C I R E D 19 th International Conference on Electricity Distribution Vienna, 21-24 May 2007 Paper 0540- CIRED2007 Session 4 Paper No 0540 Page 1 / 4 RESILIENCE OF DISTRIBUTED MICROGRID CONTROL SYSTEMS TO ICT FAULTS Tom RIGOLE Koen VANTHOURNOUT Geert DECONINCK K.U.Leuven – Belgium K.U.Leuven – Belgium K.U.Leuven - Belgium tom.rigole@esat.kuleuven.be koen.vanthournout@esat.kuleuven.be geert.deconinck@esat.kuleuven.be ABSTRACT Recent years a growing deployment of distributed generation (DG) in low voltage power grids is seen, and the amount of DG is expected to increase further in the near future. Distribution grids are however not designed to contain generators. Furthermore, no coordinated control for these generators exists and they provide little or no ancillary services. Therefore control strategies have been designed based on inexpensive ICT equipment and public communication networks. One such agent based control system is introduced in this paper. Also, several scenarios are simulated with ICT-fault injections to assess the systems’ resilience to these faults. After discussing the specific scenarios, the root causes of most limitations and vulnerabilities in generalized open, unbounded and distributed control systems are pointed out. INTRODUCTION The liberalization of the power grid, together with increasing concerns for the environment, has lead to an increasing use of DG. The biggest problem with current practices of DG is that these generators mainly produce active power, without any controls that contribute to the (distribution) grid operation, such as voltage or frequency control, or VAR compensation [1, 2]. A similar argument can be made for more general distribution grid equipment, in which there is no active control of resources. Despite the theoretical ability of many small loads to adjust their load profiles based on some external signal, there is very little active demand side management for small loads these days. To improve this passive management of distribution grid resources, we propose the use of decentralized, multi-agent systems (MAS) [3-6] employing peer-to-peer (p2p) networking protocols. More specifically, a multi-agent control system for DG was developed, which optimizes voltage levels, frequency (when islanded), and production costs. This paper first briefly presents a distributed control system for DG applications forming a microgrid. Next, an analysis is made of the impact of ICT faults on the system considering both accidental and malicious faults using computer simulations. We finally conclude by pointing out current limitations of such MAS in an open environment. The analysis is based on work in progress of the European project CRUTIAL, which covers interdependencies between the power grid and its ICT control systems. AGENTS AND P2P NETWORKING Multi-agent paradigm Although many definitions exist, one could define an agent as “an autonomous system that is situated in an environment and acts on it, based on inputs from the environment or other agents, in order to pursue its goals, and is often able to learn from previous experiences”. When many cooperating agents exist in the environment, it is called a multi-agent system. It is believed that in such system, with proper incentives given, low-level autonomous behaviour of individual agents will lead to near-optimal high-level emerging behaviour of the whole system. In the context of the deregulated power market, the multitude of autonomous parties involved can be seen as a large community of agents, each pursuing their own goals defined by economical and regulatory concerns, rendering (hopefully) the emerging behaviour of having a reliable grid operation build upon a free market. It seems quite straightforward to apply this multi-agent paradigm to power grid control, which is illustrated in this paper. Peer-to-peer networking Continuing upon the multi-agent methodology, which avoids centralized management, peer-to-peer 1 networking is an ideal structure to organize the agent community in open unbounded environments. These p2p-networks enable agents to find one another based on agent-IDs or agent attributes in the multitude of agents (resource discovery). In such p2p networks each node has only a couple of known other nodes, called its neighbours. This way, all nodes in the network are directly or indirectly connected in a graph- like structure. Well known examples of p2p-networks are file-sharing programs such as Kazaa. Over the Internet they build a p2p-network (among file-sharing peers) over which files can be queried. By routing queries from one peer to the other, no central indexing server is needed. The lack of a central server and the multiplicity of peer-to-peer links per peer make a p2p network quite resilient to peer crashes and communication network infrastructure failures [7]. A MICROGRID CONTROL SYSTEM As an example MAS implemented on a p2p network using the Internet for communications, we present a system for controlling a microgrid with high DG penetration [5, 6]. 1 A peer or node is a networked computer within a distributed ICT-system; they may be considered synonyms.