Power Flow Management in Active Networks

Transcription

1 1 Power Flow Management n Actve Networks P. H. Nguyen, W. L. Klng, Member, IEEE, and J. M. A. Myrzk Abstract--Ths paper proposes a new method to manage the actve power n the dstrbuton systems, a functon under the framework of the actve network (AN) concept. An applcaton of the graph theory s ntroduced to cope wth the optmal power generaton (DGs/Cells dspatch) and nterarea power flows. The algorthm s mplemented n a dstrbuted way supported by the mult-agent system (MAS) technology. Smulatons show how the method works n cases of optmal operaton, congeston management, and power generaton cost change. Ind Terms--actve networks; dstrbuted generator; multagent system; graph theory AN DG FACTS IPR JADE MAS PFC G(V,E) A c j r j π u j α β γ I. NOMENCLATURE Actve network Dstrbuted generaton Flble AC transmsson systems Intellgent power router Java agent development framework Mult-agent system Power flow controller Drected graph model Agent Cost of edge (,j) Resdual capacty of edge (,j) Potental of node Capacty of transmsson lne Power generaton cost Transmsson cost Load prorty cost II. INTRODUCTION HE term of actve network (AN) has been ntroduced for a Tdstrbuton system to adapt to the hgh penetraton of dstrbuted generaton (DG) [1]. Wth one more control layer, each local area network can be defned as a cell whch can manage power nsde and across cell s boundares. By dong ths, the power flow can be controlled n an effcent, flble and ntellgent way n order to overcome problems of stng dstrbuton systems. As the AN mght be meshed, mult-agent system (MAS) technology can be appled for managng autonomous control actons and coordnaton amongs cells. Wthn a cell, actve elements, for ample controllable generators and loads, wll be represented by agents (software enttes) that can operate Ths work s part of the research project Electrcal Infrastructure of the Future (Elektrsche Infrastructuur van de toekomst, n Dutch), sponsored by the Mnstry of Economc Affars of the Netherlands. The authors are wth the Department of Electrcal Engneerng, Endhoven Unversty of Technology, 5600MB Endhoven, the Netherlands (e-mal p.nguyen.hong@tue.nl; w.l.klng@tue.nl; j.m.a.myrzk@tue.nl. HV/MV ~ Mult Agent System (MAS) Platform ~ Agent ~ ~ ~ ~ Local control area (Cell) Fg. 1. Actve Network managed by mult agent system autonomously wth local targets or cooperate wth others to acheve area tasks. A superor agent s nstalled for each cell as a moderator to manage autonomous actons as well as to communcate wth other cells. Ths control archtecture s llustrated n Fg.1. In case of more than one power supply path, the nterconnecton of cells allows for power flowng through alternatve paths when certan paths are over-stressed. Thus, the AN can avod congestons when the power flow s controlled n flble way. However, a domno effect mght occur when a system falure n one part of the network can quckly spread out over the rest [1]. Therefore, desgnng the control layer for the AN needs to concern about optmzaton and securty handlng at the same tme. In ths paper, the above mentoned power flow control functon wll be presented n detal. A dstrbuted control scheme wll be proposed for each cell to adapt to varatons n the system. Power flow controllng (PFC) devces are used as nterfaces of the cells n the AN. The power routng s optmzed by a successve shortest path algorthm, an applcaton of graph theory. III. ACTIVE POWER FLOW MANAGEMENT A. Problems As stated, ncreasng nterconnecton among cells of the AN can avod bottlenecks and can mprove system relablty

2 2 and stablty. However, these meshed networks mght get n troubles wthout approprate control mechansms. The power flow n an electrcal dstrbuton network s bound to physcal laws. The passve power transmsson can easly cause congeston on low mpedance components. Another undesred ssue caused by parallel nterconnectons s the loop flow whch s defned as a flow through a network part not meant to supply local loads. Ths unntended flow can lmt power transacton schedules and ncreases power losses n the network nvolved. Along wth a large-scale mplementaton of DGs, the power flow wll gradually change from an undrectonal to a bdrectonal stream. In addton, DGs power output s fluctuatng and s hardly predctable. These uncertan characterstcs cause also operatonal problems, such as too large voltage devatons. B. Soluton revew The most popular method for controllng the network, usng the optmal power flow (OPF), s a centralzed soluton that affects the overall network. It s normally deployed at the economc dspatch stage to fnd out the optmal operaton state of the network wth respect to system constrants. The mathematcal model of the OPF problem can be presented as follows mn subject to f ( x, u) g( x, u) = 0 h( x, u) 0 where f(x,u) s the objectve functon that can be adjusted to deal wth dfferent purposes,.e., power producton cost or power loss mnmzaton. The vector of ndependent varables u presents for the state of the system, the phase angles and load bus voltages. The vector of dependent varables x presents the control varables, for ample, power generatons or tap ratos of OLTC transformers. The equalty constrant represents the power balance between supply and demand whle the nequalty constrant shows the operatonal lmts of network components. OPF requres a large-scale control overvew that s mpossble to deploy n the dstrbuton networks such as the AN. To overcome ths dsadvantage, dstrbuted OPF technques have been proposed recently [2]. However, they stll need compl nput nformaton and take relatvely long tme processng. Prce-based control can also be consdered as a dstrbuted OPF soluton. By convertng the power system parameters nto desred market sgnals, the soluton yelds nodal prces for generators that can not only deal wth congeston problem but also contrbute to other ancllary servces [3]. Ths can be presented n a mathematcal model as follows mn subject to = 1 f ( P, P P P P A A A g ( P, A ) 0 n, A, A load req = 0 0 where f ( P, P, A, A ) s the aggregated cost functon of an AN ; the equalty constrant represents for power balance; the upper bound condton denotes requrements of ancllary servces whle the lower bound condton shows the operatonal lmts of network components. In hgh voltage networks, Flble AC transmsson (FACTS) s one of the effectve means that can regulate power flows ndependently [4]. FACTS elements are categorzed nto shunt compensaton (SVC, STATCOM), seres compensaton (TCSC), and hybrd compensaton (UPFC). Regardng the dstrbuton network havng a hgh R/X rato, power electronc seres devces such as TCSC or UPFC can work effectvely [5]. Also n [5], the concept of an ntellgent node s proposed as a seres controller that connects feeders based on electronc nterfaces, such as back-to-back converters. These devces can be used to control the power flow and to lmt voltage devatons, leadng to ncreased utlzaton of network components and hgher DG penetraton possbltes. However, the nfluence of FACTS devces s just n a lmted area of the system. To obtan an optmal mpact, t s necessary to coordnate wth other controllable components of the system. Recently, the concept of an Intellgent Power Router (IPR) s proposed as a new functon n power delvery systems [6]. By connectng to generators, power lnes, and customers, an IPR not only observes the current network condton but also cooperates wth others to fnd alternatve power flow paths n necessary cases. Ths approach s qute smlar wth the deas of the nterconnecton of ANs. However, the objectve functon for makng decsons s just on mnmzng load sheddng whle satsfyng the operatng constrants. Ths smple algorthm can not reach the optmal operaton of the compl system. The applcaton of FACTS devces for control purposes makes the performance much better. C. Proposed technque Ths secton proposes a soluton based on the applcaton of graph theory and the use of power flow controllers (PFC). The method s mplemented wth support of mult-agent (MAS) technology, whch s mentoned n desgnng the AN. In general, the power flow control can be formulated mathematcally as an optmzaton problem ncludng equalty and nequalty constrants as follows. Objectve functon s mn αδpg + βδpt + γ ΔPl (1) S T D )

3 3 subject to S T ΔPl D ( T ) ΔP = ΔP + (2) g loss Pt + ΔPt Pt max (3) where, Δ Pg, ΔPt, ΔP Present a change n power generaton, l transmsson and load. α, β, γ Label the costs for producton, relablty and load prorty. Δ Gves the power loss on component. P loss t, Pt max P Are the avalable power and capacty lmt of component. S, T, D Defne the supply, transmsson and demand area sets. The objectve functon of equaton (1) s the total cost for power delvery from the generaton areas to the load parts. It reflects overall economc dspatch regardng the securty of the transmsson components and load prorty. The equalty constrant (2) represents the power balance condton. The nequalty constrant (3) represents physcal operatng lmts. Ths optmzaton can be solved n a dstrbuted way by the applcaton of graph theory. The power system, frstly, s converted to a graph G(V,E), where V presents for the set of vertces (cells n the AN) and E presents for edges (nterconnecton lnes among cells n the AN). The edge length (edge cost) c j and resdual (avalable) capacty r j assocated wth each edge (,j) s derved from the transmsson cost β and the transmsson lne capacty u j. Two vertces are added a vrtual source node (s) and a snk node (t). For each cell wth generaton, a source edge (s,) s added wth resdual capacty r s (cell generaton avalable) and cost c s (cell producton cost α ). For each cell j wth load, a snk edge (j,t) s added wth resdual capacty r jt (cell load demand) and cost c t (cell load prorty cost γ ). In the graph model, the power flow optmzaton can be defned as a mnmum cost flow problem that regards to both the shortest path (economy) [8] and the maxmum flow (capacty) [9]. A smple and effectve soluton to solve the mnmum cost flow problem s the successve shortest path algorthm [7]. A node potental π s assocated wth each vert of the graph G(V,E). The source node potental s frstly set as 0. The algorthm starts updatng the other node potentals untl they satsfy the shortest path optmalty condton π j π + r j ; for all (, j) E (4) After updatng the node potentals of all vertes, the shortest path s gettng out by trackng edges from t backward s. The algorthm then augments the flow along the shortest path from s forward t untl reachng the capacty of at least one edge. After updatng the flow, t fnds another shortest path and augments the flow agan. The algorthm s ended when there s no possble path from s to t. An ample of a 5 cell system s shown n Fg.2. The graph model of the system s shown n Fg.3. The edges among cells represent nterconnecton lnes wth assocated the P gmax1 = 15MW, α 1 = 7 P load1 = 5MW, γ 1 = 1 u 12 = 7 u 23 = β 12 = 1 β 23 = 2 7,15 2, ,7 2, ,5 β 14 = 3 β 24 = 1 β 35 = 1 u 14 = 5 u 24 = 5 u 35 = 8 P gmax4 = 0 MW, α 4 = 0 P load4 = 5MW, γ 4 = 2 3,5 1,10 s 2,5 4 5 t P gmax2 = 10MW, α 2 = 2 P load2 = 10MW, γ 2 = 1 Fg. 2. An ample of the Actve Network. 0 1,5 3,18 transmsson cost (β ) and the transmsson lne capacty (u j ). Three drected edges from s to node 1, 2, and 3 represents generaton of cell 1, cell 2, and cell 3, respectvely. Assocated numbers of these edges are cell s power generaton cost α and power generaton capacty P gmax. Fve drected edges from 5 nodes to t represents load demand of each cell, respectvely. Assocated numbers of these edges are cell s load prorty cost γ and load demand P load. A detal mplementaton of the above algorthm for ths ample wll be presented n the nt secton. D. Dstrbuted mplementaton The man dea of the dstrbuted approach s controllng the power flow based on a so-called power router system. The power router s a combnaton of an agent (software) and a power flow controller (hardware). An llustraton of ths confguaton s shown n Fg. 4. The agent, n ths case, s the moderator A of each cell. It can get local area nformaton such as the power flow on ncomng (outgong) feeders, power generaton reserve, power load demand, and costs of producton and load prorty. Besdes managng autonomous control actons, ths agent can route message to communcate wth the same level agents. 1, ,5-3 u 45 = β 45 = 2 1,10 2,5 P gmax5 = 0 MW, α 5 = 0 P load5 = 5MW, γ 5 = 2 π c j, r j π c j, r j Fg.3. Augmentng power flow along the shortest path P gmax3 = 18MW, α 3 = 3 P load3 = 10MW, γ 3 = 1 j π j Node Potental Edge cost and capacty Power flow Shortest path

4 4 Two addtonal agents, A s and A t, are created to represent the source node s and the snk node t of the graph G(V,E). The PFC mght be the applcaton of several electronc devces,.e., converters or an ntellgent node [5], that s used to control the power flow for ts feeder based on the set pont gven by the moderator. Followng up the above ample of the 5 cell system, as the source node has potental π s = 0, A s sends ts nformaton to the neghbors (A 1, A 2, and A 3 ). Ther nodes potental are updated regardng the condton (4) wth the receved nformaton π s and edge cost c s. The potentals of A 1, A 2, and A 3 are then updated as -7, -2, and -3, respectvely. Although A 2 receves two addtonal messages from A 1 and A 3 due to ncomng lnes 1-2 and 1-3, π 2 s stll kept as -2 because t satsfes (4). After gettng all the messages, A t dentfes the shortest path accordng to ts potental π t. In ths case, the shortest path s s- 2-t wth the potental π t = -3. Then, A t backwards message to augment power flow. The augmentaton must be under the lmt of the shortest path capacty (10 MW). After gettng back the confrmaton message, A s s then lookng for another shortest path wth updated data. The procedure s completed when A s can not fnd any shortest path to A t. IV. SETTING-UP SIMULATION A. Electrcal Power System Model The above ample of the 5 cell system s smulated usng Matlab/Smulnk. Each cell (subsystem) s presented by a smplfed synchronous machne, local loads, and PFCs. For the loadng cell, the synchronous machne s replaced by an equvalence source. An Embedded Matlab Functon s created for each cell as part of the power router. Local nformaton about the subsystems s transferred through ths block for beng processed at the MAS platform. The block then receves control set ponts for the generaton and the PFCs. In ths research, the PFC model s derved from a seres part of the UPFC phasor model, whch belongs to SmPowerSystem toolbox of Smulnk [10]. The man objectve of ths model s to control the actve power flow wth respect to reference values gven by MAS. Through PI regulators, error values are transferred to the V d and V q components of voltage that are used as control sgnals to the seres converters. For smplcty, PFC uses a Current Source block nstead of real power electronc devces to control the power flow. B. Mult-Agent System Model MAS s created under the Java Agent Development Framework JADE [11]. JADE has recently been used as a popular platform for applcaton of MAS n power engneerng applcatons. It supports a Graphc User Interface and uses communcaton languages that follow the Foundaton for Intellgent Physcal Agents (FIPA) standard. In ths smulaton, each subsystem s managed by a par of the agents,.e., a socket proxy agent (spa) and a server agent (SA). Whle the spa agent s used as the communcaton agent wth Matlab/Smulnk, the SA agent s a prncpal agent that Cell ~ Power router Moderator MAS Platform Fg.4. Power router confguraton. = has all functons mentoned n the prevous secton. Two addtonal server agents, SA0 and SA6, are created to represent the vrtual source node s and snk node t of the graph. C. The Protocol The protocol for communcaton between Matlab/Smulnk and JADE s based on clent/server socket communcaton. The socket proxy agent n JADE s used as a server socket. By usng the TCP/UDP/IP Toolbox, each Embedded Matlab Functon n Matlab/Smulnk can create a clent socket to send data to and receve data for the spa agents. The communcaton tme s set at 0.5 sec. V. STUDY CASES = = PFC External grd A. Optmal operaton The 5 cell system shown n Fg.2 has been nvestgated to fnd out the optmal operaton. Table I presents varatons of the power flow and the consequent cost savng before and after applyng the control method. As can be seen from the table, a major part of total cost s saved from decreasng the power generaton n cell 1. Mtgatng the power flows on lne 1-4 also reduces sgnfcantly the transmsson cost. Therefore, the total flow costs (n money-based unt) before and after controllng are p.u and p.u, respectvely. The total cost savng s p.u. Dynamc behavour of the system when the proposed method starts workng s shown n Fg.5. At t = 5 s, each agent starts collectng and sharng nformaton across the MAS platform. At t = 10 s, new reference values are set for the generaton and the PFC devces The generators and PFC devces start controllng the power to reach new set ponts. The transent state occurs wthn around 10 sec and the system reaches a new optmal state.

5 5 TABLE I POWER FLOW VARIATION AND THE COST SAVING OPTIMAL OPERATION Before control After control From To Cost P Cell Cell g, Power P g, Power dff. MW flow, MW MW flow, MW Total cost dfference B. Congeston management To see the capablty of the method to cope wth congestons, the capacty of lne 3-5 s decreased from 8 MW to 4 MW. Although there s no change of generaton dspatch, the power flows are dfferent from the prevous case due to the restrcton of the lnes. Therefore, the total flow cost s hgher than prevous case ( p.u). The power flow varatons and transmsson cost changes are shown n Table II. The power flow n lne 3-5 reaches ts capacty of 4 MW. TABLE II POWER FLOW VARIATION AND THE COST SAVING CONGESTION MANAGEMENT Before control After control From To Cost P Cell Cell g, Power P g, Power dff. MW flow, MW MW flow, MW Total cost dfference C. Producton cost varaton Wth large-scale mplementaton of DGs n the dstrbuton networks, the producton costs wll fluctuate frequently. To see the capablty of the method to deal wth producton cost change, power generaton costs of cell 1, cell 2, and cell 3 are changed from 7, 2, and 3 to 3, 4, and 5, correspondngly. The dfference n producton costs establshes a new optmal operaton state of the network. Those varatons are presented n Table III. Wth new producton costs, the total flow costs before and after controllng are p.u and p.u, respectvely. Cost savng s accumulated manly from mtgatng the power flow on lne 1-4 and decreasng power generaton of cell 3. Pgen, MW Controlled power flow, MW tme, s 8 P14 P TABLE III POWER FLOW VARIATION AND THE COST SAVING PRODUCTION COST VARIATION Pgen - Cell 1 Pgen - Cell 2 Pgen - Cell tme, s Fg. 5. Power generaton and controlled power flow Before control After control From To Cost P Cell Cell g, Power P g, Power dff. MW flow, MW MW flow, MW Total cost dfference 9.97 VI. CONCLUSION Ths paper ntroduces the concept of Actve Networks as an effectve, flble and ntellgent soluton for the future. In ths respect, the functon of power flow management has been developed. Ths functon s mplemented n a dstrbuted way supported by the Mult-Agent System technology. The algorthm used for dstrbuted control comes from the applcaton of the graph theory. The smulatons show that the method can allow both the generaton and the PFC devces to operate optmally. Although the method s ntroduced as an applcaton for the Actve Network concept, ths technque can be used for systems on varous scales wth smlar structures. In partcular, t could be appled for the transmsson networks wth avalable FACTS devces. The drected graph model represents a power system wth a certan power flow drecton. It mght get bad condtons when

6 6 the power flow s changed drastcally. An applcaton of undrected graph model could mtgate ths problem. As usng a straghtforward algorthm of the graph theory, the number of messages followng among agents (the computaton tmes) s sgnfcant. Further study s needed reduce ths computaton burden. VII. REFERENCES [1] F. van Overbeeke, Actve networks Dstrbuton networks facltatng ntegraton of dstrbuted generaton, In Proc. of 2 nd nternatonal symposum on dstrbuted generaton power system and market aspects, Stockholm, [2] B.H. Km, and R. Baldck, A comparson of dstrbuted optmal power flow algorthms, IEEE Transacton on Power Systems, vol. 15, pp , [3] A. Jokc, Prce-based Optmal Control of Electrcal Power Systems, Phd dssertaton, Dept. Elect. Eng., Endhoven Unv., Endhoven, the Netherlands, [4] X.P. Zhang, C. Rehtanz, and B. Pal, Flble AC Transmsson Systems Modelng and Control, Sprnger, [5] R.d. Graaff, J.A.M. Myrzk, and W.L. Klng, Seres controllers n dstrbuton systems A survey of benefts n relaton to DG, In Proc. of Internatonal Conference on Future Power Systems, Amsterdam, the Netherlands, [6] I. J. Laurens, A decentralzed negotaton framework for restorng electrcal energy delvery networks wth Intellgent Power Routers IPRs, MS thess, Unversty of Puerto Rco, [7] R.K. Ahuja, T.L. Magnant, and J.B. Orln, Network flows theory, algorthm, and applcatons, Prentce-Hall Inc., [8] P. We, Y. Yan, Y. N, Y. Yen, and F.F. Wu, A decentralzed approach for optmal wholesale cross-border trade plannng usng mult-agent technology, IEEE Transacton on Power Systems, vol. 16, pp , [9] A. Amrbruster, M. Gosnell, B. McMlln, and M.L. Crow, Power transmsson control usng dstrbuted max-flow, In Proc. of 29 th Annual Internatonal Computer Software and Applcatons Conference, [10] SmPowerSystems Matlab/Smulnk, Unfed Power Flow Controller. [11] JADE Java Agent DEvelopment Framework [Onlne]. Avalable http//jade.tlab.com/. VIII. BIOGRAPHIES Phuong H. Nguyen was born n Hano, Vetnam n He receved hs M.Eng. n Electrcal Engneerng from the Asan Insttute of Technology, Thaland n From 2004 to 2006 he worked as a researcher at the Power Engneerng Consultng Company No. 1, Electrcty of Vetnam. In the end of 2006 he joned the Electrcal Power System Research group at Endhoven Unversty of Technology, the Netherlands as a Phd student. He s workng under the framework of the Electrcal Infrastructure of the Future project. Wl L. Klng (M 95) was born n Heesch, The Netherlands n He receved the M.Sc. degree n electrcal engneerng from the Endhoven Unversty of Technology, The Netherlands, n From 1978 to 1983 he worked wth Kema and from 1983 to 1998 wth Sep. Snce then he s wth TenneT, the Dutch Transmsson System Operator, as senor engneer for network plannng and network strategy. Snce 1993 he s a part-tme Professor at the Delft Unversty of Technology and snce 2000 he s also a part-tme Professor n the Electrc Power Systems Group at the Endhoven Unversty of Technology, The Netherlands. From December 2008 he s apponted as a full-tme professor and a char of EPS group at the Endhoven Unversty of Technology. He s leadng research programs on dstrbuted generaton, ntegraton of wnd power, network concepts and relablty. Mr. Klng s nvolved n scentfc organzatons such as Cgre and IEEE. He s the Dutch Representatve n the Cgre Study Commttee C6 Dstrbuton Systems and Dspersed Generaton. Johanna M.A. Myrzk was born n Darmstadt, Germany n She receved her MSc. n Electrcal Engneerng from the Darmstadt Unversty of Technology, Germany n From 1993 to 1995 she worked as a researcher at the Insttute for Solar Energy Supply Technology (ISET e.v.) n Kassel, Germany. In 1995 Mrs. Myrzk joned the Kassel Unversty, where she fnshed her PhD thess n the feld of solar nverter topologes n Snce 2000, Mrs. Myrzk s wth the Endhoven Unversty of Technology, the Netherlands. In 2002 she became an assstant professor and snce 2008 she s an assocate professor n the feld of resdental electrcal nfrastructure. Her felds of nterests are power electroncs, renewable energy, dstrbuted generaton, electrcal power supply.