A Primary and Backup Cooperative Protection System ... - IEEE Xplore

1222

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 21, NO. 3, JULY 2006

A Primary and Backup Cooperative Protection System Based on Wide Area Agents Renan Giovanini, Kenneth Hopkinson, Denis V. Coury, Member, IEEE, and James S. Thorp, Fellow, IEEE

Abstract—This paper presents a study of wide area agents based on communication for primary and backup coordinated protection. Agents are used to give each protection component control capacity as well as the ability to communicate with other agents. We feel that this method naturally points toward a new philosophy for primary and backup protection. Simulations are used to illustrate concepts, using a simulation engine called EPOCHS that combines the EMTDC/PSCAD power simulator with the NS2 network communications simulator. Results show the improved performance of our protection scheme. In this new protection system, agents were embedded in each of the conventional protection components to construct an IED relay. The agent searches for relevant information by communicating with other agents. Agent communication can take place at the same substation or at remote substations. This information can be used to detect primary and remote faults, relay misoperation, breaker failures, and to compensate such problems with a much better performance than can be done in traditional schemes. Preliminary results give us hope that the proposed protection scheme may be able to contribute toward the mitigation of wide-area disturbances and the power blackouts that frequently follow them. Index Terms—Agents, communication, cooperative systems, power system protection, relaying.

I. INTRODUCTION

D

URING the last decade, a new policy arose for many power system utilities around the world. The old monolithic system was dismantled into a decentralized model. As a result, these power systems are now being operated closer and closer to their limits. Problems such as transmission congestion, power fluctuations and smaller generation reserves are new drawbacks in this scenario. Moreover, faster, more reliable, and better coordinated protection and stability control are even more critical in this new environment than they have been in the past. New methods are needed to overcome this challenge.

Manuscript received September 8, 2005. The authors were supported in part by the University of São Paulo at São Carlos, in part by the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP), in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), and in part by Cornell University. The views expressed in this document are those of the authors and do not reflect the official policy or position of the United States Air Force, Department of Defense, or the U.S. Government. Paper no. TPWRD-00533-2005. R. Giovanini is with Operador Nacional do Sistema Elétrico (ONS), Rio de Janeiro, Brazil (Brazilian ISO) (e-mail: [email protected]). K. Hopkinson is with the Department of Electrical and Computer Engineering, Air Force Institute of Technology, Wright-Patterson Air Force Base, OH 45433-7765 USA (e-mail: [email protected]). D. V. Coury is with the Department of Electrical Engineering, University of São Paulo, São Carlos 13566-590, Brazil (e-mail: [email protected]). J. S. Thorp is with the Department of Electrical and Computer Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061-0202 USA (e-mail: [email protected]). Digital Object Identifier 10.1109/TPWRD.2006.876984

Traditional protection relays are based on standalone units. These relays make decisions based on their local inputs and data from remote units are rarely used in their internal logic. Communication plays a minor role in these systems. However, relay engineers are beginning to study and access the benefits of wide area communication. The success of the Internet has attracted the attention of the protection community. In particular, the Internet has shown the capabilities and advantages of IP based networks. With these advantages in mind, the power industry has started to consider this type of communication as a reliable way to improve the protection of the electric power grid. Faster responses, better coordination and increased correctness are all expected features from communication. Despite the great interest in the electric power community in protection systems involving communication, no software has offered the possibility to simulate IP networks for protection purposes. For this reason, a platform called electric power and communication synchronizing simulator (EPOCHS) was built. EPOCHS combines simulators from different domains (power systems and communication networks) in order to evaluate protection schemes based on communication. This paper presents the study of wide area agents based on communication for primary and backup coordinated protection. An IEEE base power system was used in conjunction with a fiber-optic Ethernet network. The simulations presented show the performance of these agents. Furthermore, aspects pertaining to power systems and communication networks are analyzed. Characteristics such as traffic congestion and link losses are considered, as well as agent and breaker failures. In all cases, the communication approach proves to be superior to the legacy system. II. AGENTS FOR PROTECTION AND CONTROL OF POWER SYSTEMS The electric power grid has traditionally been made up of a large number of protection and control devices that act on local information to respond to problems [1]. This approach works well in some cases, but is inefficient in others. Agents have begun to be recognized as a natural solution to this problem in the electric power research community. Their autonomous nature, ability to share information and coordinate actions, and the potential to be easily replaced from remote facilities make them potentially valuable [2]. Some practical applications of agents in power systems can be found in literature [3]–[7]. The protection and control scenarios that interest us use geographically distributed agents located in a number of Intelligent Electronic Devices (IEDs) as shown in Fig. 1. An IED is a hardware environment that has the necessary computational,

0885-8977/$20.00 © 2006 IEEE

GIOVANINI et al.: PRIMARY AND BACKUP COOPERATIVE PROTECTION SYSTEM

1223

between the old and new systems in this type of situation. As shown in Fig. 2, agents have the ability to communicate through a LAN in order to interact with other agents directly located on that same LAN, or can pass information along to the utility WAN, i.e. the utility Intranet, ultimately communicating with more remote IEDs. A. Structure of a Utility Communication Network

Fig. 1. Location of the agent-based IEDs within the utility intranet infrastructure.

Networked computing systems are becoming increasingly prevalent in many areas and we believe that this growth will occur within electric utility systems as well. Technology is constantly changing, but we can make some educated guesses about what utility communication systems will look like in the near future. First, the network systems will almost certainly be built from standard commercial off-the-shelf components. To do otherwise would be expensive both in terms of initial cost outlay and system maintenance. This means that these networks will be based on Internet standards even if the systems remained independent of the global network conglomeration. We can already see hints that such changes are coming due to recent standardization efforts such as the IEC 61 850. We believe fiber-optic Ethernet networks in conjunction with IP-based communication protocols will be heavily used in utility communication for these reasons. III. EPOCHS A. Overview

Fig. 2. Structure of an agent-based IED.

communication, and other I/O capabilities needed to support a software agent. An IED can be loaded with agents that can perform control and/or protection functionality. These agent-based IEDs work in an autonomous manner where they interact both with their environment and with each other. An example of this might be digital relays where each one has its own thread of local control, but they perceive a more global scope of the system and act in response to their nonlocal environment by communicating with other agents either via local area networks (LANs) or via wide area networks (WANs). The agent-based IEDs structure is depicted in Fig. 2. Agents within an IED perceive their environment through local sensors and act upon it through the IED’s actuators. Examples of sensor inputs might include local measurements of the current, voltage, and breaker status. Actuator outputs might include breaker trip signals, adjusting transformer tap settings, and switching signals in capacitor banks. Agents might even interface with systems such as supervisory control and data acquisition (SCADA) systems. The host computer shown in Fig. 1 could act as a bridge

EPOCHS is a distributed simulation platform that links commercial and high quality simulators through the use of a runtime infrastructure (RTI) to allow modelers to investigate electric power scenarios that involve network communication. EPOCHS seamlessly links simulation systems from a modeler’s perspective, enabling them to investigate power protection and control scenarios that combine communication with the ability to sense the state of a power system and to react to it in real-time [8]. EPOCHS is particularly valuable for evaluating the communication requirements of new protection and control schemes and the impact of common Internet behavior, such as traffic congestion, in a power system’s operation [9]. B. Architecture The EPOCHS system is shown in Fig. 3. It is composed of five main components. 1) PSCAD/EMTDC: It is used for electromagnetic transient simulation. EMTDC is a well-known electric power simulator produced by the Manitoba HVDC Research Centre [10]. 2) PSLF: It is an electromechanical transient simulation software used for stability studies. It is produced by General Electric [11]. 3) NS2: It is a communication network simulator that was created through a joint effort between the University of California at Berkeley, Lawrence Berkeley Labs, the University of Southern California, and Xerox PARC [12]. 4) AgentHQ: It is a module that we developed to present a unified environment to our agents and acts as a proxy for those agents when interacting with other EPOCHS

1224


the agents. For this system in particular, we assumed that these packets are one of the following five types of messages:

Fig. 3. Relationship between EPOCHS’s five components.

components. Through it, the agents can get and set power system values and send and receive messages to one another. 5) RTI: It acts as the “bond” between all other components. It is responsible for simulation synchronization and for routing communication between EPOCHS components. C. Component Interaction The synchronization between the various simulation components follows a straightforward algorithm. At the beginning of each time step, the RTI waits for synchronization messages from both the power system simulator and NS2. Once the messages are received, the RTI yields control to the AgentHQ. The AgentHQ allows each of the agents to execute. When they are activated, the agents are capable of sending communication messages and getting/setting power system variables. Once all agents are done, the AgentHQ returns the control back to the RTI. Finally, the RTI notifies both the NS2 and the power system simulator that the current time step has been completed. At this point, the two simulation engines run for an additional time step. Special attention must be paid to NS2. Messages may be received in between two synchronization points within NS2. If a message arrives, NS2 will immediately pass it along to the RTI bound for the AgentHQ. The AgentHQ will, in turn, pass the message on to the appropriate agent. The agent can process the message and send another in response. If the message requires power system state to be read or changed then the agent must keep the message in a queue until the next synchronization point occurs. IV. AGENT-BASED PROTECTION SYSTEM A. Message Structure Used by the Agents First, we must give attention to the type and the size (in bytes) of the transmitted messages on the network, as their values directly influence the total traffic bandwidth. The packet size is set through one of the network simulation parameters and is directly bound to the amount of information exchanged among

All messages share the first two fields as well as the last two. The first field is assigned to the message type, as it can be a command to trip a breaker (SET), a message containing the current phasors (GET_RESPONSE) or the three different kinds of trip (INTERTRIP, BACKUP_TRIP or NEIGHBOR_TRIP). The second field of all types of messages depicts the time the message was created. The last two fields are related to the UDP/IP protocol. It should be mentioned that the field before the last is the UDP header, whilst the last one is the IP header version 4. Message 1 is the command to open the breakers. Field #3 identifies the breaker ID, and field #4 comprises the command to be executed (OPEN or CLOSE). Message 2 is the main message used by the differential logic. It contains the values for current phasors A, B, and C, seen by the local agents. The last three types of messages correspond to the three different types of trip present in this system. The first field determines which type of trip it is, while the remaining fields have similar meanings to those present in Message 1. B. Protection Strategy In the current implementation, agents are responsible for transmission line protection. These agents receive information such as local voltages and currents from the local IED or acquire information by communicating with remote agents. Three types of agents were implemented: primary agents, backup agents, and load agents. Primary agents are responsible for first zone protection, covering 100% of the transmission line. Backup agents are responsible for the third zone protection, which covers the first zone plus all the transmission lines connected to the remote end of the first zone. Load agents are only responsible for sending their current state (usually their current phasors) to the backup agents. An agent can either receive the list of agents, which are in its protection zone and with which it will communicate, at initialization or it might learn this information through some type of network topology discovery algorithm. Primary and backup agents follow a differential philosophy to detect a fault. At every time-step, they read their local current


1225

TABLE I RULES FOR PRIMARY AND BACKUP AGENT BEHAVIOR

phasors and send this information to the agents in their protection zone. Once an agent receives the phasors from its protection zone remote end/ends, it calculates the differential current and decides whether a fault has occurred or not. After detecting a fault, the agents take action based on the preset rules depicted in Table I. As shown, if a primary agent detects a fault (rule 1), it sets its internal variable FaultStatus to DETECTED, sends an INTERTRIP to all agents in its primary agent list and starts a timer. As we can see, no action is taken to open the associated breaker. Traditional protection would open the breaker at this point, but since a communication network is available, the agent will wait for a message confirming the fault before performing this operation. If an INTERTRIP or BACKUP_TRIP is received, the primary agent will open its breaker (rule 6). On the other hand, if a INTERTRIP_RESPONSE = NEGATIVE is received (rule 8), it means the primary agent counterpart has not detected the fault, which might signal an agent misoperation. In this case, the agent disables its internal timer and it waits for another message. If a BACKUP_TRIP message is received, the agent follows rule 6, and opens the breaker. Finally, if, after detecting a fault, no message is received within 15 ms, the primary agent assumes that a communication problem might be occurring and triggers its breaker (rule 5). In all cases, after opening a breaker, the primary agent starts a second timer. This timer will then check to see if the breaker was really opened.

If a current is still flowing into the primary protection zone after 50 ms, the agent sends a NEIGHBOR_TRIP to all primary agents located at the same bus, in order for them to open their breakers (rule 2). A primary agent can receive an INTERTRIP, a BACKUP_TRIP or a NEIGHBOR_TRIP without having detected a fault. For example, if a primary agent receives the first two types of messages, it will assume that it is defective, and will open its breaker without detecting a fault by itself (rule 7). On the other hand, if a primary agent receives a NEIGHBOR_TRIP and a BACKUP_TRIP, it will infer that a problem has occurred with one of its primary agent neighbors, and will trip its breaker (rule 9). The backup agent sends a BACKUP_TRIP to all primary agents inside its backup protection zone and starts a timer (rule 3) if a fault is detected. If, after 100 ms of fault detection, a differential current is still present in its backup protection zone, it assumes all first zone relays have failed to clear the fault, and opens its breaker (rule 4).

V. UTILIZED POWER SYSTEM AND THE AGENT SYSTEM CONFIGURATION All tests conducted with the agent system were based on the well-known IEEE 14-bus system. The complete data set for this

1226

Fig. 4.


IEEE 14-bus system.

system can be found in [13]. The structure of the 14-bus system is illustrated in Fig. 4. All transmission lines were modeled based on the PI model of the line, and all sources were modeled as constant power sources. The communication links were created in parallel with the transmission lines, resulting in a communication system with almost the same topology as the power system grid. Since buses 5 and 6, and 4, 8 and 9 are transformers and not transmission lines, we assumed that buses inside the same substation define only one node in the communication system. All links were assumed to be full-duplex with a traversal time of 2 ms, and initially have a bandwidth of 100 Mb/s. All communication was based on the UDP/IP standard. This communication system can be seen in Fig. 5. In the following examples, we deployed 30 primary agents, 30 backup agents, and 16 load agents. All agents calculate the phasor currents seen on their terminals based on a moving window of one cycle with a sampling rate of 1 kHz. Additional effects such as anti-aliasing filters, current transformers and digital sampling were taken into account for all agents.

VI. CASE STUDIES Our case studies were divided into two different categories. First we show the communication network performance and the analysis of the network parameters. Later we present the agent system’s performance when a fault occurs during different situations. It is important to remember that the communication network and the agent system are linked to each other. A network parameter change directly influences the agent system and vice-versa. The design of the agent system and the communication network must be done together. If a parameter of one of them is changed, the other system must be re-evaluated in order to assure the correctness of the protection system.

Fig. 5.

Communication system.

A. Communication Tests The first characteristic we must concentrate on in an agent system based on communication is the communication network itself. Link failures can cause delays in packet transmissions or increase the traffic magnitude to unacceptable levels. Packet losses and disruptions to the communication process are common consequences in this scenario. The communication system must be designed in order to support the loss of one or more links, ensuring that the communication process will work even in abnormal situations. One possible solution would be to duplicate all communication links. However, even with this setup, substation blackouts could cause communication interruptions. Another solution is to reserve enough bandwidth in order to deal with situations where one or more links are inoperative, to allow the affected traffic to be rerouted to the sound links. The next two figures illustrate a case before and after a communication failure occurs. Fig. 6 presents a situation where the communication system is working correctly. In it, the bandwidth used is the least possible. During a normal situation, seven different connections go through link 2-3. For the sake of simplicity, only the traffic related to the primary protection of transmission lines 2-3, 2-4 and 3-4 are depicted. The total bandwidth used by the associated links (links 2-3, 3-2, 2-4, 4-2, 3-4 and 4-3) are shown in Table II. In the next figure (Fig. 7), links 2-3 and 3-2 are down. In this situation, the routers located at nodes 2 and 3 detect that these links are down, and through a standard routing algorithm, they rebuild their routing tables. After this procedure, if router 2 wants to send a message to router 3, it sends it through link 2-4 and router 4 forward it through link 4-3. The new traffic level is shown in Table II. As we can see, the system still works correctly because there is enough bandwidth on the links. Had this not been the case, the amount of traffic generated by the


1227

TABLE III TRAVERSAL TIME

Fig. 6. Communication system working correctly.

TABLE II TRAFFIC BANDWIDTH

Fig. 7.

Links 2-3 and 3-2 are down.

agents would have surpassed the link capacities, causing packet losses. From this example we can conclude two facts. First, if enough bandwidth is available, agents can communicate with each other even though one or more links (but not all of them) are down at

the same time at the same node. Second, if all links connected to a node are down, then communication becomes impossible, and the protection system must switch to a primary protection system based on a standalone technology. Another aspect of the communication system that merits attention is the traversal time. When a link goes down, it not only affects the bandwidth of the remaining links, but it also changes the time it takes for certain messages to go from one node to another. For example, in the previous case all messages that were routed through links 2-3 and 3-2 had to be rerouted after these links went down, as shown in Fig. 7. As shown in Table III, all messages in a normal situation take 2.007 ms to move between neighboring nodes. After the rerouting, messages from node 2 to 3 and from 3 to 2 take more than double the time. Messages from node 2 to 4, 3 to 4, 4 to 2 and from 4 to 3 are not affected. The cases above illustrate that the communication link capacities chosen can have a great impact on the operation of a protection system. In order to determine an ideal bandwidth, we initially assumed all links had a very high bandwidth (100 Mb/s) in order to avoid packet losses caused by buffer overflow originated by inoperative links. We followed the initial experiment by simulating 70 cases where one or more links were down. The bandwidth used after rerouting occurred in all links was calculated for all cases. One thing we should point out is the fact that since all links are full-duplex, the amount of traffic in link 1-2 is not the same as it is in link 2-1 and if link 1-2 goes down, its counterpart (link 2-1) also goes down. The simulated cases consisted of the following possible situations: • one link down; • two links down; • three links down; • all links down, but one; • all links down (substation communication blacked-out). During this set of simulations, we assumed that links connected to different nodes would never become inoperative at the same time. E.g., links 2-3 and 2-4 being inoperative is an example of 2 links down, while links 2-4 and 5-7 being inoperative is not part of the set of simulations. Table IV shows the maximum bandwidth in all links for all tests. As can be seen, the traffic volume during abnormal situations can reach values 16 times higher than in normal traffic, as denoted for link 7-6 in Table IV. This fact is very important when designing the communication network. Even if we had set all the links to hold up to 100% more traffic than the normal situation, we would experience a complete system failure, due to

1228


TABLE IV MAXIMUM TRAFFIC BANDWIDTH

traffic rerouting. The normal traffic for link 7-6 is 1.408 Mb/s and this amount goes up to 23.232 Mb/s when links 5-1, 5-2, 5-4, 5-8 and 5-9 are down. Based on Table IV, all links were set to 30 Mb/s for the next set of tests. As we can see, a detailed and careful design of the communication network must be implemented for applications with real-time requirements. Once the bandwidth of the links is defined, the next step is to analyze the maximum traversal time experienced in the network. As presented, a minimum bandwidth is necessary for the communication system to survive abnormal situations. However, having enough bandwidth is not sufficient if the network topology does not have the redundancy required to overcome abnormal situations. If after a failure, a message takes a much longer time to reach its destination, then the insertion of new links or link duplication should be considered. The 70 tests just described were repeated, but this time all links were set to 30 Mb/s. The traversal time between all neighbor nodes in the new set of tests was calculated and the results are shown in Table V. As can be observed, the traversal time for a specific link may increase up to 6 times for this communication system (link 1-5). It is suggested in the specialized literature that communication should be carried out in less than 25 ms, which leads us to conclude that this communication system’s topology is able to cope with real-time applications even in abnormal situations. Had it not been the case, alternative links would need to be added to the network in order to provide enough redundancy to ensure a small traversal time.

TABLE V MAXIMUM TRAVERSAL TIME

B. Protection Tests Our second category of tests are focused on the agent system itself. The next four tests show how the agent system behaves when a fault occurs. 1) Correct Primary Protection: In the first test, we present an example of primary protection when it is operating correctly. A three-phase fault occurs at 0.1500 s in the middle of transmission line 13-14 (Fig. 4). In this example, agents PRIM_13-14 and PRIM_14-13 detect a fault at 0.157 s and after 1.4 ms and 1.2 ms respectively, they receive an INTERTRIP message, which leads them to open their breakers (rule 6 of Table I). The sequence of events can be seen in Table VI. 2) Link Failure: In this case, we consider the same fault but a link failure occurs at 0.100 s, making link 9-10 inoperative (Fig. 5). At first glance, this could cause problems for the agents. However, through a standard dynamic routing algorithm, a new path is found. Messages that should go from node 9 to node 10, now go from nodes 10 to 4, from 4 to 5, from 5 to 9, and vice-versa. In this scenario the primary protection is delayed, but still works correctly. Initially, the fault is detected at 0.1590 s by agents PRIM_13-14 and PRIM_14-13. At 0.1592 s, agent PRIM_13-14 receives a BACKUP_TRIP and opens its breaker (rule 6 of Table I). The same happens to PRIM_14-13 at 0.1594 s. The sequence of events can be seen in Table VII. 3) Agent PRIM_13-14 Fails: In the third case, a test is performed where agent PRIM_13-14 fails to detect a fault in line


TABLE VI SEQUENCE OF EVENTS FOR CASE 1

1229

TABLE VIII SEQUENCE OF EVENTS FOR CASE 3

TABLE VII SEQUENCE OF EVENTS FOR CASE 2

TABLE IX SEQUENCE OF EVENTS FOR CASE 4

13-14. As can be seen, despite being incapable of detecting the fault, the agent system still works correctly. At first, the agent PRIM_13-14 receives an INTERTRIP at 0.1584 s. Later on, it receives a BACKUP_TRIP at 0.1595 s and finally opens its breaker (rule 7 of Table I). Agent PRIM_14-13 detects the fault at 0.1590 s, receives a BACKUP_TRIP at 0.1592 s and finally opens its breaker (rule 6 of Table I). This example shows the strength of a cooperative system based on communication. Despite having a defective agent, the agent system was able to detect and precisely eliminate the fault only 0.6 ms later than case 1, where no problems occurred. The sequence of events can be seen in Table VIII. 4) Breaker Failure: In the final test, we present a breaker failure, where the associated breaker to agent PRIM_13-14 refuses to open. As shown in Table IX, the sequence of events for agents PRIM_13-14 and PRIM_14-13 are exactly the same as case 1 up to 0.1584 s. At 0.2070 s, agent PRIM_13-14 realizes there is still a current flowing through its breaker, which leads it to contact its primary agent neighbors to open their breakers (rule 2 of Table I). Agents PRIM_13-06 and PRIM_13-12 receive this message almost instantaneously, and trip their breakers at 0.2070 s (rule 9 of Table I). VII. CONCLUSIONS In this paper we have described the use of wide area agents for primary and backup protection. Initially, we defined the concept of agents for power systems, and pointed out that one of their main strengths is their ability to communicate. We believe that power utilities will have private communication

networks as depicted in this paper in the near future. Based on this, we defined a utility intranet on top of a power grid. This utility intranet is based on TCP/IP and UDP/IP standards. To analyze the agent technology for power system protection, EPOCHS, a platform that integrates a power system simulator (PSCAD), and a network communication simulator (NS2) was created and implemented. Initially, our sets of tests concentrated on the communication network. The network parameters (bandwidth and traversal time) were analyzed in order to evaluate the network performance when different conditions of link failures took place. Next, we analyzed the agent system behavior and its performance for different situations. Our first two tests showed how agents can perform a primary protection

1230


scheme by exchanging basic information. As depicted in case 2, even though a communication link failure exists, a primary protection scheme based on a utility intranet can perform its goal successfully. The last two examples showed how agents can cooperate through communication to overcome different kinds of failures. In these cases, the ability to communicate shows the power of agents over traditional systems. In all cases, the agent approach proved to be faster and more reliable than the traditional standalone alternatives. ACKNOWLEDGMENT The authors would like to thank the Department of Electrical Engineering—Engineering School of São Carlos, University of São Paulo (Brazil), and the School of Electrical Engineering, Cornell University (USA) for research facilities provided to conduct this research project. Our thanks also to the financial support given by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) and Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP). REFERENCES [1] A. G. Phadke and J. S. Thorp, Computer Relaying for Power Systems. Baldock, Hertfordshire, U.K.: New Research Studies, 1988. [2] M. Genesereth and S. Ketchpel, “Software agents,” Commun. ACM, vol. 37, pp. 48–52, July 1994. [3] Y. Tomita, C. Fukui, and H. Kudo, “A cooperative protection system with an agent model,” IEEE Trans. Power Del., vol. 13, no. 4, pp. 1060–1066, Oct. 1998. [4] D. V. Coury, J. S. Thorp, K. M. Hopkinson, and K. P. Birman, “An agent based current differential relay for use with a utility intranet,” IEEE Trans. Power Del., vol. 17, no. 1, p. 47, Jan. 2002. [5] S. K. Wong and A. Kalam, An Agent Approach to Designing Protection System, IEE Conf. Publ. 434, pp. 373–376, 1997. [6] C. Rehtanz, Autonomous Systems and Intelligent Agents in Power System Control and Operation. New York: Springer, 2003. [7] T. Hiyama, H. Esaki, and T. Funabashi, “Experimental studies on multiagent based GC for isolated power system with dispersed power source,” Eng. Intell. Syst. Elect. Eng. Commun., vol. 13, no. 2, pp. 135–140, Jun. 2005. [8] K. M. Hopkinson, R. Giovanini, X. Wang, J. S. Thorp, and D. V. Coury, “EPOCHS: integrated commercial off-the-shelf software for agent-based electric power and communication simulation,” in Proc. Winter Simulation Conf., New Orleans, LA, 2003. [9] R. Giovanini, D. V. Coury, J. S. Thorp, and K. M. Hopkinson, “Improving local and backup protection using wide area agents,” in Proc. 8th Int. Conf. Developments in Power System Protection, Amsterdam, The Netherlands, 2004.

[10] EMTDC/PSCAD User’s Guide, Manitoba HVDC Research Centre, Winnipeg, MB, Canada, 2002. [11] PSLF Manual, General Electric, 2003. [12] Network Simulator—NS2, Univ. California, Berkeley, CA, 2004. [13] Power System Test Case Archive, Electrical Engineering—Univ. Washington, Seattle, WA, 2004.

Renan Giovanini was born in Brazil in 1974. He received the B.Sc. degree in electrical engineering and the M.Sc. degree from the Escola de Engenharia de São Carlos (EESC), University of São Paulo, São Paulo, Brazil, in 1998 and 2000, respectively. He received the Ph.D. degree from the EESC from the University of São Paulo in 2005. Currently, he is with Operador Nacional do Sistema Elétrico (ONS), Rio de Janeiro, Brazil. His research interests are power systems protection, control, and automation.

Kenneth Hopkinson received the B.S. degree in computer science from Rensselaer Polytechnic Institute, Troy, NY, in 1997, and the M.S. and Ph.D. degrees in computer science from Cornell University, Ithaca, NY, in 2002 and 2004, respectively. Currently, he is an Assistant Professor of Computer Science with the Air Force Institute of Technology, Wright-Patterson AFB, OH. His research interests include distributed systems, networking, and simulation.

Denis V. Coury (M’92) received the B.Sc. degree in electrical engineering from the Federal University of Uberlandia, Uberlandia, Brazil, in 1983, the M.Sc. degree from the University of São Paulo, São Carlos, Brazil, in 1986, and the Ph.D. degree from Bath University, Bath, U.K., in 1992. Currently, he is a Full Professor in the Department of Electrical Engineering, University of São Paulo, São Carlos, Brazil. His research interests are power system protection and control, expert systems, and neural networks.

James S. Thorp (F’89) is Chair of the Department of Electrical and Computer Engineering at Virginia Polytechnic Institute and State University, Blacksburg, VA. Dr. Thorp is a member of the National Academy of Engineering and a member of the IEEE Power System Relaying Committee, CIGRE, Eta Kappa Nu, Tau Beta Pi, and Sigma Xi. He was an Associate Editor for IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS from 1985 to 1987.