. CONCLUION he man am of LFC s balancng between generaton and consumpton. Due to complexty and mult varable power system, n abnormal condtons conventonal PI controller cannot have proper response, so robust and ntellgent controllers such as ELIC and fuzzy logc are used. o evaluate the PI, ELIC and fuzzy logc controllers n normal and abnormal system condtons, crterons such as: Lower ntegral error (IE), lower over shoot, better transent response, zero steady state error and hgher velocty n response are consdered. mulaton results show superorty and robustness of ntellgent controllers. Appendx: wo area power system parameters ELIC parameters controller EFEENCE [] P.Kundur Power system stablty and control,mcgrawhll, 994. []. Venkata Prasanth,. V. Jayaram Kumar Load Frequency Control For a wo Area Interconnected Power ystem Usng obust Genetc Algorthm Controller Journal of heoretcal and Appled Informaton echnology, pp 04, 005. [3] anuva Nageswara ao, P.ama Krshna eddy, PO based tunng of PID controller for a Load frequency control n two area power system Internatonal Journal of Engneerng esearch and Applcatons (IJEA), Vol., Issue 3, pp , 05. [4] K.abah, M.A.Nekou, M. eshnehlab, M. Alyar and M. Mansour Load Frequency Control n Interconnected Power ystem Usng Modfed Dynamc Neural Networks proceedngs of the 5th Medterranean on Control & Automaton, July 79, AthensGreece, 007. [5] K. Vrdolak, N. Per c, I. Petrov c ldng mode based loadfrequency control n power systems Electrc Power ystems esearch, Volume 80, Issue 5, Pages 54 57, May 00. [6] C.. Pan, C.M. Law An adaptve controller for power system load frequency control Power ystems, IEEE ransactons on power system, Volume 4 No, pp 8, Feb 989. [7] Had aadat, Power ystem Analyss McGrawHll, 999. [8] J. Moren, Emotonand learnng: A computatonal model of the Amygdala, Ph.D. dssertaton, Lund Unv., Lund, weden, 00. [9] L.A. Zadeh, "Fuzzy sets," Informaton and Control, Vol. 8, pp , 965. [0] Yager, onald., Zadeh, Lotf A. An Introducton tofuzzy Logc Applcatons n Intellgent ystems prnger U 99. [] J.Harrs An Introducton to Fuzzy Logc Applcatons prnger Netherlands, 000. []. Maren, Longerm Potentaton n the Amygdala A Mechansm foremotonal Learnng and Memory, rends n Neuroscences, Vol., No.,pp , 999. [3] J. Moren, C. alkenus, A Computatonal Model of Emotonal Learnng n the Amygdala, Cybernetcs and ystems, Vol. 3, No. 6, pp , 000. [4] C. Lucas, D.hahmrzad, N.hekhleslam, Introducng ELIC: ran Emotonal Learnng ased Intellgent Controller, Internatonal Journal of Intellgent Automaton and oft Computng, Vol. 0, NO., pp., 004. [5] D. hahmrzad, Computatonal Modelng Of he ran Lmbc ystem And Its Applcaton In Control Engneerng, Master dssertaton, exas A&M Unversty, U..A., 005. [6] M. Fatourech, C. Lucas and A. Khak edgh, educng Control Effort by means of Emotonal Learnng, Proc. of 9th Iranan Conf. on Electrcal Engneerng, ICEE 0, ehran, Iran, Vol. 4, pp. 8, 00. [7]. Mohammad, C. Lucas and. N. Araab, A novel controller for a powersystem based ELIC, Proceedngs of World Automaton Congress, Vol 8, pp , 004. [8] E. Daryabeg, G. Arabmarkadeh and C. Lucas, multaneously, speed and flux control of an nducton motor, wth bran emotonal learnng based ntellgent controller (ELIC), IEEE Internatonal Conference on Electrc Machnes and Drves, pp , 009. [9] A.. Mehraban, C. Lucas, Emotonal Learnng ased Intellgent obust Adaptve Controller for table Uncertan Nonlnear ystems, Internatonal Journal of Intellgent echnology (IJI), pp. 3440, 005. [0] C. Lucas, D. hahmrzad and M. glarbegan, Coevolutonary Approach to GraphColorng Problem, echncal Journal of Amrkabr Unversty of echnology, Vol. 4, NO.54, pp ,003. [] Mamdan, E.H. and. Asslan, An experment n lngustc synthess wth a fuzzy logc controller, Internatonal Journal of ManMachne tudes, Vol. 7, No., pp. 3, 975. [] ugeno, M., Industral applcatons of fuzzy control, Elsever cence Pub. Co., 985. IOGAPHIE Davoud Ghanbar was born n Ahwaz, Iran. He receved the. n Azad Islamc Unversty Dezful ranch n Dezful, Iran n 008 and M.c electrcal engneerng from the hahrekord Unversty, hahrekord, Iran n 0. Hs man research s n the area of applcaton mult phase electrc motor drves and ntellgent system n general. He has worked n hahd Abbaspour Dam & 000 MW Hydro Power Plant Operaton & Generaton Co, Masedsoleyman, Iran snce0. abak Mahmood was born n Izeh, ran. He receved.c n Islamc Azad Unversty, Dezful branch, Iran n 008 And M.sc electrcal engneerng from the same place at 0. Hs man research nterest s power system dynamc and control, Facts devces And Intellgent systems; he has worked n Islamc Azad Unversty, Izeh branch as a teacher, Izeh, Iran snce 009. Copyrght to DOI 0.748/