Aalborg Universitet. Published in: I E E E Transactions on Smart Grid. DOI (link to publication from Publisher): /TSG.2015.

Transcription

1 Aalborg Unverstet A Multagent-based Consensus Algorthm for Dstrbuted Coordnated Control of Dstrbuted Generators n the Energy Internet Sun, Quye; Han, Renke; Zhang, Huaguang; Zhou, Janguo; Guerrero, Josep M. Publshed n: I E E E ransactons on Smart Grd DOI (lnk to publcaton from Publsher):.9/SG Publcaton date: 5 Document Verson Early verson, also known as pre-prnt Lnk to publcaton from Aalborg Unversty Ctaton for publshed verson (APA): Sun, Q., Han, R., Zhang, H., Zhou, J., & Guerrero, J. M. (5). A Multagent-based Consensus Algorthm for Dstrbuted Coordnated Control of Dstrbuted Generators n the Energy Internet. I E E E ransactons on Smart Grd, 6(6), /SG General rghts Copyrght and moral rghts for the publcatons made accessble n the publc portal are retaned by the authors and/or other copyrght owners and t s a condton of accessng publcatons that users recognse and abde by the legal requrements assocated wth these rghts.? Users may download and prnt one copy of any publcaton from the publc portal for the purpose of prvate study or research.? You may not further dstrbute the materal or use t for any proft-makng actvty or commercal gan? You may freely dstrbute the URL dentfyng the publcaton n the publc portal? ake down polcy If you beleve that ths document breaches copyrght please contact us at vbn@aub.aau.dk provdng detals, and we wll remove access to the work mmedately and nvestgate your clam. Downloaded from vbn.aau.dk on: May 8, 6

2 ths document downloaded from s the preprnt verson of the paper: Q. Sun, R. Han, H. Zhang, J. Zhou, and J. M. Guerrero, "A mult-agent-based consensus algorthm for dstrbuted coordnated control of dstrbuted generators n the energy nternet," IEEE rans. Smart Grd, 5. A Mult-Agent-based Consensus Algorthm for Dstrbuted Coordnated Control of Dstrbuted Generators n the Energy Internet Quye Sun, Member, IEEE, Renke Han, Huaguang Zhang, Fellow, IEEE, Janguo Zhou, and Josep M. Guerrero, Fellow, IEEE Abstract Wth the bdrectonal power flow provded by the Energy Internet, varous methods are promoted to mprove and ncrease the energy utlzaton between Energy Internet and Man-Grd. hs paper proposes a novel dstrbuted coordnated controller combned wth a mult-agent-based consensus algorthm whch s appled to dstrbuted generators n the Energy Internet. hen, the decomposed tasks, models, and nformaton flow of the proposed method are analyzed. he proposed coordnated controller nstalled between the Energy Internet and the Man-Grd keeps voltage angles and ampltudes consensus whle provdng accurate power-sharng and mnmzng crculatng currents. Fnally, the Energy Internet can be ntegrated nto the Man-Grd seamlessly f necessary. Hence the Energy Internet can be operated as a spnnng reserve system. Smulaton results are provded to show the effectveness of the proposed controller n an Energy Internet. Index erms Energy nternet, dstrbuted coordnated control, spnnng reserve, mult-agent consensus algorthm, dstrbuted generators (DGs) I a Ir I v od voq od oq P Q NOMENCLAURE actve crculatng current reactve crculatng current total crculatng current number of dstrbuted generators (DGs) angle of voltage from DG the angular velocty of DG to represent the frequency d-axs output voltage components for DG q-axs output voltage components for DG d-axs output voltage components for DG q-axs output current components for DG average actve power for DG average reactve power for DG hs work was supported by the Natonal Natural Scence Foundaton of Chna (6386, 64334), the Fundamental Research Funds for the Central Unverstes (N44). Q. Sun, R. Han, H, Zhang, and J. Zhou are wth the School of Informaton Scence and Engneerng, Northeastern Unversty, Shenyang 89, Chna (sunquye@se.neu.edu.cn, hanrenke.neu@gmal.com, zhanghuaguang@se.neu.edu.cn, janguozhou.neu@gmal.com) J. M. Guerrero s wth the Insttute of Energy echnology, Aalborg Unversty, Aalborg 9, Denmark (Joz@et.aau.dk ) Correspondng author (Renke Han). el: ; fax: ; E-mal address: hanrenke.neu@gmal.com c m n set v set v od v oq od oq F C f ld lq k pvod kqvod k Vod k Voq d q ld lq v d v q L f cutoff frequency of power flter angle-actve droop parameter for DG voltage-reactve droop parameter for DG set value of angle-actve droop control for DG set value of voltage-reactve droop control for DG reference of voltage angle for DG d-axs reference of voltage ampltude n the voltage q-axs reference of voltage ampltude n the voltage d-axs auxlary state varables n the voltage q-axs auxlary state varables n the voltage feedforward parameter n the voltage controller flter capactor for DG d-axs current reference n the current controller for DG q-axs current reference n the current controller for DG d-axs proporton parameters n the voltage q-axs proporton parameters n the voltage d-axs ntegral parameters n the voltage controller for DG q-axs ntegral parameters n the voltage controller for DG d-axs auxlary state varables n the current q-axs auxlary state varables n the current d-axs flter current components for DG q-axs flter current components for DG d-axs output of current q-axs output of current flter nductance for DG k d-axs proporton parameters n the current pld k q-axs proporton parameters n the current plq k pld klq park d-axs ntegral parameters n the current controller for DG q-axs ntegral parameters n the current controller for DG Angle used for Park transformaton

3 W I. INRODUCION IH the large penetraton of renewable energy n power systems, flexble energy management and power sharng among dfferent dstrbuted generators (DGs) have rased a major concern. herefore, the concept of Energy Internet, motvated by the Internet, has been proposed recently [] and can provde promsng solutons. Prevous lteratures about Energy Internet manly focused on ts archtecture, system ntegraton, and control of the sold-state transformer [] [3]. However, n order to realze the envsoned Energy Internet concepton, the crtcal desgn and control objectves should be consdered: () mantanng flexble and proportonal powersharng among DGs; () mantanng system contnuous synchronzaton wth the Man-Grd consderng load varatons; (3) mnmzng crculatng currents between DGs, v) achevng seamless energy transtons between Energy Internet and Man-Grd f necessary. o realze proportonal power-sharng among DGs n the Energy Internet, varous control schemes have been proposed. he most mportant one s the droop control [4], an attractve dstrbuted control scheme that has been wdely studed recently snce ts operaton does not requre for hghbandwdth communcaton systems. In addton, a power sharng unt (PSU) s proposed to acheve power management strategy n a hybrd mcrogrd archtecture n [5]. However, the conventonal droop controller has several drawbacks such as load-dependent frequences and voltage ampltudes, large crculatng currents among DGs [6], the tradeoff between power-sharng accuracy and voltage synchronzaton [7] [8]. o overcome these shortcomngs, a modfed droop control strategy [9] has been proposed to mprove the power-sharng accuracy consderng the lne-mpedance effect through desgnng a proper vrtual mpedance. o restore the voltage and frequency of DGs to nomnal values, the so-called secondary control has been recently nvestgated n the lterature [] [], whch demonstrate that proper communcaton systems are necessary to realze the control goals. he secondary control can be classfed nto centralzed control and dstrbuted control n general. he centralzed control [] [3]requres a central unt to receve all the nformaton and to broadcast the decsons. Due to the centralzed nature, t presents a sngle pont falure whch can reduce the relablty and stablty of the whole system. Alternatvely, the mult-agent system as a knd of dstrbuted control structure has drawn much attenton due to ts flexblty and computatonal effcency [4] [5]. In [6], the partcpaton of a mult-agent-system-based mcrogrd nto the Energy Market s proposed. In order to charge electrc vehcles (EVs) at low electrcty prces, an agent-based control system that coordnates the battery chargng of electrc vehcles n dstrbuton networks s presented n [7]. In [8], the multagent system for EV chargng control s proposed based on the Nash Certanty Equvalence n order to solve the grd mpact. Meanwhle, the mult-agent consensus algorthm has been appled nto control system based on the mult-agent system structure [9] []. However, these applcatons only solved a sngle problem of voltage restoraton or frequency restoraton. Furthermore, Energy Internet always suffers from large crculatng currents caused by the slght dfferences among phases and ampltudes of the output voltages [], whch cannot be elmnated only by mantanng the synchronzaton of the output voltage angles or ampltudes. In order to overcome the aforementoned challenges, ths paper proposes a novel dstrbuted coordnated controller combned wth the mult-agent-based consensus algorthm to control DGs n the Energy Internet. wo man control objectves are acheved by the proposed control scheme: ) keep angles and ampltudes of all DGs output voltages beng synchronzed wth the Man-Grd nformaton (restore to ther nomnal values), whle keepng accurate proportonal powersharng; ) elmnate (mnmze) crculatng currents among DGs n the Energy Internet. Specfcally, P-δ and Q-V droop controls are adopted to suppress the crculatng currents, whle achevng proportonal power-sharng. hs can be acheved by regulatng both the angles (δ) and ampltudes (V) at the same tme. Due to the nonlnear feature of DGs n the Energy Internet, the proposed control approach s desgned based on the nput-output feedback lnearzaton control prncple [] and related stablty analyss approach [3]. he man features and benefts of the proposed controller are gven as follows: () the structure of the proposed controller related to the concepton of the mult-agent system s proposed ncludng tasks decomposton, types of agents and nformaton flow; () the novel dstrbuted coordnated controller combned wth the mult-agent consensus algorthm s proposed to control DGs n the Energy Internet; (3) the power can be shared n proportonal and the angles and ampltudes of output voltages can be synchronzed wth the Man-Grd; (4) the crculatng currents among DGs can be suppressed effectvely; (5) the control method requres only a sparse communcaton structure whch means each DG only needs ts local nformaton and ts neghbor s nformaton to acheve control objectves, then the controller can be more relable and less expensve. It s worthwhle to remark here that the proposed controller can brng extra benefts: () Energy Internet can be operated as a spnnng reserve system when the leader nformaton s from Man-Grd, then t can acheve seamless ntegraton nto Man- Grd; roughly speakng, the spnnng reserve means that the bdrectonal power flow between Energy Internet and Man- Grd could respond voluntarly to load dsturbances wthn a gven perod of tme [4] [5]; () facltated by ths advanced control and communcaton scheme, Energy Internet can provde opportuntes to other smart end users (flexble loads such as Intellgence Data Centers) n our daly lves to satsfy the needs of ther power demands whle mnmzng ther energy cost [6]; (3) the system can supply other ancllary servces such as market partcpaton as well requred by Man- Grd [7] [8]; (4) wth the ncreasng number of renewable sources and the development of Energy Internet, the fuel crss and envronmental problems can be solved gradually. hs paper s organzed as follows. In Secton II, the archtecture of the proposed controller based on mult-agent system s presented; the concepton of dfferent agents and nformaton flow among agents are descrbed. In Secton III, the dynamc model of the sngle-nverter-based DG s establshed. Secton IV dscusses the dstrbuted coordnated

4 3 controller desgned by the nput-output feedback lnearzaton and the mult-agent consensus algorthm. In Secton V, smulaton results verfy the effectveness of the proposed controller. In Secton VI, the comparsons of proposed dstrbuted approach versus centralzed approach are llustrated. Secton VII concludes the paper. Secton VIII llustrates the future work. II. HE ARCHIECURE OF HE DISRIBUED COORDINAED CONROL BASED ON MULI-AGEN SYSEM A. Requrements of the Proposed Dstrbuted Coordnated Control In order to acheve voltage synchronzaton, proportonal power-sharng and spnnng reserve requrements for Energy Internet, the dstrbuted coordnated controller combned wth mult-agent consensus algorthm s proposed. A typcal archtecture of Energy Internet ntegrated wth the proposed controller s shown n Fg.. he energy sources n an Energy Internet consst of dstrbuted renewable energy resources (DRERs), dstrbuted energy storage devces (DESDs), Man- Grd (MG). In ths paper, the desgnaton of the controller structure used n an Energy Internet s based on the mult-agent system structure [9] [3] [3]and each DRER, DESD, MG s controlled by each dfferent agent takng advantages of the autonomous, ntellgent, cooperatve proactve and adaptve features of the mult-agent system. Man Grd Communcaton Flow Control nformaton Flow Electrcal Connecton Energy Router Man-Grd Agent V AC Bus IFM AC DC BES DESD ESA Energy Swtcher AC DC SMES DESD ESAn Energy Swtchern FID LOAD AC DC PV DRER PDGA SDGA Energy Swtcher AC DC PV DRER PDGA SDGA Energy Swtcher LOAD Energy Internet AC DC W DRER PDGAn SDGAn Energy Swtchern Fg.. Archtecture of the Energy Internet he proposed Energy Internet structure and dstrbuted coordnated controller should meet the followng requrements: ) Acheve power-sharng n proportonal among dfferent DGs dynamcally and restore the ampltudes and angles of output voltages to nomnal value to keep trackng wth the nformaton from the leader (Man-Grd or one DG). ) Ensure the hgh relablty of the proposed algorthm and mantan seamless transton when Energy Internet swtches between the grd-connected mode and slanded mode, whch s called spnnng reserve condton. 3) Mnmze the crculatng current between dfferent DGs to enhance effcency of energy transmsson. 4) Acheve sparse communcaton structure to enhance the relablty and effcency of the control system. B. Defntons of Agents Each agent n the proposed controller has ts own goals and functons. Accordng to dfferent goals and functons, there are three dfferent agents desgned for the dstrbuted coordnated controller and two addtonal agents n the normal mult-agent system. he three dfferent agents nclude the Man-Grd Agent (MGA), the DG Agents (DGAs) whch consst of the Prmary-DGA (PDGA) and the Secondary-DGA (SDGA), the Energy Storage Agents (ESAs). he MGA s also called energy router whch s used to regulate the power flow between Energy Internet and Man-Grd. he ESA and SDGA are all called energy swtcher whch s used to regulate the power flow nsde Energy Internet. he other two necessary agents n a mult-agent system nclude the agent management servce agent (AMSA) whch s compulsory and the drectory facltator agent (DFA) [9]. In addton, the ntellgent fault management (IFM), and fault solaton devce (FID) are also nstalled on the transton lnes, whose functons are not dscussed n ths paper. he reference archtecture of multagent system s shown n Fg.. Message ransport System Drectory AMS Energy Internet MG Agent Man- Grd Sparse Communcaton Protocol IFM ES Agent Energy Storage FID SDG Agent PDG Agent Fg.. Reference archtecture of mult-agent system ) Man-Grd Agent (MGA): In ths paper, t s manly used to choose the operaton modes (grd-connected mode or slanded mode) for Energy Internet and calculate crculatng currents between Man-Grd and Energy Internet. It provdes the control nformaton for the ESA and DGA. he general archtecture of MGA s llustrated n Fg. 3. MG Agent Crculatng Current Calculaton Mode Selecton Decson Makng Module Communcaton Module Operaton Mode DG Power Defcency Fg. 3. Archtecture of MGA ) DG Agent (DGA): It conssts of two sub-agent called Prmary-DGA (PDGA) and the Secondary-DGA (SDGA) respectvely. he PDGA s used to acheve local control objectves for each DG and the SDGA s used to acheve the dstrbuted coordnated control between dfferent DGs. he general archtecture of DGA s llustrated n Fg. 4. 3) Energy Storage Agent (ESA): It s used to control the energy storage to compensate the power unbalance tmely by whch the control system can provde enough tme for DGA to response to the load dsturbance. he archtecture of ESA s shown n Fg. 5. 4) Agent Management Servce Agent (AMSA): It acts as a whte page, mantanng a drectory of agents regstered wthn the control system.

5 4 5) Drectory Facltator Agent (DFA): It acts as a yellow page, mantanng a drectory of agents and the servces they can offer other agents. he general archtectures of AMSA and DFA are llustrated n [9] and [3]. DG Agent Ampltude and Angles of output voltage Fg. 4. Archtecture of DGA ES Agent Power Defcency Fg. 5. Archtecture of ESA SDGA Status Collecton Module Physcal Model PDGA Status Collecton Module Physcal Model Ampltude and Angles of output voltage from neghborng DG Communcaton Protocol Control module Control module C. Implementaton of Proposed Dstrbuted Coordnated Control Energy Internet can be operated under two modes. he frst mode s that Energy Internet has no responsblty to connect wth Man-Grd and can be operated n the solated mode. In ths mode, one of the DGs should be chosen as the leader n the control system. In addton, the leader DG should be controlled to output rated power n order to guarantee ts leader nformaton wth a constant and normal value. he second mode s that Energy Internet should be connected to the Man-Grd f necessary; the nformaton from Man-Grd should be as the leader nformaton. In ths mode, Energy Internet can be operated under the spnnng reserve condton wth bdrectonal power flow to compensate the power dsturbance n both Man-Grd and Energy Internet. he man dfference between the two modes s whether Energy Internet should be connected wth the Man-Grd. Except for that, other operatons of the proposed controller of Energy Internet are same. Fg. 6 llustrates the flowchart of the proposed controller combned wth the algorthms, task decomposton and nformaton flow requred for the control system. Step ) Energy Internet confrms the operaton modes. If Energy Internet s operated under the frst mode, the MGA should choose one DG as the leader n the system. If Energy Internet should be operated under the second mode, the leader s nformaton should be detected from the Man-Grd. Meanwhle, the actve and reactve crculatng current n Energy Internet should be calculated accordng to the equaton () and (3). If I or a I suppress the standard value r I or anorm I, rnorm other agents wll be actvated to mplement ther tasks. Step ) he large crculatng current caused by the power unbalance means the bg devatons of ampltudes and angles of voltages among DGs n Energy Internet. In order to compensate the power unbalance tmely, the ESAs should be actvated. Meanwhle, ths operaton can provde enough tme for DGAs to compensate power unbalance. o be specfc, n ths paper, t s assumed that the DGs have enough power to provde for the load, thus the ESAs are used temporarly. he detal algorthm about how to operate ESAs s not dscussed n ths paper. Step 3) SDGAs can communcate wth ts neghbors nformaton by mult-agent communcaton protocol whch s desgned accordng to the mult-agent consensus algorthm. he devatons of angles and ampltudes of voltages between DGs are calculated by each SDGA through mult-agent consensus algorthm as shown n equaton () and (). hen, the devaton should be multpled wth the couplng gans and feedback control gans as the auxlary varables shown n equaton (5) as the control nput nto PDGAs. Step 4) PDGAs need only local nformaton wthout communcaton. Each PDGA should establsh the nonlnear model of each DG as shown n Secton III. hrough the feedback lnearzaton method, the nonlnear model can be transformed to the lnear model as shown n Secton IV-B. hs nformaton s embedded n each PDGA whch s used to control the local DG. hen, by usng the nformaton sent from the SDGAs, the PDGAs can calculate the local control sgnal accordng to the equaton (43). Fnally, the ampltudes and angles of voltages can be synchronzed wth the leader s nformaton thus the crculatng current can be effectvely mnmzed and the output power of DGs can be shared n proporton accordng to the equaton (7) and (8) and the power unbalance can be compensated effectvely. It should be emphaszed that not every DGA needs to receve the leader s nformaton because f one DGA can receve the leader nformaton, the other DGAs can produce the reasonable control sgnal to ts own DGs. hus, Energy Internet can be hghly effcent and relable. c, c k, k III. HE ANALYSIS OF CIRCULAING CURREN AND DESIGN OF HE PDGA BASED ON NONLINEAR MODEL OF DG he control loop, ncludng the power calculaton, voltage and current controllers, second order generalzed ntegrator, droop control, s used to acheve local control about the ampltude and angle of output voltage produced by dfferent DGs. A. Crculatng current analyss hrough the crculatng current analyss between DGs n [7] the crculatng current between two parallel connected DG s dvded nto actve crculatng and reactve crculatng current as I Ia jir ()

6 5 N Grd Connecton Demand? Y Swtchng on the Grd-Connecton swtcher 4 Confrm the operaton mode; Detect Man Grd nformaton Detect one DG nformaton Calculate crculatng current; Equaton () () (3) Y Ia<Ianorm Ir<Irnorm N yleader, yleader, 3 Suppresson of Crculatng Current; Compensatng the devaton of ampltudes and angles of voltages Calculate the devatons wth Man Grd or One Leader DG by Equatons () and () Coordnated power-sharng; Managng DG model (Desgned n Secton III) by feedback lnearzaton accordng to secton IV-B Regulatng the reference value to PDGAs by Equaton (43) Power unbalance (actve power and reactve power) Regulatng auxlary varable n order to calculate reference value to PDGAs by Equaton (5) Recoverng the devatons of ampltudes and angles of voltages by Equatons (7) and (8) Man-Grd Agent (MGA) Energy Router Compensated by DESD temporary Secondary DG Agents (SDGAs) Prmary DG Agents (PDGAs) Energy Storage Agents (ESAs) Fg. 6. Flowchart of the proposed controller for Energy Internet based on Mult-Agent System X 4 ' ' sn sn k Ro Ia E E () R A o X 4 ' ' cos cos k Ro Ir E E (3) R A o where s output voltage of the frst DG, s output voltage of the second DG,, are lne parameters between two DGs, E X k X k ' Xk ' Xk R o s the load between two DGs, E E, E E, R R / k k k k o A X X X X R. From equatons () and (3), t can be found that actve and reactve crculatng currents cannot be effectvely elmnated only by controllng the ampltudes or phase angles of the output voltages. hus the P and Q V droops are used to control output voltages ampltudes and angles. In order to mmc the behavor of a synchronous generator, the P droop control represents the lnear relatonshps between actve power and angle of output voltage and the Q V droop control represents the lnear relatonshps between reactve power and ampltude of output voltage. B. he desgn of PDGA combned wth nonlnear DG model In ths paper, dc-bus dynamcs can be safely neglected, snce a dc-lnk feed forward loop can be used [3]. he block dagram of a DG based on a sngle-phase nverter s presented n Fg. 7. he nonlnear dynamcs of DGs are controlled under d-q reference frame. In order to mmc a synchronous generator, the d-axs represents drecton of rotor magnetc flux lnkage and the q-axs s of 9 degrees ahead of d-axs. hus d-q reference usually forms a rotatng orthogonal reference frame used by three-phase DG. However, n ths paper sngle-phase nverters are consdered, so that the nverters output voltage and current should be dvded frst nto - components E o o through a second order generalzed ntegrator, as shown n Fg. 8 [33]. o be more specfc, the components are the transton reference frame from sngle-phase reference frame to the d-q reference frame. Voltage Controller Current Controller Power Controller - Second Order generalzed Integrator Second Order generalzed Integrator Fg. 7. Block dagram of a DG based on a sngle-phase nverter he phase of output voltage of one DG can be expressed as (4) Second Order Generalzed Integrator Fg. 8. Second Order Generalzed Integrator Because the P droop control s used n the system, no frequency devatons would occur under load dsturbances. he power controller shown n Fg. 9 conssts of the second order generalzed ntegrator, park transformaton, phaselocked loop (PLL), power calculaton, low-pass flter and droop controller. he power controller can provde the ampltude reference of voltage v for the frst stage brdge od and the angle reference of output voltage for the second

7 6 stage brdge. he dfferental equatons of the actve and reactve power can be expressed as Second Order Generalzed Integrator vodod voqoq P cp c (5) vodoq voqod Q cq c (6) PLL Fg. 9. he block dagram of power controller he P and Q V droop controls are presented as mp (7) set v v n Q (8) od set v oq By nsertng the (7) and (8) nto the (5) and (6), t yelds to c set c m (9) vodod voqoq P () c od set c n vodoq voqod Q v v () Durng the process of our model, the normal control about voltage and current should be ncluded accordng to [34], as shown n Appendx A. From the above equatons (4) and ()-(), and the equatons (7), (3)-(3), and (35)-(4) shown n Appendx A, the model of -th DG can be rewrtten nto a matrx formaton as x F x gu () y hx where the state vector s x, P, Q, d, q, d, q, ld, lq, vod, voq, od, oq In order to keep the angle and the ampltude DG beng synchronzed wth other DGs, the and are selected to be the control nputs. Snce v oq s kept to be zero, the ampltude of output voltage s v v (3) o od set v o of the -th herefore, the nputs of the system are u v and the outputs of the system are y v. In the next Secton, the feedback lnearzaton method combned wth the multagent consensus algorthm s used to desgn the SDGAs. IV. HE DESIGN OF HE SDGA BASED ON MULI-AGEN CONSENSUS ALGORIHM In ths Secton, the desgn of the SDGA based on the multagent consensus algorthm s proposed by usng feedback lnearzaton, whch can transform the nonlnear model to the lnear model. Each SDGA has two control nputs and two outputs. he consensus problem about angles and ampltudes of output voltages among DGs s the synchronzaton trackng od set set v set problem, whch needs a leader n the system. he leader nformaton s decded by the Energy Router. Each SDGA only requres local and ts neghbors nformaton, whch can be used to produce the control decson sgnal. o be more specfc, f the leader nformaton n the Energy Router comes from Man-Grd, the Energy Internet can be operated under spnnng reserve condton, whch means t can be connected to Man-Grd and mmedately provde power f necessary. A. Graph heory he dstrbuted coordnated controller should use a communcaton network called drected graph, whch can be expressed as Gr VG, EG, AG [35]. he set of nodes n the network can be expressed as VG v, v, vn, the set of edges can be expressed as and the weghted adjacency E V V G G G matrx can be expressed as adjacency element node s denoted by a j A G a j nn. An edge rooted at node v, v j wth nonnegatve j and ended at, whch means that nformaton can flow from the node to node. For a graph wth - adjacency elements, the n-degree and out-degree of node v are defned as follows: deg j n v a deg v a (4) n j out j j j he degree matrx of dagraph j where j for all j G r n s a dagonal matrx and deg v. he out Laplacan matrx assocated wth the dgraph s defned as LG A (5) B. he desgnaton of SDGA r From the dscusson n Secton III-B, the s used as the nput to control angles of output voltages among dfferent DGs to track the leader and the s used as the nput to y leader, control the ampltude of output voltage to track the voltage y. leader, he feedback lnearzaton method can establsh the relatonshp between the control outputs and nputs. For the sake of space, the feedback lnearzaton procedures are shown n Appendx B accordng to [36]. Accordng to process n Appendx B, the relatve degree should be calculated frst. he relatve degree of the nonlnear system s,. he decouplng matrx s calculated as he matrx calculated as A x c set v set c V set G r set A x s nonsngular. he matrx (6) b x s (7) b x L h x L h x f f where L h x m f c set c v od od v oq oq v v L h x v v n v od oq oq od f c set od c oq ld od C f

8 7 From (6) and (7), the control law u x can be calculated as n (43) (see Appendx B). he v x can be calculated as follows to acheve the synchronzaton of. v x A xu x b x (8) where v v, v,. he nonlnear system can be transformed nto the followng lnear system y, v, y (9), v, where y y, y,, y,, y, v od y,, B he nonlnear system () s transformed to the lnear system (9) by usng the well-known feedback lnearzaton method. Note that (9) presents the -th DG lnear dynamcs. he control of angles of output voltage s the trackng synchronzaton problem wth a leader n the system. he am s to make all the angles be equal to the leader. he trackng error s gven by e, aj y, y j, +, y, yleader, () where a j jn s the element of adjacency matrx, e, represents the error of angles of output voltage about the -th DG. If the node can receve the leader s angle nformaton, and edge v v s sad to exst wth weghtng gan. he, leader, node angle. wth, = he E dag R, s used as a pnned and controlled node about, NN,. can be wrtten nto matrx as he control of ampltude of the output voltage s also the trackng synchronzaton problem wth the system leader. hus, the trackng error s gven as e, aj y, y j,, y, yleader, () where e, jn represents the error of ampltudes of output voltage about the -th DG. If the node can receve the leader s ampltude nformaton, and edge v, leader, v s sad to exst wth weghtng gan. he node s used as a pnned, wth, = and controlled node about ampltude. he nto matrx as, E dag R NN., can be wrtten he errors can be rewrtten by usng matrxes as followng e L E Y Y () where + leader, L E Y Y leader, e (3) e = e, e, e n,, e = e, e, e n,, Y y, y, y n,, Yleader, N yleader, Y y, y, y n,, Yleader, N yleader, he system (9) can be rewrtten as Y INv Y INv. (4) where v v, v, v n, v, v, v n, v. he relatonshp between, and, should be desgned n order to keep the system stable wth the control nput. Let the auxlary control be defned as follows where c, c R e v, and e v, v v v, ck e, v, ck e, s the couplng gan and k, k (5) R s the feedback control gan. For sake of space, the detaled desgn process s not ncluded n ths paper. Remark : Accordng to the IEEE [37], f the ampltudes and angles of output voltages can be kept trackng wth the Man-Grd, the system can be connected wth Man- Grd f necessary and provde power, whch s called spnnng reserved condton. Under ths condton, the system can compensate the power unbalance wth the help from Man- Grd and can also provde the surplus power for Man-Grd. Along wth the novel dstrbuted coordnated controller ()- (5), Energy Internet can be ntegrated nto the Man-Grd seamlessly to provde reserve servces at any tme to ensure the relablty of the power system as long as the leader nformaton comes from Man-Grd. o be specfc, from the perspectve of power management, the bdrectonal power flow between Man-Grd and Energy Internet can be acheved to compensate power unbalance n both sdes. hen Energy Internet can be operated under spnnng reserve condton. From the perspectve of control method, the angle and ampltude devatons between Man-Grd and Energy Internet are wthn a small range all the tme, thus Energy Internet can be connected to the Man-Grd at any tme. Furthermore, varous methods on the optmzaton of spnnng reserve condton have also been nvestgated n [38], [39] and [4], whch s out the scope of ths paper. Remark : Mcrogrds and vrtual power plants (VPPs) are two recently proposed concepts. Even though lots of control methods for mcrogrd have been studed, the framework that may encompass a large number of control methods s not completed yet. It means each new control method needs a new control archtecture. Meanwhle, VPPs are used to ntroduce to energy and CO -reducton markets more aggressvely [4], n whch varous power producton unts can cooperate and behave as a sngle aggregated unt. he power market n mcrogrds s ntroduced by the concepton of VPPs [4]. Compared wth the above two conceptons, the control archtecture of Energy Internet ncludng energy router, energy swtcher s proposed based on mult-agent system. Meanwhle, the algorthms appled n each agent are studed n ths paper. V. SIMULAION RESULS In order to test the proposed approach, a number of smulatons under dfferent scenaros have been performed through MALAB/SIMULINK. he electrcal part of the smulaton s establshed through the SIMULINK, meanwhle the control method ncludng the feedback lnearzaton and mult-agent consensus algorthm s programmed through the S- functon. Dfferent confguratons of loads and DGs are

9 8 consdered n ths secton. Fg. shows the confguraton of the Energy Internet test system. In addton, the communcaton structure between agents s also shown n Fg. by dotted lnes. In the system, the DG- and DG- are connected n seres and the and are connected n parallel. hs confguraton conssts of seres and parallel connected DGs, whch are common n Energy Internet, thus the smulaton results can be more convncng. he system data used for smulaton are lsted n able I n Appendx C. he numercal values of smulaton results are shown n ables II and III n Appendx C. ESA ES t can be found that the output power from four DGs are not accordng to the desred ratos. Reactve Power (Var) DG- DG- Energy Router Z Ld Z Ld Z3 Z5 Ld4 Z6 L L Z4 Ld3 L4 Ld5 Leader Value DG- PDGA SDGA DG- PDGA SDGA L3 PDGA4 SDGA4 Fg.. Reactve Power Sharng n Case me (Sec) PDGA3 Communcaton Lne SDGA3 Power Lne Fg.. Confguraton of the Energy Internet est System A. Case : Conventonal Controller In ths case, the desred actve and reactve rato of DG-, DG-, and s::.33:.33. o nvestgate the load sharng wth reduced system, s dsconnected at.4s and the load power s shared by DG-, DG- and. At.7s, load-4 and load-5 are also dsconnected. he three DGs connected to the Energy Internet supply the 3 kw and 8 kvar load. Angle (Degree) DG- DG- me (sec) Fg. 3. Angles of Output Voltages from 4 DGs n Case DG-. Actve Power (W) DG- me (sec) Fg.. Actve Power-Sharng n Case It can be seen that the system operaton s stable. As shown n Fgs. and, the DGs cannot share loads n the desred rato of ::.33. he numercal results can be seen n able II n Appendx C. In able II, the output actve power and reactve power from four DGs are shown n numercal results and the rato between each DG output power to DG- output power are also shown n able II. he ratos n () are desred ratos between each DG to DG-. From the numercal results, he Ampltude of Voltage (pu) DG- DG- DG me (sec) Fg. 4. he p.u. value of Output Voltages from 4 DGs n Case From Fgs 3 and 4, the devatons of output voltages from dfferent DGs are large. For one thng, from Fg. 3, at the begnnng of the smulaton, the angle devatons among four DGs output voltages exst. After.4s, the devatons are becomng larger. For another thng, from Fg. 4, after.4s, the ampltude devatons between the output voltages are

10 9 becomng larger. hus the conventonal control method cannot be effectve to control the Energy Internet. From the above dscusson, the devaton angles among the four output currents shown n Fg. 5 are large. Accordng to the equaton ()-(3), the crculatng currents between DGs exst very largely n the system. hus wth bg devatons between output voltages, f the Energy Internet need to be connected to the Man-Grd, another synchronzaton algorthm should be added to control the Energy Internet. hus through the conventonal method, the Energy Internet cannot be operated under the spnnng reserve condton. And the bg crculatng currents n Energy Internet can cause unrelablty and low power effcency of the whole system. four DGs are shown n numercal results and the rato between each DG output power to DG- output power are also shown n able III. he ratos n () are desred ratos between each DG to DG-. Compared wth the numercal results n able II, t can be found that the output power-sharng ratos from four DGs are same wth the desred ratos. he accuraces about power sharng can be verfed. DG- 5 5 DG- DG- Reactve Power (Var) DG- Current (A) me (sec) Fg. 5. he Output Current n Case B. Case : Proposed Controller o valdate the performance of the proposed controller, the Energy Internet s operated at smlar stuaton as descrbed n Case. me (Sec) Fg. 7. Reactve Power Sharng n Case Angle (Degree) DG- DG- DG- me (Sec) Actve Power (W) DG- me (Sec) Fg. 6. Actve Power Sharng n Case he proposed controller s started at.s. From the communcaton structure, only DG- s SDGA can receve the Man-Grd nformaton from the Energy Router. he leader s nformaton s shown n able. I, Appendx C. he numercal values of power can be seen n able III, Appendx C. In able III, the output actve power and reactve power from Fg. 8. Angles of Output Voltages from 4 DGs n Case he Ampltude of Voltage (pu) data DG-.99 data DG- data data me (sec) Fg. 9. he p.u. value of Output Voltages from 4 DGs n Case

11 Compared wth results n Case, DGs can share the actve power and reactve power n the desred rato as shown n Fgs. 6 and 7. After.s, the ampltudes and angles of output voltages from DGs begn to reach a consensus wth the leader shown n Fgs. 8 and 9. hus the ampltude of output currents of four DGs are n desred rato and the devaton angles between four output currents are very small as shown n Fg.. From the equaton () and (3), the crculatng current between DGs can be mnmzed. Furthermore, snce both angles and ampltudes of DGs output voltages n the Energy Internet can be controlled to reach consensus wth the Man-Grd nformaton due to the proposed method. Wth enough energy storage, the Energy Internet can be operated under spnnng reserve condton. Current (A) me (sec) Fg.. he Output Current n Case Remark 3: Snce large crculatng currents n the Energy Internet can cause large unnecessary power losses, to reduce the crculatng current can effectvely decrease power losses n the Energy Internet. VI. COMPARISON BEWEEN HE PROPOSED DISRIBUED APPROACH WIH HE CENRALIZED APPROACH A. Comparson about computatonal power When a centralzed controller s used, all the system nformaton should be communcated to the central unt. Wth the ncreasng number of DGs and other energy sources, the communcaton burden may ncrease due to the addtonal varables and constrants. Meanwhle, the computaton tme of such a central controller s hghly dependent on the number of devces n the control system. hus large communcaton delays may occur n the system whch may cause serous negatve nfluence on the stablty and system performances, thus reducng the system scalablty drastcally. By comparson, n proposed controller, there s no centralzed controller n the system and the communcaton load for each controller s uncorrelated wth the ncreasng number of controllers n the system. Snce the mult-agent consensus algorthm s appled n ths paper, each controller only needs ts local nformaton and at least one of ts neghbors nformaton. hus each controller s computaton DG- DG- load s nearly constant when controllers are added or removed from the control system. he same results about the comparson of the computatonal power between dstrbuted controller and centralzed controller can be also found n []. B. Comparson wth the performance, reslence and scalablty When the system does not present faults, the performances between dstrbuted or centralzed controllers are almost equal. Indeed the centralzed controller gves all the local DGs the same control sgnal, whle the dstrbuted controller make control decson of each DG accordng to ts neghbors nformaton. hus the response speed of the dstrbuted controller can be faster than that of centralzed controller. Under fault condtons, f the centralzed controller fals down, the whole system wll lose the control and may become unstable. By comparson, once one dstrbuted controller fals down, t cannot cause serous nfluence on the whole system. he reason of ths s that each controller only needs ts local and neghbors nformaton based on sparse communcaton structure. Furthermore, f ths controller cannot be recovered n a short tme, we can dsconnect t to mantan the stablty of the whole system. However f the centralzed controller cannot be recovered tmely, t may not be possble to mantan stable operaton. he stablty about mult-agent consensus algorthm consderng swtchng topology has been proved n several prevous works lke those n [43] [44]. In the future, ths algorthm may be extended to the dstrbuted controllers. hus reslence of the dstrbuted controller can be much better than the centralzed controller one. Based on the proposed algorthm combned wth the multagent system, the dstrbuted controller n ths paper s stll open for mprovement, because several condtons are not ncluded n ths paper. Secton VIII dscusses some possbltes of mprovement for reference to the nterested reader. Meanwhle, the scale of the dstrbuted system can be enlarged wthout communcaton lmts, computaton lmts and algorthm lmts. By comparson, the scale of the centralzed controller s lmted by ts communcaton and computaton ablty. hus the scalablty of dstrbuted controller s much hgher than that of centralzed controller. VII. CONCLUSION he control ssues of n Energy Internet were nvestgated n ths paper. he mult-agent consensus algorthm and multagent system archtecture are combned to desgn the control structure and control method appled n Energy Internet. he decomposed tasks, establshed and transformed model, nformaton flow for each agent are studed. he benefts of ths paper are as follows: ) he combnaton of mult-agent system and mult-agent consensus algorthm wll provde an nfrastructure for future research used n Energy Internet. ) he crculatng current reducton n Energy Internet and proportonal power-sharng to the desred rato among DGs can be guaranteed by combnng wth the P- and Q-V controllers and mult-agent consensus algorthms. 3) he output voltages n Energy Internet can be recovered whle beng synchronzed wth the leader nformaton from Man-

12 Grd, whch means the Energy Internet can operate as spnnng reserve system. 4) For daly lfe, wth sparse communcaton systems and bdrectonal power flows, Energy Internet provdes customers a relable and effcent power supply, whle mnmzng energy costs and provdng the dfferent types of renewable energy resources the possblty to plugn/out at any tme. VIII. FUURE WORK Future work wll consst of, but wll not be lmted to, the followng aspects: ). he cost, locaton elements should be consdered when the proposed algorthm decdes the power sharng n Energy Internet. In ths paper, the power sharng between dfferent DGs only base on ther rated power, but n practcal the economc element s another mportant ndex. Meanwhle, the power market should be studed n Energy Internet n order to make the system operated more economcal. ). he changes of communcaton topology should be consdered n the future. In ths paper, the communcaton topology n the control system s kept constant. In the future, the swtchng topology mult-agent consensus algorthm should be appled n the control system. Wth ths applcaton, the comparson about reslence between dstrbuted controller and centralzed controller could be more obvous. 3). he cost about hardware durng the process of desgnaton should be consdered n the future. In ths paper, the cost about hardware between dstrbuted controller and centralzed controller are not compared whch can llustrate benefts about dstrbuted controller furthermore. 4). In ths paper, we only ntroduce a lttle about the energy storage wthout ts control strategy. Because the energy storage s very mportant under some extremely condtons, the control strategy for energy storage should be studed based on the proposed control archtecture n the future. APPENDIX A he block dagram of the voltage and current controller s shown n Fg.. Fg.. he Block dagram of the voltage and current controller he dfferental equatons of the voltage controller are presented as v v (6) od od od v v (7) oq oq oq he outputs of the voltage controller are presented as F C v k v v k (8) ld od f f oq pvod od od Vod od lq oq f f od pvoq oq oq Voq oq F C v k v v k (9) In addton, nsertng (8) nto (6), we obtan od vset nq vod (3) he dfferental equatons of the current controller are presented as (3) d ld ld q lq lq (3) he outputs of the current controller are presented as v L k k (33) d f f lq pld ld ld ld d v L k k (34) q f f ld plq lq lq lq q Fg. shows that the connecton between nverters are nductve. he dfferental equatons for the output LCL flter and and the output nductor are expressed as follows. L f C f L C R f ld ld lq vd vod (35) L L f f R f lq lq ld vq voq (36) L L h f v v C f (37) od oq ld od f voq vod lq oq (38) C x v v L f (39) od oq od bd c oq od voq vbq (4) L c APPENDIX B he process of establshng the relatonshp n a MIMO nonlnear system descrbed n () s explaned below. Frst, the relatve degree r, r,, r m should be calculated. Second, accordng to the relatve degree, a decouplng matrx A xwhch should be nonsngular should be calculated as gven n (4), another vector should be calculated at the same tme as n (4). In (4) and (4), Lfh x s the Le dervatve of wth respect to and s defned as Lf h x h / x f. r r Lg L f h x Lg L m f h x A x (4) r r L L h x L L h x b x g f m gm f m f r r m (4) b x L h x L h x f f m hrd, based on the defned relatve degree, the control law of a MIMO nonlnear system s defned as ux A x bx vx (43) where r r m m m v x v v y y. Appendx C ABLE I SYSEM PARAMEERS System Quanttes System Frequency DC voltage Leader nformaton Voltage Values 5Hz 8V p.u

13 Angle 57.4 degree Lne mpedance Z =Z =Z3 =Z4 =Z5 =Z6.5+j.5-5 Ω Load Ratngs Ld. KW and 5 kvar Ld. KW and 5 kvar Ld3. KW and 5 kvar Ld4 KW and 6 kvar Ld5 KW and 7 kvar DG ratngs DG- KW and 5 kvar DG- KW and kvar 3.3 KW and 6.6 kvar 3.3 KW and 6.6 kvar Output nductances L μh L 5 μh L3 7.5 μh L4 7.5 μh Droop Coeffcents Actve Power-Angle m -4 rad/w m 5-5 rad/w m rad/w m rad/w Reactve Power-Voltage n 4-4 rad/var n -4 rad/var n3 3-4 rad/var n4 3-4 rad/var ABLE II NUMERICAL RESULS OF CASE I CASE-: Secton V. A Reduced System of Fg. 7 wthout Proposed Controller Actve Power Intal Value (s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG- 9. kw 9.8 kw 6 kw PDG- 7. kw kw 4kW P.6 kw kw kw P.6 kw kw kw Actve Power Rato of P DG- Intal Value (s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG-. (.). (.). (.) PDG-.879 (.).43 (.).333 (.) P.6 (.33).4(.33).83 (.33) P.6 (.33). (.). (.) Reactve Power Intal Value (s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG- 5.4 kvar 5.8 kvar 3.kVAr PDG- 8.7 kvar kvar 6.kVAr P 6.8 kvar 3 kvar 4.7 kvar P 6.9 kvar kvar kvar Reactve Power Rato of P DG- Intal Value (s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG-. (.). (.). (.) PDG-.6 (.).6 (.).9 (.) P.9 (.33).4 (.33).469 (.33) P.7 (.33). (.). (.) ABLE III NUMERICAL RESULS OF CASE II CASE-: Secton V. B Reduced System wth Proposed Controller Actve Power Intal Value (s-.s) Intermedate Value (.s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG- 9. kw 9. kw.7 kw 7. kw PDG- 7. kw 8. kw 3.6 kw 3.9 kw P.6 kw kw 5.7 kw 9.3 kw P.6 kw kw kw kw Actve Power Rato of PDG- Intal Value (s-.s) Intermedate Value (.s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG-. (.). (.). (.). (.) PDG-.879 (.).978 (.).7 (.).986 (.) P.6 (.33).34 (.33).34 (.33).38 (.33) P.6 (.33).34 (.33). (.). (.) Reactve Power Intal Value (s-.s) Intermedate Value (.s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG- 5.4 kvar 4.9 kvar 6.45 kvar 3.5 kvar PDG- 8.7 kvar 9.9 kvar.95kvar 6.87 kvar P 6.8 kvar 6.6 kvar 8.55 kvar 4.63 kvar P 6.7 kvar 6.5 kvar kvar kvar Reactve Power Rato of PDG- Intal Value (s-.s) Intermedate Value (.s-.4s) Intermedate Value (.4s-.7s) Fnal Value (.7s-s) PDG-. (.). (.). (.). (.) PDG-.6 (.). (.). (.).96 (.) P.6 (.33).34 (.33).33 (.33).3 (.33) P.4 (.33).33 (.33). (.). (.) IX. REFERENCES [] A. Q. Huang, M. L. Crow, G.. Heydt, J. P. Zheng, and S. J. Dale, he future renewable electrc energy delvery and management system: he energy nternet, n Proc. IEEE, vol. 99, no., pp , Jan.. [] X. Yu, X. She, X. N, and A. Q. Huang, System ntegraton and herarchcal power management strategy for a sold-state transformer nterfaced mcrogrd system, IEEE rans. Power Electron., vol. 9, no. 8, pp , Aug. 4. [3] X. Yu, X. She, X. Zhou, and A. Q. Huang, Power management for dc mcrogrd enabled by sold-state transformer, IEEE rans. Smart Grd, vol. 5, no., pp , Mar. 4. [4] M. C. Chandorkar, D. M. Dvan, R. Adapa, Control of parallel connected nverters n standalone ac supply systems, IEEE rans. Ind. Appl., vol. 9, no., pp , Jan/Feb [5] Q. Sun, J. Zhou, J. M. Guerrero, and H. Zhang, Hybrd threephase/sngle-phase mcrogrd archtecture wth power management capabltes, IEEE rans. Power Electron., Publshed on-lne, 4, DOI:.9/PEL [6] S. V. Iyer, M. N. Belur, and M. C. Chandorkar, Analyss and mtgaton of voltage offsets n mult-nverter mcrogrds, IEEE rans. Energy Convers., vol. 6, no., pp , Mar.. [7] Z. Ye, P. K. Jan, P. C. Sen, Crculatng current mnmzaton n hgh frequency AC power dstrbuton archtecture wth multple nverter modules operated n parallel, IEEE rans Ind. Electron., vol. 54, no. 5, pp Oct. 7. [8] J. M. Guerrero, J. C. Vasquez, J. Matas, J. G. de Vcuna, and M. Castlla, Herarchcal control of droop-controlled AC and DC mcrogrds a general approach toward standardzaton. IEEE rans Ind. Electron., vol. 58, no., pp Jan.. [9] Y. W. L, and C.-N. Kao, An accurate power control strategy for powerelectroncs-nterfaced dstrbuted generaton unts operatng n a low-voltage multbus mcrogrd, IEEE rans. Power Electron., vol. 4, no., pp , Dec. 9. [] A. Bdram, A. Davoud, F. L. Lews, and S. S. Ge, Dstrbuted adaptve voltage control of nverter-based mcrogrds, IEEE rans. Energy Convers., to be publshed. [] J. M. Guerrero, L. Hang, J. Uceda, Control of dstrbuted unnterruptble power supply systems, IEEE rans Ind. Electron., vol. 55, no. 8, pp Aug. 8. [] X. Lu, J. M. Guerrero, K. Sun, J. C. Vasquez, R. eodorescu, and L. Huang, Herarchcal control of parallel ac-dc converter nterfaces for hybrd mcrogrds, IEEE rans. Smart Grd, vol. 5, no., pp , Mar. 4. [3] M. Savagheb, A. Jallan, J. C. Vasquez, and J. M. Guerrero, Secondary control scheme for voltage unbalance compensaton n an slanded droop-controlled mcrogrd, IEEE rans. Smart Grd, vol. 3, no., pp , Mar. 4. [4] H. Zhang,. Feng, G. Yang, and H. Lang, Dstrbuted cooperatve optmal control for multagent systems on drected graphs: an nverse optmal

14 3 approach, IEEE rans. Cyber., publshed on-lne on st, Sep. 4, DOI:.9/CYB [5] H. Zhang, J. Zhang, G. Yang, and Y. Luo, Leader-based optmal coordnaton control for the consensus problem of mult-agent dfferental games va fuzzy adaptve dynamc programmng, IEEE rans. Fuzzy Syst., vol. 3, no., pp. 5-63, 5. [6] A. L. Dmeas, N. D. Hatzargyrou, Operaton of a multagent system for mcrogrd control, IEEE rans. Power Syst., vol., no. 3, pp , Aug. 5. [7] P. Papadopoulos, N. Jenkns, L. M. Cpcgan, I. Grau, E. Zabala, Coordnaton of the chargng of electrc vehcles usng a mult-agent system, IEEE rans. Smart Grd, vol. 4, no. 4, pp. 8-89, Dec. 3. [8] E. L. Karfopoulos, N. D. Hatzargyrou, A mult-agent system for controlled chargng of a large populaton of electrc vehcles, IEEE rans. Power Syst., vol. 8, no., pp. 96-4, May 3. [9] A. Bdram, A. Davoud, F. L. Lws, and J. M. Guerrero, Dstrbuted cooperatve secondary control of mcrogrds usng feedback lnearzaton, IEEE rans. Power Syst., vol. 8, no. 3, pp , Aug. 3. [] W. Lu, W. Gu, W. Sheng, X. Meng, Z. Wu, and W. Chen, Decentralzed mult-agent system-based cooperatve frequency control for autonomous mcrogrd wth communcaton constrants, IEEE rans. Sustan. Energy, vol. 5, no., pp , Apr. 4. [] W. Yao, M. Chen, J. Matas, J. M. Guerrero, and Z. M. Qan, Desgn and analyss of the droop control method for parallel nverters consderng the mpact of the complex mpedance on the power sharng, IEEE rans. Ind. Electron., vol. 58, no., pp , Feb.. [] J. E. Slotne and W. L, Appled Nonlnear Control. Upper Saddle Rver, NJ, USA: Prentce-Hall, 9. [3] H. Zhang, D. Lu, Y. Luo, and D. Wang, Adaptve Dynamc Programmng for Control-Algorthms and Stablty. London: Sprnger-Verlag, 3. [4] M. Q. Wang, and H. B. Goo, Spnnng reserve estmaton n mcrogrds, IEEE rans. Power Syst., vol. 6, no. 3, pp , Aug.. [5] Y. Rebours and D. S. Krschen, What s Spnnng Reserve?, 5. [Onlne]. Avalable: eportstheses/aoe/rebours et al_tech rep_5a.pdf. [6] L. Rao, X. Lu, M. D. Ilc, and J. Lu, Dstrbuted coordnaton of nternet data centers under multregonal electrcty markets, n Proc. IEEE, vol., no., pp. 69-8, Jan.. [7]. L. Vandoorn, J. C. Vasquez, J. D. Koonng, J. M. Guerrero, and L. Vandevelde, Mcrogrds: herarchcal control and overvew of the control and reserve management strateges, IEEE Ind. Electron. Mag., vol. 7, no. 4, pp. 4-55, Dec. 3. [8] C. Yuen, A. Oudalov, and A. mbus, he provson of frequency control reserves from multple mcrogrds, IEEE rans. Ind. Electron., vol. 58, no., pp , Jan.. [9] E. M. Davdson, S. D. J. McArthur, J. R. McDonald,. Cummng, I. Watt, Applyng mult-agent system technology n practce: automated management and analyss of SCADA and dgtal fault recorder data, IEEE rans. Power Syst., vol., no., pp , Aug. 6. [3] J. A. Hossack, J. Menal, S. D. J. McArthur, J. R. McDonald, A multagent archtecture for protecton engneerng dagnostc assstance, IEEE rans. Power Syst., vol. 8, no., pp , May 3. [3] M. Wooldrdge, N. R. Jennngs, and D. Knny, he gaa methodology for agent-orented analyss and desgn, Journal of Autonomous Agents and Mult-Agent Systems, vol. 3, pp. 85-3,. [3] J. C. Vasquez, J. M. Guerrero, M. Savagheb, J. Eloy-Garca, R. eodorescu, Modelng, analyss, and desgn of statonary-reference-frame droop-controlled parallel three-phase voltage source nverters, IEEE rans Ind. Electron., vol. 6, no. 4, pp Apr. 3. [33] M. Cobotaru, R. eodorescu, F. Blaabjerg, A new sngle-phase PLL structure based on second order generalzed ntegrator, n Proc. 6 IEEE Power Electroncs Specalsts Conf., 6, pp. -6. [34] N. Pogaku, M. Prodanovc, and. C. Green, Modelng, analyss and testng of autonomous operaton of an nverter-based mcrogrd, IEEE rans. Power Electron., vol., no., pp , Mar. 7. [35] C. Godsl and G. Royle, Algebrac Graph heory. New York: Sprnger- Verlag,, vol. 7, Graduate exts n Mathematcs. [36] F. L. Lews and V. L. Syrmos, Optmal Control. New York: USA: Wley, 995. [37] IEEE Standard for nterconnectng dstrbuted resources wth electrc power system[s]. Amercan, IEEE commttee, 3 [38] L.-R. Chang-Chen, L. N. An,.-W. Ln, and W.-J. Lee, Incorporatng demand response wth spnnng reserve to realze an adaptve frequency restoraton plan for system contngences, IEEE rans. Smart Grd, vol. 3, no. 3, pp , Sep.. [39] A. Ahmad-Khatr, M. Bozorg, and R. Cherkaou, Probablstc spnnng reserve provson model n mult-control zone power system, IEEE rans. Power Syst., vol. 8, no. 3, pp , Aug. 3. [4] S. Lou, S. Lu, Y. Wu, and D. S. Krschen, Optmzng spnnng reserve requrement of power system wth carbon capture plants, IEEE rans. Power Syst., vol. pp, no. pp, -8, to be publshed. [4] A. L. Dmeas, N. D. Hatzargyrou, Agent based control of Vrtual Power Plants, n Proc. Int. Conf. Intellgent Syst. Applcatons to Power Syst. Jan. 7. [4] O. Palzban, K. Kauhanem, J. M. Guerrero, Mcrogrds n actve network management-part I: Herarchcal control, energy storage, vrtual power plants, and market partcpaton, Renewable and Sustanable Energy Revews, vol. 36, pp , Jan. 4. [43] U. Munz, A. Papachrstodoulou, F. Allgower, Consensus n multagent systems wth couplng delays and swtchng topology, IEEE rans. Autom. Control, vol. 56, no., pp , Dec.. [44] J. Qn, C. Yu, S. Hrche, Statonary consensus of asynchronous dscrete-tme second-order mult-agent systems under swtchng topology, IEEE rans. Industral Informatcs, vol. 8, no. 4, pp , Nov.. Quye Sun (M ) receved the B.S. degree n power system and ts automaton from the Northeast Danl Unversty of Chna, Jln Cty, Chna, n, the M.S. degree n power electroncs and drves, and the Ph.D. degree n control theory and control engneerng from the Northeastern Unversty, Shenyang, Chna, n 4 and 7, respectvely. Snce 4, he has been a Full Professor wth the School of Informaton Scence and Engneerng, Northeastern Unversty, Chna. Hs man research nterests are optmzaton and control of smart grd, and network control of dstrbuted generaton system, mcrogrds, and Energy Internet. He has authored and coauthored over 8 journal and conference papers, sx monographs and co-nvented 9 patents. Renke Han was born n Anshan, Laonng Provnce, Chna, n 99. He receved the B.S. degree n automaton from the Northeastern Unversty, Shenyang, Chna, n 3. He s currently workng toward the M.S. degree n control theory and control engneerng from the Northeastern Unversty, Shenyang, Chna. Hs current research nterests nclude network control, herarchcal and dstrbuted coordnated control, and power flow management strategy n mcrogrd and Energy Internet. Huaguang Zhang (M 3 SM 4 F 4) receved the B.S. degree and the M.S. degree n control engneerng from Northeast Danl Unversty of Chna, Jln Cty, Chna, n 98 and 985, respectvely. He receved the Ph.D. degree n thermal power engneerng and automaton from Southeast Unversty, Nanjng, Chna, n 99. He joned the Department of Automatc Control, Northeastern Unversty, Shenyang, Chna, n 99, as a Postdoctoral Fellow for two years. Snce 994, he has been a Professor and Head of the Insttute of Electrc Automaton, School of Informaton Scence and Engneerng, Northeastern Unversty, Shenyang, Chna. Hs man research nterests are fuzzy control, stochastc system control, neural networks based control, nonlnear control, and ther applcatons. He has authored and coauthored over 8 journal and conference papers, sx monographs and co-nvented 9 patents. Dr. Zhang s Char of the Adaptve Dynamc Programmng & Renforcement Learnng echncal Commttee on IEEE Computatonal Intellgence Socety. He s an Assocate Edtor of AUOMAICA, IEEE RANSACIONS ON

15 4 NEURAL NEWORKS, IEEE RANSACIONS ON CYBERNEICS, and NEUROCOMPUING, respectvely. He was an Assocate Edtor of IEEE RANSACIONS ON FUZZY SYSEMS (8-3). He was awarded the Outstandng Youth Scence Foundaton Award from the Natonal Natural Scence Foundaton Commttee of Chna n 3. He was named the Cheung Kong Scholar by the Educaton Mnstry of Chna n 5. He s a recpent of the IEEE ransactons on Neural Networks Outstandng Paper Award. Janguo Zhou was born n Yunnan, Chna, n 987. He receved the B.S. degree n automaton, and the M.S. degree n control theory and control engneerng from the Northeastern Unversty, Shenyang, Chna, n and 3, respectvely. He s currently workng toward the Ph.D. degree n control theory and control engneerng from the Northeastern Unversty, Shenyang, Chna. Hs current research nterests ncludes power electroncs, herarchcal and dstrbuted cooperatve control, and power qualty mprovement of mcrogrds, and synchronzaton of complex/mult-agent networks and ther applcatons n mcrogrds and Energy Internet. Josep M. Guerrero (S -M 4-SM 8-F 4) receved the B.S. degree n telecommuncatons engneerng, the M.S. degree n electroncs engneerng, and the Ph.D. degree n power electroncs from the echncal Unversty of Catalona, Barcelona, n 997, and 3, respectvely. Snce, he has been a Full Professor wth the Department of Energy echnology, Aalborg Unversty, Denmark, where he s responsble for the Mcrogrd Research Program. From he s a guest Professor at the Chnese Academy of Scence and the Nanjng Unversty of Aeronautcs and Astronautcs; and from 4 he s char Professor n Shandong Unversty. Hs research nterests s orented to dfferent mcrogrd aspects, ncludng power electroncs, dstrbuted energy-storage systems, herarchcal and cooperatve control, energy management systems, and optmzaton of mcrogrds and slanded mngrds. Prof. Guerrero s an Assocate Edtor for the IEEE RANSACIONS ON POWER ELECRONICS, the IEEE RANSACIONS ON INDUSRIAL ELECRONICS, and the IEEE Industral Electroncs Magazne, and an Edtor for the IEEE RANSACIONS on SMAR GRID. He has been Guest Edtor of the IEEE RANSACIONS ON POWER ELECRONICS Specal Issues: Power Electroncs for Wnd Energy Converson and Power Electroncs for Mcrogrds; the IEEE RANSACIONS ON INDUSRIAL ELECRONICS Specal Sectons: Unnterruptble Power Supples systems, Renewable Energy Systems, Dstrbuted Generaton and Mcrogrds, and Industral Applcatons and Implementaton Issues of the Kalman Flter; and the IEEE RANSACIONS on SMAR GRID Specal Issue on Smart DC Dstrbuton Systems. He was the char of the Renewable Energy Systems echncal Commttee of the IEEE Industral Electroncs Socety. In 4 he was awarded by homson Reuters as ISI Hghly Cted Researcher, and n 5 he was elevated as IEEE Fellow for contrbutons to dstrbuted power systems and mcrogrds.