AD-HOC vs. SUPERVISORY WIDE AREA BACKUP RELAY PROTECTION VALIDATED ON POWER/NETWORK CO-SIMULATION PLATFORM

Transcription

1 1 AD-HOC vs. SUPERVISORY WIDE AREA BACKUP RELAY PROTECTION VALIDATED ON POWER/NETWORK CO-SIMULATION PLATFORM Hua Lin, Santhoshkumar Sambamoorthy, Sandeep Shukla, James Thorp, Lamine Mili Department of Electrical and Computer Engineering, Virginia Tech Blacksburg, Virginia, USA {birchlin, ssan, shukla, jsthorp, vt.edu Abstract The functional and computational capabilities of modern digital relays currently deployed in the power systems are usually underutilized. There is great potential to non-intrusively upgrade the existing relay protection infrastructure, for example, by constructing or properly utilizing a network infrastructure connecting the relays. In this paper 1, we propose two agent-based backup relay protection schemes built on low latency communication networks. Our schemes are non-intrusive in the sense that they don t modify anything in the current relay settings and structures, but rather provide additional functionality and robustness through communicating software agents associated with relays. The operation of the protection schemes are fully discussed and then validated on a power/network co-simulation platform. The comparison of two schemes based on the co-simulation provides guidance for possible future implementations of wide area protection in real power systems. Key Words: Backup Relay Protection, Network Infrastructure, Software Agents, Co-Simulation 1 INTRODUCTION Power system protection is crucial to the robustness, resilience, sustainability and security of the power grids[1]. Nowadays power systems are equipped all kinds of relays to protect the lines and the various equipments. Generally the relays monitor local operation status and control local breakers to isolate faults. Communication is usually excluded during the entire protection with few exceptions[2-6]. Although pilot relays may coordinate the adjacent relays to make protection decision through the proprietary network, the communication is still limited by low channel bandwidth. However, high bandwidth, low latency communication is no longer a rarity. Under the vision of Smart Grid, the prospective power grids will be empowered by extensive network infrastructure and stateof-the-art communication and computing, especially the Internet-related techniques. Hence, the current power system protection infrastructure may greatly benefit from these potential upgrades and reach an unprecedented level of robustness. The problem of robustness in the protection system comes from two possibilities: (i) when a relay fails to trip when it should; and (ii) when a relay trips when it should not. The first case is usually mitigated by backup relays (zone 2 and zone 3 relays). However, the second problem is often considered as a non-issue because it does not damage the equipments. In fact, it has been found that 1 The authors acknowledge the partial support of NSF under grant NSF EFRI many of the cascading blackouts in the recent past involved false tripping of some relays[7, 8]. The false tripping may comes from hidden failures of the relays[9, 10]. But it is also possible to happen when the system is under normal operation. For instance, the zone 3 relays on lines between Pennsylvania and New York tripped according to their settings during the 2003 blackout because a huge amount of power flow from south to north induced low voltage and huge current flow and led to perceived impedance lowering[11]. In such a case, zone 3 relays behave as per their settings and they don t have enough wide area system visibility to do otherwise. A supervisory relaying system was proposed in [12] based on software agents monitoring faults and breaker status additionally through their interaction with the relays. These agents, by virtue of low latency communication, have wide area visibility and thus are able to advise the relays against false trips when appropriate. While this was a tentative first step towards utilizing communication to make backup relays robust to false trips, it didn t detail the distributed software agent architecture, their communication and peer relationships. In this paper we accordingly model the supervisory protection scheme with much more concrete details and further propose a new ad-hoc type of protection scheme. The new schemes not only add more robustness to false tripping but also improve the backup protection performance for the primary relay failure situation. Our protection schemes have shown faster reaction time than traditional backup protection as well as the potential to cover very large areas. The paper is organized as follows: In the next section we introduce the two wide area backup relay protection schemes and explain their operations in detail. Section 3 discusses several problems that we meet towards the implementation of these protection schemes including the design of a power/network co-simulation platform. The schemes are then simulated and validated on the cosimulation platform in section 4. In section 5, the two schemes are compared using several criteria. Then the entire work is summarized in the concluding section. 2 AGENT-BASED BACKUP PROTECTION Directional distance relays are widely deployed in current power systems. By properly adjusting zone 1, zone 2 and zone 3 settings they can achieve both primary and backup protection of the transmission lines. Usually zone 1 protection operates instantaneously while zone 2 and zone 3 protections is associated with an intended time delay as backups. It is common practice to use longer time

2 2 delay for longer reach of the relays so that they can provide effective system protection without unnecessary power loss. The zone 3 relay time delay may be around one second which is much longer than the primary relay. Although system may suffer from instability issues during the delay, without full system visibility, the backup relays can t do anything but wait. Plus false tripping problems today s relay protection system needs adding more robustness to such unpredictable incidents. We accordingly propose new power system protection schemes that leverage the present distance relay protection framework with the addition of an underlying network infrastructure. Modern microprocessor-based digital relays are more reliable and efficient than traditional electromechanical ones, thus it is possible to attach software agent to them to attain more elaborate protection schemes. We manage to make the distance relays communicate with each other from which a coordinated system protection is formed. By virtue of extensive communication our protection schemes have faster backup relay protection and additional robustness to false tripping. There are two types of communication types between relays hence two configurations of protection schemes are discussed. 2.1 Supervisory Protection (master to slave) We discussed a supervisory backup relay protection scheme in [12] preliminarily. Here we provide more details for the potential implementation of this scheme. In such a supervisory protection scheme, there is one central protection controller called master agent which coordinates the operations of digital relays in the system. The digital relays discussed here are all directional distance relays which protect the high voltage transmission lines. Each of the relays is assumed to have a software agent called slave agent attached to it. The central controller and all the relays are connected by a network infrastructure so that the master agent and the slave agents can communicate with each other. The primary protection in our scheme is the same as the traditional ones. However the backup protection is distinguished from others. When the relay sees zone 2 or zone 3 faults, instead of waiting for a time delay to trip, our scheme proactively collects information from other relays to evaluate the status and make the decision. This procedure is done by communications between slave agents and the master agent which can be seen in Figure 1. Firstly the slave agent whose associated relay sees a remote fault submits a request to the master agent for decision. The master agent then asks other responsible slave agents to see if others also see the fault. Based on the feedback from other slave agents, the master agent sends the final decision to the original slave agents. Figure 1: Supervisory protection communication. Detailed operations of this master-slave mode supervisory protection scheme are shown in Figure 2 using a finite state machine (FSM) representation. In the FSM, a circle represents a certain state. An arrow line represents a transition from one state to another. There is a fraction associated with each transition. The numerator position shows the event which causes the transition and the denominator position shows the action taken because of that event. (a) Slave agent (b) Master agent Figure 2: FSM of supervisory protection. On the slave agent side, the relay starts from the normal monitoring state and keep sensing the transmission line. If a zone 1 (local) fault is observed the relay should trip the transmission line immediately. If a zone 2 or zone 3 fault (remote) is observed, the slave agent should send a decision request to the master agent and enter the wait for decision state. If the fault disappears during the wait state, the slave agent should go back to the normal state and block whatever decision received from the master agent since this condition indicates the fault may have been cleared by its own primary protection. If the fault persists and the slave agent receives a trip decision from the master agent, the relay should trip the line since the primary protection may have failed. If the slave agent receives a block decision from the master agent but still sees the fault, this indicates the relay may have a hidden failure. Then the slave agent should put the relay out of service and call for future maintenance. In this manner the slave agents can both expedite the backup protection and prevent hidden failure induced false tripping.

3 3 On the master agent side, when it receives a decision request from a slave agent, it enters the processing decision request state. A group of responsible relays which may protect the fault area are selected and queried by the master agent. When the software agents of the selected relays receive the queries, no matter which state they are in, they should report if they see a fault back to the master agent. The master agent will try to make the final decision whenever it receives a feedback. As long as a final decision could be made, the master should send it to the original slave agent to take action. Figure 2 (b) only shows the master agent operation for one slave agent. Actually when a fault happens in the system, multiple slave agents could send request to the master agent. Therefore, the master agent should be a multi-threaded program which can handle all the requests simultaneously. Although there is extensive communication as shown in Figure 1, the total communication time could still be shorter than the traditional time delay set for the zone 2 or zone 3 protections. However the time delay associated with zone 2 or zone 3 should not be eliminated since the network itself may fail which can hamper the communication. Either link failure or traffic congestion may significantly increase the communication delay and even drop messages. Hence if our communication-based protection couldn t complete within a certain time, the relays should trip the line to protect the system as shown in Figure Ad-Hoc Protection (peer to peer) In the supervisory protection scheme, the master agent is the most crucial component since it coordinates all the slave agents. However if the master agent fails due to any reason, the entire protection scheme will fail and put system in danger although the relays still have the traditional time-delayed protection function. To address this drawback, a backup master agent may be necessary, that is, add redundancy and additional financial cost to improve robustness. Another issue with the supervisory scheme is, since the slave agents always communicate with the master agent as shown in Figure 1, the communication time could be long and unstable depending on how far the slave agent is from the master agent. The coordination of the protection could be difficult to predict. Accordingly we propose an ad-hoc protection scheme on the basis of the supervisory protection scheme. In this scheme, we remove the master agent and transplant its functions to each slave agents. Now the slave agents can directly communicate with each other in a peer-to-peer manner as shown in Figure 3. Figure 3: Ad-hoc protection communication. Figure 4 shows the FSM representation of the ad-hoc protection operations. The only agents in this scheme are the peer agents. Each peer agent actually combines the operations of the slave agent and the master agent as shown in Figure 2. The main difference compared to the supervisory protection is when the peer agent sees a remote fault it should send queries directly to other responsible peer agents. Whenever the peer agent receives a report from other peer agents, it will try to make a decision itself. Hence this is a fully distributed and autonomous application based on ad-hoc communication. Ad-hoc communication has been investigated extensively in computer engineering. We believe power applications can also make use of the studies especially in the Smart Grid era. Our protection scheme takes a tentative step towards this direction. Figure 4: FSM of ad-hoc protection. 3 IMPLEMENTATION In this section, we describe more details about how to implement and validate the protection schemes in a cosimulation environment. 3.1 Relay Searching Algorithm In both of the protection schemes proposed in previous section, a relay searching process is needed for the agent to determine the responsible relay agents group when a fault is observed. It is possible to preset a list of relays for each agent in memory however if the power systems of interest cover relatively large areas and consist of large number of relays, the storage capacity needed could be cumbersome. Furthermore, if the power system topology changes the updating work in turn could be timeconsuming. We therefore design a relay searching algorithm working on a graph which has better feasibility and flexibility. The step by step instruction of this algorithm is shown in Figure 5. Firstly we need to convert the power system topology into an undirected graph GV (, E ). Since the distance relays here protect the transmission lines only, we should eliminate other unrelated components like transformers and generators etc. Thus all the transmission lines should be converted as edges in the graph G and only the buses which connected to a

4 4 transmission line should be converted as vertices in graph G. All the electrical attributes of the lines and buses should be neglected. The relay in this graph can be represented by an ordered pair (( mn, ), mwhich ) means the relay locates at bus m side of the transmission line from bus m to bus n. The searching algorithm basically consists of two major steps. First, based on the relay who submits the decision request, the algorithm find out the possible faulted lines. Then for each possible faulted line, the algorithm finds out two primary protection relays and all the backup relays for this line. Algorithm Input: A modified undirected system graph GV (, E ) A relay represented by (( mn, ), m ) Output: A relay set R Steps: 1. Find the possible faulted lines set L : for each edge ( uv, ) E except ( mn, ), if n= uor n= v, add ( uv, ) to L 2. Find the responsible relays for each line in L : a. For each ( u', v') L, add (( u', v'), u ') and (( u', v'), v') to R b. If n= u', for each edge ( u", v") E except ( u', v '), if v' u" (( u", v"), v") to R, if v' v" (( u", v"), u") to R c. If n= v', for each edge ( u", v") E except ( u', v '), if u' u" (( u", v"), v") to R, if u' v" (( u", v"), u") to R Figure 5: The relay searching algorithm. We use an example to better illustrate this algorithm. We implement this algorithm for the New England 39-bus system as shown in Figure 6. We assume there is a fault on transmission line (4,14) and its backup protection relay ((14,15),15) senses the fault. We then find all the responsible relays for relay ((14,15),15) following the steps in Figure 5. In step 1, we can find all the lines connected to bus 14 except (14,15) where we get lines (4,14) and (13,14). These two lines are all possible faulted lines although the real fault should be located at only one line. Then within step 2, for line (4,14), in step 2.a we find its primary protection relays ((4,14),4) and ((4,14),14). Next, in step 2.b or 2.c we find all the lines connected to bus 4 except (4,14) which are (3,4) and (4,5) and then find out the backup protection relays ((3,4),3) and ((4,5),5). For line (13,14), following the similar steps we find the final relay set for ((14,15),15) which are ((4,14),4), ((4,14),14), ((3,4),3), ((4,5),5), ((13,14),13), ((13,14),14) and ((10,13),10). Figure 6: Relay searching on 39-bus system[13]. 3.2 Decision Making So far we haven t explained how the backup protection decision is made by either the master agent or the peer agent itself. The decision is made based on the feedbacks from the slave agents or the peer agents which are selected using the relay searching algorithm. Based on the FSM in Figure 2 and Figure 4 there are totally four types of feedback reports that can be received from the agents: remote fault, no fault, fault tripped and out of service. Note that there is no local fault feedback report because if the primary protection relay works normally, a local fault should be tripped immediately. Our decision making works as follows: a. If the agent receives a remote fault report, the trip decision can be made immediately. Since the possibility of two relays have hidden failure simultaneously is very low, there is no need to keep waiting for other reports in this condition. b. If the agent receives a no fault report, it should first check if all the relays in the responsible relay set have reported back. If so the block decision can be made because none of other relays see a fault implying a potential hidden failure. Otherwise the agent should keep waiting for other reports. c. If the agent receives a fault tripped report, no decision should be made. But the agent should remove this reporting relay from the responsible relay set since this report won t affect the decision any more. d. If the agent receives an out of service report, no decision should be made and the reporting relay should also be removed as well. 3.3 Co-Simulation Platform Both of the supervisory and ad-hoc protection schemes need communication between software agents. It is crucial for these schemes to guarantee the communication time between relays to meet the system protection requirements. The communication time consumed from a relay sees a fault to the relay gets the decision should be within a certain threshold like the traditional zone 3 time delay. Usually the communication delay consists of

5 5 transmission delay, queuing delay, process delay and propagation delay. It is very difficult to accurately estimate total communication delays simply based on the mathematical delay models especially when the network is complex. A simulation environment is naturally necessary for power engineers to estimate the communication time in two protection schemes before the real system testing. However most of today s power system softwares can t simulate the power system and its network infrastructure together[14]. A co-simulation of power system and communication network is desired for the simulation and validation of our protection schemes. We designed a co-simulation platform in [15] which integrates PSLF and NS2 softwares. PSLF is a GE product which can simulate power flow, short circuit and dynamics of the power system. NS2 is a well-known open source network simulator which has good reputation in both industry and academia. We choose these two softwares because they are all recognized in their own domains and have good user-extension capabilities. The software architecture of this co-simulation is shown in Figure 7. We integrate PSLF and NS2 using a rigorous synchronization mechanism so that there are no extra errors introduced by the interface between the two softwares. A new dynamic model epcmod is written in PSLF as an interface to transfer power data to NS2. Also new agent models are implemented in NS2 using C++ and Tcl. When we run the co-simulation, the power system is created in PSLF and the communication network is built in NS2. The software agents in NS2 periodically collect power data from PSLF and do their own impedance calculations. When a fault is observed the software agents will start communication in NS2. New network transport protocol which can carry power data is also realized in NS2. We choose UDP as the transport protocol in this simulation for its simplicity and efficiency. 4.1 Supervisory Protection For the supervisory protection, we create one fault on each transmission line sequentially and intentionally disable one of the primary protections so that other backup relays can take in charge. In each fault case, there could be multiple backup relays contacting the master agent at bus 16. We measure and record the communication time for each relay in each fault case and take the average. The average communication time distribution for all the relays is plotted in Figure 8. We find that the communication time significantly varies depending on the location of the relays. Even so, all the communication can be done within five cycles which is about s in the US. Figure 8: Time delay distribution in supervisory protection Figure 7: Power/network co-simulation platform. 4 SIMULATION SETTINGS AND RESULTS With the help of the co-simulation platform, we simulate and validate our protection schemes in this section. The protection schemes are implemented for the New England 39-bus system as shown in Figure 6. We place one relay at both ends of each transmission line in the system so that they provide a full coverage of the system. For the supervisory protection scheme, a master agent is also placed at bus 16. The communication network is constructed such as it has the same topology of the power system. The communication link is modeled as having 100 Mbps bandwidth and 3 ms latency. We assume that this network is dedicated to the protection schemes only so that there is no extra background traffic. 4.2 Ad-Hoc Protection For the ad-hoc protection, we create one remote fault for each relay agent and also assume the primary protection fails to trip. We measure the communication time needed for both trip decision and block decision. The trip decision can be made as long as another peer agent sees the fault and the block decision will be made when all the responsible peer agents report back. The final time delay distribution is plotted in Figure 9 where we find immediately that the time consumed during peer-to-peer communication is very short and stable. The only exception happens to the relays which are on the lines connecting bus 26, 28 and 29. These lines form a small transmission loop so that other responsible peer agents could be very close to the original one. The ad-hoc protection could react very fast: trip decision can be made within one cycle and block decision can also be made within two cycles.

6 6 link failures are close to one agent, the communication time for that agent could be much higher than the normal condition. Therefore, our assumption actually may help us to find the worst case results which should be of more interest to system designers. Figure 9: Time delay distribution in ad-hoc protection Figure 11: Time delay distribution in ad-hoc protection with communication link failure The communication time distributions for both of the schemes under link failure condition are shown in Figure 10 and Figure 11. For supervisory protection the distribution is still unstable. The communication time increases a little bit for most of the relays but also increases a lot for several others. Generally speaking, the communication time increases by one cycle on average. For ad-hoc protection, the time distribution becomes unordered. Most of relays need much more communication time to make a trip or block decision. There are also some relays whose decision time doesn t change too much. This is also due to transmission line loops in the system. 5 COMPARISON AND DISCUSSION Figure 10: Time delay distribution in supervisory protection with communication link failure 4.3 Protection under Network Failure Similar to the power system, the communication network may suffer from failure as well. Network link failure, node failure and traffic congestion can all undermine the normal communication and affect the applications on top of it. Since our protection schemes highly rely on the network infrastructure, it is also very important to study the robustness to network failure of our protection schemes. In order to study such scenarios, in our next set of experiments we assume that when a fault happens, the communication link at the same location will also be out of service. In practice, this may be rare since most of the network devices have backup power supply. But the failure possibility of that link should be higher than the other. Furthermore our simulation shows if the 5.1 Difficulty of Real System Implementation Most of today s digital relays can be equipped with network interface as an option. International communication standards are being designed to realize compatible communication between devices from different vendors. An underlying communication network could be constructed to connect the relays right away. The cost of a new network infrastructure could be high but the power system may share bandwidth with existing networks like telephone network, cell phone network or the Internet, although security issues need to be considered. The supervisory protection scheme is a non-intrusive upgrade of existing protection infrastructure and only one extra master agent needs to be designed. The coordination between relays can be realized by slave software agents which wouldn t require any redesign of current digital relays. As long as the digital relays are able to send the fault status and receive trip or block decisions, all the other work could be completed by slave software agents and the master agent. The power system information

7 7 should be stored in the master agent only so that the investment for the design of those software agents could be justified. The ad-hoc protection scheme may need more investment, but only for the design of the peer agents. The existing relays don t have to be replaced either. Since each peer agent operates in an autonomous manner, the system information should be installed in each peer agent to find the responsible peer agents. However if the system topology changes the system information stored in the peer agents have to be updated one by one which will limit the flexibility of this application. A compromise solution could be using a hybrid protection scheme based on both supervisory and ad-hoc protection. The master agent is still maintained in this hybrid mode to update the system information for each peer agent when necessary. But if a fault is observed in the system the peer agents still communicate with each other in an ad-hoc manner. 5.2 Reaction Time From the simulation results in the last section we find that the ad-hoc protection has much smaller and more stable communication time than the supervisory. This is an advantage indicating that this protection scheme could be deployed in a very large system since the communication time is almost independent of the relay location. In contrary, the supervisory protection may be limited within a small area to satisfy the communication requirement. When considering the protection of a larger area, multiple master agents may be needed to overlap their reaches. However this may induce unwanted contention and coordination problems. A fast reaction time of the backup protection may not be always good. In our simulation, the ad-hoc communication can be finished within one cycle. This reaction time may be too fast for a backup relay since the backup relay may trip even faster than the primary protection. More rigorous design specifications could be posed for the relay designer to solve oversensitive issues. We also need to mention that, our simulation results are obtained based on a relatively simple network model and the real network could be more complex and sparse. However our co-simulation platform provides the users the capability of modeling their own complex networks in NS Robustness to Network Failure The simulation results in the last section show that both of the protection schemes have extra communication delays due to the network link failure. In supervisory protection scheme the average extra delay is about one cycle. Compared to the original communication time this is about 30% increase. However in the ad-hoc protection, the communication time increases by two or three times which implies the ad-hoc protection performance declines more than the supervisory protection under such kind of failure. Also the time distribution of the ad-hoc protection becomes unstable under link failure hence it increases the protection design difficulty when considering system fault tolerance. 6 CONCLUSION In this paper we discuss the potential to upgrade the current protective relay infrastructure with non-intrusive methods. A supervisory and an ad-hoc backup relay protection schemes based on a network infrastructure and communication are proposed accordingly. The details of the principles, operations and implementations of these two protection schemes are discussed thoroughly. We further simulate and validate the two protection schemes on a power/network co-simulation platform. Both of the normal condition and operation under network link failure of the schemes are considered. The simulation results provide the design guidance for the potential protection deployment. The pros and cons of both schemes are compared and discussed subsequently. REFERENCES [1] W. R. Lachs, "A new horizon for system protection schemes," Power Systems, IEEE Transactions on, vol. 18, pp , [2] S. Song, et al., "Multiagent And WAN Based Adaptive Coordinated Protection System," in Transmission and Distribution Conference and Exhibition: Asia and Pacific, 2005 IEEE/PES, 2005, pp [3] T. Shono, et al., "A remote supervisory system for a power system protection and control unit applying mobile agent technology," in Transmission and Distribution Conference and Exhibition 2002: Asia Pacific. IEEE/PES, 2002, pp vol.1. [4] W. Dong-qing, et al., "Design of a novel wide-area backup protection system," in Transmission and Distribution Conference and Exhibition: Asia and Pacific, 2005 IEEE/PES, 2005, pp [5] X. Tianqi, et al., "A novel communication network for three-level wide area protection system," in Power and Energy Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st Century, 2008 IEEE, 2008, pp [6] M. M. Eissa, et al., "A Novel Back Up Wide Area Protection Technique for Power Transmission Grids Using Phasor Measurement Unit," Power Delivery, IEEE Transactions on, vol. 25, pp , [7] H. Wang and J. S. Thorp, "Optimal locations for protection system enhancement: a simulation of cascading outages," Power Delivery, IEEE Transactions on, vol. 16, pp , [8] J. De La Ree, et al., "Catastrophic Failures in Power Systems: Causes, Analyses, and Countermeasures," Proceedings of the IEEE, vol. 93, pp , [9] D. C. Elizondo, et al., "Hidden failures in protection systems and their impact on wide-area disturbances," in Power Engineering Society Winter Meeting, IEEE, 2001, pp vol.2. [10] Y. Fang, et al., "Effects of Protection System Hidden Failures on Bulk Power System Reliability," in Power Symposium, NAPS th North American, 2006, pp [11] J. Sung-Kwan, et al., "Empirical Analysis of the Impact of 2003 Blackout on Security Values of U.S. Utilities and Electrical Equipment Manufacturing Firms," Power Systems, IEEE Transactions on, vol. 22, pp , [12] S. Garlapati, et al., "Agent Based Supervision of Zone 3 Relays to Prevent Hidden Failure Based Tripping," presented at the First IEEE International Conference on Smart Grid Communications, [13] I. Hiskens, "Significance of Load Modeling in Power System dynamics," presented at the x symposium of specialists in electric operational and expansion planning, [14] K. Hopkinson, et al., "EPOCHS: a platform for agent-based electric power and communication simulation built from commercial offthe-shelf components," Power Systems, IEEE Transactions on, vol. 21, pp , [15] H. Lin, et al., "Power System and Communication Network Co- Simulation for Smart Grid Applications," will be presented at the Conference on Innovative Smart Grid Technologies, 2011.