A Distributed Algorithm for Improved Dependability a in Electricity Networks by Integrating Point-to-Point a Communication and Control

Publication REE REE 2005-8
OAI : oai:www.see.asso.fr:1301:2005-8:20226
contenu protégé  Document accessible sous conditions - vous devez vous connecter ou vous enregistrer pour accéder à ou acquérir ce document.
- Accès libre pour les ayants-droit


A Distributed Algorithm for Improved Dependability a in Electricity Networks by Integrating Point-to-Point a Communication and Control


1.77 Mo


Creative Commons Aucune (Tous droits réservés)
<resource  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xsi:schemaLocation="http://datacite.org/schema/kernel-4 http://schema.datacite.org/meta/kernel-4/metadata.xsd">
        <identifier identifierType="DOI">10.23723/1301:2005-8/20226</identifier><creators><creator><creatorName>Geert Deconinck</creatorName></creator><creator><creatorName>Jan van de Vyver</creatorName></creator><creator><creatorName>Adrian Dusa</creatorName></creator><creator><creatorName>Ronnie Belmans</creatorName></creator></creators><titles>
            <title>A Distributed Algorithm for Improved Dependability a in Electricity Networks by Integrating Point-to-Point a Communication and Control</title></titles>
        <resourceType resourceTypeGeneral="Text">Text</resourceType><dates>
	    <date dateType="Created">Wed 11 Oct 2017</date>
	    <date dateType="Updated">Wed 11 Oct 2017</date>
            <date dateType="Submitted">Tue 22 Jan 2019</date>
	    <alternateIdentifier alternateIdentifierType="bitstream">d9307330d6b8d3f348d47fc7beb077e65bc6054e</alternateIdentifier>
            <description descriptionType="Abstract"></description>

Dossier INFRASTRUCTURES CRITIQUES A Distributed Algorithm for Improved Dependability a in Electricity Networks by Integrating Point-to-Point a Communication and Control Mots clés Distributed Algorithm, Communication andControl, Dependability Par Geert DECONINCK, Jan VAN DEVYVER,Adrian DUSA, Ronnie BELMANS KULeuven - ESATIELECTA, Belgium Introduction (An earlier version of this work was presented at the CRIS conference on Securing Critical Infrastructures, [Vyve04]). The trend towards renewable energy poses a serious threat for the electricity infrastructure. The current infra- structure is designed to be controlled centrally. The elec- tricity network is semi-static : only occasionally new large nodes are added to the system (long-term planning). Classic generators are also quite predictable in terms of how much power will be produced. However, including renewable energy sources in the electricity infrastructure creates three problems. 'It yieids a dynamic network topology due to small scale of renewable energy sources, as generators can be added whenever the system requires it (evolution from long-term planning to short-term planning) . The unpredictability of some renewable energy sources (wind, photovoltaics) makes it difficult to plan beforehand which generator will produce how much energy ZD There is an increasing need for communication due to the larger number of nodes and the complex way in which they interact. Current centralised commu- nication methods are no longer sufficient. The liberalisation of the electricity market has increased the stress on generators and on the transport and distri- bution network. Moving the operating point closer to this critical state creates a less dependable system [Gene04]. The large scale power failures in the United States, Italy and London in the second half of 2003 are partly a consequence of shifting the operating point closer to its limits [UCTE03, PSOT03]. [HeydOl] presents a possible solution. This paper presents a distributed algorithm [Vyve03a, Vyve03b] that can be used to control the flow of electri- city. This algorithm avoids congestion, handles transient node failures and proves to be as scalable as the centrali- sed approach as far as communication is concerned. The algorithm is called emergent [Fish99] because nodes can only communicate with their immediate neighbours, but it still yields a global (system-wide) effect. ESSENTI SYNOPSIS L'objetdecetarticleestunalgorithme distribué, basésurdespro- tocolesde communication point-à-point, destinéà répondre aux nouvellesexigencesdes réseaux, tels que les changements rapidesde la consommationet de la production.Basé sur l'échangede messages,l'algorithmerépartitla chargede telle sortequelachargerelative de chaque noeudsoitapproximative- ment égale.L'algorithme et le nombrede messageséchangés est une fonctionlinéairedu nombrede noeuds.Cependant,la redondance de l'information rendle réseaumoinsvulnérable aux attaquesdirigées contreunnoeudet éliminelespointsde panne uniques. Lesfauteslocales sonttraitéesde façonadéquate. This paper presents a distributed algorithm, based on point-to- point communication, to deal with future challengesin networks (suchasdynamicchangesin consumptionandproduction).Based on messageexchanges,the algorithm distributes loadssuchthat the relative load of nodes is approximately equal. The algorithm and the number of exchanged message scale linearly with the numberof nodes. However,the distributedinformationexchange makes the network less vulnerable to directed attacks and eli- minates single points-of-failure. Local failures are handled ade- quately. REE No 8 Septembre 2005 A Distributed Algorithm for Improved Dependability in Electricity Networks... Model of the electricity network a) Topology In a simple model, the electricity network is divided into three levels : high, medium and low voltage. The transport network generally operates at high or medium level voltage. The distribution network operates at medium or low voltage. Such example of a small network is shown in figure 1. transport nctwork iiocle distribution network node t 6 ". Y i o 0 .... y 3 4 }' t " 1 le 9 s 8 7 Figure 1. Exaiiple topology of electricity network : iiode 1 and 4 are generators, node 2 and 3 are medium voltage sta- tions, iiodes 5 to 10 are consumers. In the transport network nodes are typically connec- ted in a mesh. In the distribution network nodes are connected radially or in a (generally open) ring. The lines in figure 1 are power fines ; the boxes and circles depict nodes. If there is an electrical connection between nodes, we assume that a point-to-point communication between these two nodes is also possible. b) Functions of nodes Let's assume that nodes have the following functions with respect to each other : supplier and/or client. The fol- lowing list presents the functions nodes can have and also some examples of each group : . supplier : generators (nuclear, CHP, coal, gas, rene- wable energy sources...) . supplier and client : transformer stations (high to medium voltage and medium to low voltage) . client : consumers. The first group is straightforward : generators can only provide electrical energy to the network. Transformer stations receive energy from one node and forward it to another node. Therefore they simultaneously are supplier and client. Most consumers are only clients. In the future we can expect more customers to install some renewable power source at home. These customers can sometimes be client, sometimes supplier, depending on their own electricity demand. c) Uncoupling The distributed algorithm assumes that there is uncoupling between nodes, i.e. the flow of resources can be controlled at the supplier side. For the electricity infra- structure this can be achieved by means of power elec- tronics, FACTS, etc. The inclusion of power electronics can also facilitate the control of reactive power needed to stabilise the electricity network. Nodes in figure 1 are uncoupled from their neighbouring nodes. In the current electricity infrastructure demand is controlled through voltage stabilisation. If the voltage drops, more current is injected. Within one node such a system can be present. If all connections of that node are uncoupled from the rest of the electricity network the dis- tributed algorithm as explained further can be applied. At the boundaries of this node, the demand can be measured and propagated to the higher levels in such a way consis- tent with the distributed algorithm. The possibility of having a hybrid system (partially controlled through voltage stabilisation and partially through the use of the distributed algorithm) makes it more likely to be used in the future. The author is well aware of the high price for implementing an electrical network controlled completely by the distributed algo- rithm (and the required investment in power electronics). In the future the prices for power electronics are likely to fall. Broader application areas for power electronics can also facilitate the decrease in price. Currently however the distributed algorithm is an interesting exercise to explore possible application domains. d) Communication Communication is required between clients and sup- pliers. In this paper we assume that there is a bidirectio- nal communication link available wherever required. How communication is established is outside of the scope of this paper. Algorithm The primary function of the algorithm is to deliver resources (e.g. electrical power) from suppliers to clients in a dependable manner. The algorithm also requires the following characteristics : . avoid electrical energy congestion at node level . handle nodes that are temporarily down . good scalability properties The proposed algorithm requires bidirectional com- munication between a supplier and all its clients. The clients will demand the required electrical power to their REE NO 8 Septembre 2005 Dossier INFRASTRUCTURES CRITIQUES suppliers so that the relative load on the suppliers is as equal as possible. The relative load (RR) of a node is the division of the load over the maximum allowed load or : P RR = clIrrenl p iii (I'VI/Jili/l The messages that are exchanged are the following : . generators send their load to their clients ; . clients send their demand to their suppliers ; . transformer stations send their demand to their suppliers and send their load to their clients. The demand that is sent by a client to a supplier is cal- culated locally in function of the following parameters : 1. total demand of that node 2. for every supplier : . current demand sent to that supplier 'total load of that supplier* . maximum allowed load of that supplier* The items without an asterisk are known locally and do not have to be received from the supplier. The items with an asterisk are not known locally, the client bas to wait for the data to arrive from the supplier. Every client calculates the demand to every supplier from these values. The sum of the demands to the sup- pliers of a client is equal to the total demand of the client. The demand values are chosen so that the relative loads of the suppliers are as equal as possible. The demand values are sent periodically 1 to the suppliers which can update their total load. The suppliers simultaneously send their total load and maximum allowed load to their clients (these values will be used in the next iteration step). Upon reception of the data of all suppliers, the client will calculate the values of the demand to every supplier and send these values to the respective suppliers. The values which were received from all suppliers in the previous iteration loop will be used to calculate the current demands to all suppliers of a client. Let us illustrate the algorithm with the example shown in figure 2. SI S2 ",P 12 Pli P' ? 1 , P-1 P 14' y t t Cl C2 Figvcre 2. Two clients and two suppliers. The system consists of four nodes. SI and S2 are sup- pliers, CI and C2 are consumers. The maximum allowed load and the load of the two suppliers are known (PSI,,,,,, and PS2,jjiax), along with the demands of CI and C2 (P and P The demand that is going to be sent over every connection (PII, P21, P12 and P) is then calculated as follows : 1) (1 + 1,) - (1) - (t) - P,,(1)+ P, (1 » + (t) - ( " 11 (1) --çi (I' 1 p é n il 1 > -il 'nU'o n/.\. n/.\ /.\. \ (l) + PÇ2...... (1) l (t) + pl2,,,, (1 2, (1 + t) (1) - (l] (1)- " 2 1 (t) + li1(1 » 1> 1 (1) - (Àl 1 (t) -'t2 (/) 1) Çl (1) + l. 2..... (1) ll, (t) + (1) CLn,nvf/+s :.n,u,O PSt.n,usltl+yc2,mns\l 1) (1) + l. 2..... (1) (t) + (1) (t) - J) 1, (1 + i,) = 1 1 f,., + 1> é,,, (l 1) (1) + (t) (t) + (t) 1), (1 + 1 () : 2 )' ('Il (1) -'> " (1) + l 2 (1 » l (1) (-) " (1) -'12 «) (t) + (1) (1) + I) s (1) In these equations, to is the length of an iteration step. The values for PI,, P21, P12 and P22 can be negative after these calculations (when there the difference in relative load between suppliers is large). The demand can physi- cally not be negative. The algorithm will adjust the demands so that they are not negative (the negative demand will be set to zero). The relative load will not be equal then, but as equal as possible, given the constraint. These calculations will also make sure that the maxi- mum allowed load of a node is never exceeded. This way security is ensured. This means that a node requesting more resources to a congested node will not get these resources. Currently it is the first node applying for resources (after the node was congested) that will be star- ved, but in the future the algorithm can be easily adjusted to add priorities. A more important node could then be satisfied while a less important node would be the first to be deprived of its required resources. The adding of prio- rities also creates the possibility of using different pricing schemes. A client can allow a certain amount of downti- me if they can purchase cheaper electricity. A company that can cope with a transient electrical failure (e.g. because they have their own generator or because their business allows a certain downtime) can then make an agreement with the electricity distributor that they can be shut off for some time each day. If this lowers their elec- tricity bills significantly, they will consider this possibility. a) Congestion avoidance When the load of a node is equal to the maximum From now on we will refer to the period in between the sending of messagesas the iteration step. REE No 8 Septembre 2005 A Distributed Algorithm for Improved Dependability in Electricity Networks... allowed load, the node is congested. Every iteration step the suppliers send their total load and their maximum allowed load to their clients. If the clients demand is unchanged and the relative load of the suppliers is equal, no action is undertaken. If the relative load of the sup- pliers is different, the load will be redistributed (through the calculation of the demands) so that the relative load equalizes again. Nodes that are under heavy load will thus be relieved. Congestion will never appear when there is enough room for redistributing power to another node 2. As an example the simulation results of the system from figure 3 with the inputs given in figure 4 are shown in figure 5. Sopplier L Stippliei 2 ('liclit 1 Client 1 1 \\44, Figure 3. Exarnple system to illustrate cor2gestion avoictavce. The first graph of figure 5 shows the demands of client 2 to supplier 1 and supplier 2. This graph illustrates how demands to both suppliers vary throughout the simu- lation. When the demand of client 1 is high, client 2 will demand more from supplier 2. When the demand of client 3 is high, client 2 will demand more from supplier 1. When both demands are evolving in the opposite direction (one goes up, the other goes down), client 2 is redistribu- ting power to keep the relative loads of supplier 1 and sup- plier 2 as equal as possible (between iterations stap 0 and 30, 90 and 120 and iteration steps 220 and 260). C enll \/ u \ ! o 5 50 100 150 200 250 300 Clien12 3ÛO --- \/ " o so,oo,sa zoo zso aoe Clieiil 2 400 300 300 -,------ \-/- .' 4 20D \/ " "./ 0 50 100 150 203 250 300 Clplit 3 300 " \.. zoo r 1, " \. \ /'\.' 200 \ -.-..- -' " \ o 5c 100 15G zoo 250 0 50 100 150 MO 250 300 300 - 200 100 û 100 0 1 5 1 - 05- Demands ofclient 2tosupplier 1and supplier 2 tosupplier 2 to supplier 1 50 100 150 200 250 RL ative ! oads of supplier 1 a [) d supplier 2 300 - RL supplier 1 o PL supplier 2 o 50 100 150 2UG 11, JUU o --so- ioa oso zao - Sum oflie total demands ofc [ents 1 , alld tue SLJI) L of the oacis si tiie supp les 1 2 soc --- - 600 400 ÉUU 1 aL sum otloads sum oidemands 50 100 150 zoo Figatre 4. Iiilgiits bisedfor client 1, client 2 and client 3. Figure 5. Siiiiulatioîz resliltsfor the systeiii infigure 3. The second graph of figure 5 represents the relative loads of supplier 1 and supplier 2. The graph shows that they are always equal except between iteration steps 120 and 175 and iteration steps 260 and 300. From figure 4 we notice that between the aforementioned times the demand from client 1 is quite high. Client 2 reacts by demanding all its power to supplier 2 (client 3 was alrea- dy demanding all its power to supplier 2). Even with all demand of client 2 directed to supplier 2, the relative loads of supplier 1 and supplier 2 cannot be equalised. Although supplier 1 is only congested between iteration steps 145 and 160, the algorithm will already redistribute power at time 90 (graph 1 in figure 5). If it would not, supplier 1 would already be congested before iteration step 145. The algorithm does not only avoid congestion when it appears but tries to avoid it beforehand. The rea- son why congestion could not be completely avoided bet- ween iteration step 145 and 160 is that the network is designed poorly. Supplier 1 can not provide the resources that are required by client 1. The third graph of figure 5 shows how the sum of the total loads of supplier 1 and supplier 2 tracks the sum of the demands of client 1, client 2 and client 3. Between iteration steps 30 and 70 and between iteration steps 255 and 300 there is a difference between the two graphs. The sum of the total loads of supplier 1 and supplier 2 is bounded by the sum of their maximum allowed loads (respectively 200 and 350). When the sum of the demands of client 1, client 2 and client 3 exceeds this physical limit tracking is not possible. b) Transient failure of nodes When a supplier is temporarily unavailable due to maintenance the algorithm will make the node sent a 2 If the nodesare often close to congestion or congested, this means that the network itself was badly designed. Only a redesign or the installation of additional capacity can solve the problem. REE No 8 Septembre 2005 Dossier INFRASTRUCTURES CRITIQUES maximum allowed load of zero. Thus the clients of this node will no longer request resources. If maintenance has finished the maximum allowed load can go back to the previous value (or a higher one if e.g. an upgrade was done). The clients of that node will automatically detect this and reroute power through that node again. When an internal failure is detected at the node, then the same action can be taken (setting the maximum allo- wed load to zero). The clients of this node will detect a decreased or zero maximum allowed load. The clients will then request their resources to another node. c) Scalability The algorithm only requires local communication between nodes. A node communicates only with its clients and its suppliers. In practice there is an upper limit to the number of clients and suppliers : the maximum number of clients and suppliers to a node are: Ms and Mc In our architecture there are three types of nodes : consumers, transformer stations and generators. The generators only have clients and the consumers only have suppliers, the maximum number of clients and suppliers respectively are: McJ !,and Ms .. The transformer stations have both clients and suppliers. The maximum number of clients and suppliers are Mc_t and M, 1. Nodes send the necessary data to their clients and suppliers at every iteration step. The number of genera- tors, transformer stations and consumers is Ng, N, and Ne Every interval the maximum number of sent messages is: Ng.M (._g + Nc.Ms -c + Nt. (Mu + Mu). The first term comes from the generators sending their load and maxi- mum allowed load to their clients. The second term refers to consumers sending their demand to their suppliers. The third term refers to the transformer stations sending their load and maximum allowed load to their clients. The fourth term refers to the transformer station sending their demand to their suppliers. If one generator is added, the amount of extra mes- sages is bounded by 2.M,__g. If a transformer station is added the amount of extra messages is bounded by 2. (Me t +Al-t). If one consumer is added the amount of extra mes- sages is bounded by 2.Ms.. If we define moveial, as the maximum of 2.M,-g, 2. (Mc_t + Ms -) and 2.M., -,,, then the amount of extra messages when one node is added to the system never exceeds When we define N,,,e,,,, as Ng + Ne + NI then the amount of messages per time inter- val never exceeds This formula indicates a linear increase of the number of messages with the number of nodes in the system or the system is 0 (n). Compared to a centrally controlled system, the total number of messages per time interval is also a linear function of the total number of nodes. The distributed algorithm approach does not worsen the scalability of the controlling algorithm. Current approach Currently, controlling the electricity infrastructure is done centrally or pseudo centrally (not distributed at every node but a few of them, as a way to improve depen- dability through redundancy). A centrally controlled sys- tem can be operated close to allowed tolerances since one node knows the entire system state and can calculate the best state for the system (cost, usage, load distribu- tion...). The amount of exchanged messages increases linear with the total number of nodes or is 0 (n). A major drawback of the centralized approach is that the central command is a single-point-of-failure. If that node fails, a large part of the system (or the entire system) can fail. The system is vulnerable to directed attacks by outside sources. Eliminating this single-point-of-failure eliminates this vulnerability and raises the overall depen- dability of the system. If communication is done in an efficient manner a centrally controlled system is able to avoid congestion quite well (because of the view of the entire system). Communication is also very important in case of an emergency. Several reports on the power failures in the U.S. and Italy [Gene04, UCTE03, PSOT03] show that the human factor (and especially interpersonal communi- cation) plays a very important role. A distributed algo- rithm would make communication an integral part of the control strategy. This would eliminate human misunders- tandings and hesitation and raise overall dependability of the electricity infrastructure. Conclusion The distributed algorithm described in this paper can handle the flow of electricity in the aforementioned cases. Congestion is avoided effectively whenever pos- sible. The algorithm can also handle transient failures of nodes. There is the possibility of including priorities so that more diverse pricing schemes can be used. The algo- rithm scales linearly with the number of nodes which is similar to the centralized approach. Communication and exchanging information between nodes is an integral part of the algorithm, it can be done in an efficient manner. Automation of communication eliminates (mostly human) misunderstandings. The distributed nature of the control strategy makes the network less vulnerable to directed attacks. The single point-of-failure is eliminated. In future work, the impact of network failures on the operation of the electricity network will be investigated in more detail. For this a model has been created of the electricity network (based on the Belgian and Dutch net- REE NO 8 Septembre 2005 A Distributed Algorithm for Improved Dependability in Electricity Networks... work). Simulations on limited systems have been done in Matlab/Simulink to verify the proposed algorithm. References [Gene04] The Geneva Association. " Vulnerabilities and Criticalities of Technical and Organisational Systems in the New Service Economy ". Working paper series of the Geneva asso- ciation. Etudes et Dossiers no. 280, special issue, Mar. 2004. fUCTE031 UCTE. " Interim Report of the Investigation Committee on the 28 September 2003 Blackout in Italy ", Oct. 2003. [PSOT031 The U.S.-Canada Power System Outage Task Force. " Interim Report : Causes of the August 14th Blackout in the United States anc/Canada ". Nov. 2003. [Heyd01] G.T. HEYDT, C.C. LIU, A.G. PHADKE, V. VITTAL. " Solutions for the Crisis in Electric Power Supply ". IEEE Computer Applications in Power, pp.22-30, July 2001 [Vyve03a] J. VAN DE VYVER, G. DECONINCK, R. BELMANS. " Using a Distributed Architecture to Tolerate Interdependencies in the Electricity Infrastructure ". 10th European Conference on Power Electronics and Applications (EPE 2003), Toulouse, France, Sep. 2003, 8 pages. [Vyve03bi J. VAN DE VYVER, G. DECONINCK, R. BELMANS. " The Need for a Distributed Algorithm for Contrai of the Electrical Power Infrastructure ". IEEE international Symposium on Computational Intelligence for Measurement Systems and Applications (C ! MSA), Lugano, Switzerland, Jul 2003, pp. 211- 215. [Fish991 D. A. FISHER, H. F. LIPSON. " EmergentAlgorithms-A new Method for Enhancing Survivability in Unbounded Systems ". Proc. 32nd Annual Hawaii International Conference on System Sciences (HICSS-32), Maui, Hi, Jan. 1999. [Vyve04] J. VAN DE VYVER, G. DECONINCK, R. BELMANS, " Improving Dependability in the Electricity Network by Integratin_q Communication in the Control Strategy Using a Oistributed Algorithm " Proc. 2nd Int. Conf. on Critical Infrastructures (CRIS- 2004), Grenoble, France, Oct. 2004, 6 pages. LI a u mu Geert Deconinck is Associate Professor at the K. U Leuven (Belgium) since 2003. He is a staff member of the research group ELECTA (Electrical Energy and Computing Architectures) of the Department of Electrical Engineering (ESAT). His research interests include the design and assessment of software-based solutions to meet dependability, real-time, and cost constraints for embedded applications on distributed systems. He received his M.Sc. in Electrical Englneering and his Ph.D. ln Engineering from the KU Leuven in 1991 and 1996 respectiveiy. He was a visiting professor at the K.U.Leuven since 1999 and a postdoc- toral fellow of the Fund for Scientific Research - Flanders (Belgium) in the period 1997-2003. He is a member of the Royal Flemish Engineering Society, a senior member of the IEEE and of the IEEE Reliability, Computer and Power Engineering Societies. Jan Van de Vyver was researcher at the K.U. Leuven (Belgium) in 2001-2004. He has a M.Sc. in Electrical Engineering from K.U.Leuven (2001) and is a student member of IEEE. He is cur- rently with the company Fluxys. Adrian Dusa is Ph.D. student at K.U. Leuven (Betgium) since 2001. He is a staff member of the research group ELECTA (Electrical Energy and Computing Architectures) of the Department of Electrical Engineering (ESAT). He graduated the Computer Science Engineering program at the Technical University Cluj-Napoca, (Romania) in 2000. He received hls Master of Engineering in Elecirical Engineering from the K.U. Leuven, Belgium in 2001, He is a student member of IEEE. Ronnie Belmans received the M.S. degree in electrical engi- neenng in 1979 and the Ph.D. in 1984, both from the Katholieke Universiteit Leuven, Belgium, the special Doctorate in 1989 and the Habilitierung in 1993, both from the RWTH Aachen, Germany. Currently, he is a full professor with the K. U. Leuven, teaching electrical energy systems including electrical machines. His research interests include electrical energy trans- mission systems and the implications of the liberalised market, renewable energy systems, distributed power and storage. He is visiting professor at Imperial College, London. Currently, he is also chairman of the board of Elia, the Belgian transmission sys- tem operator. Dr. Belmans is a member of the ! EE (U.K.). thé International Compumag Society and the Koninklijke Vlaamse Ingenieursvereniging jKV ! V). From 1 January 2005 he has been elected Fellow of the IEEE. REE NO 8 Septembre2005