Abstract: Using data from the North American Electric Reliability Council (NERC) for 1984–2006, we find several trends. We find that the frequency of large blackouts in the United States has not decreased over time, that there is a statistically significant increase in blackout frequency during peak hours of the day and during late summer and mid-winter months (although non-storm-related risk is nearly constant through the year) and that there is strong statistical support for the previously observed power-law statistical relationship between blackout size and frequency. We do not find that blackout sizes and blackout durations are significantly correlated. These trends hold even after controlling for increasing demand and population and after eliminating small events, for which the data may be skewed by spotty reporting. Trends in blackout occurrences, such as those observed in the North American data, have important implications for those who make investment and policy decisions in the electricity industry. We provide a number of examples that illustrate how these trends can inform benefit-cost analysis calculations. Also, following procedures used in natural disaster planning we use the observed statistical trends to calculate the size of the 100-year blackout, which for North America is 186,000 MW.
Abstract: Cascading failures in electrical power networks often come with disastrous consequences. A variety of schemes for mitigating cascading failures exist, but the vast majority depend upon centralised control architectures. Centralised designs are frequently more susceptible to communications latency and bandwidth limitations and can be vulnerable to random failures and directed attacks. This paper proposes a decentralised approach. We place control agents at each substation in a power network, each of which uses decentralised Model Predictive Control (MPC) to select emergency control actions. When making decisions, the control agents consider not only their own goals, but also the goals of nearby agents. Thus the agents act with Reciprocal Altruism (RA). Results from simulations of extreme cascading failures within the IEEE 300 bus test network indicate that this approach can dramatically reduce the average size and social cost of large cascading failures. Simulations also show that the bandwidth required for message passing is well within the limits of current technology.
Abstract: Cascading failures in electricity networks often result in large blackouts with severe social consequences. A cascading failure typically begins with one or more equipment outages that cause operating constraint violations. When violations persist in a network, they can trigger additional outages which in turn may cause further violations. This paper proposes a method for limiting the social costs of cascading failures by eliminating violations before dependent outages occur. Specifically, our approach places one autonomous software agents at each bus of a power network, each of which is tasked with solving the global control problem with limited data and communication. Each agent builds a simplified model of the network based on locally available data and solves its local problem using model predictive control and cooperation. Through extensive simulations with IEEE test networks, we find that the autonomous agent design meets its goals with limited communication. Experiments also demonstrate that allowing agents to cooperate can vastly improve system performance.
Abstract: Most large blackouts are caused by cascading failures—sequences of equipment outages, one set of outages precipitating another. We study the application of distributed, autonomous agents for shortening such sequences. Each agent controls a single variable—the consumption of a load or the output of a generator. Each agent uses model predictive control and cooperates with its neighbors in making its decisions. Experiments using the IEEE 118 bus test case illustrate the effectiveness of this method.