Reducing the Costs of Disturbances to the Electric Power Network

Electric power networks are operated by human and mechanical agents, each with some autonomy. This paper describes two methods for coordinating the efforts of mechanical agents (relays, voltage regulators, and governors) using software agents distributed throughout the power system in order to reduce the social costs of disturbances. As evidenced by recent blackouts in Europe and North America, relatively minor disturbances can initiate cascading failures with catastrophic effects. We propose a method to limit such cascades, and thereby reduce the social costs of the resulting blackouts. The proposed methods have several important features. Firstly, they allow for varied interruption costs to be assigned to loads, thereby decreasing the social costs of the required load shedding. Secondly, the methods are distributed and cooperative, giving benefits in actuation speed and scalability and increasing the ability of the system to react well to problems that occur near seams between control areas. The mechanical agents considered here are situated at the nodes of the network. Each has control over a few variables, such as the terminal voltage of a generator or the load fed from a substation. After a disturbance, software agents communicate and decide how to adjust their control-variables in order to minimize the spread and cost of the disturbance. They do this by solving a local optimization problem while cooperating (exchanging valuable information) with their neighbors. Preliminary results suggest that cooperation schemes exist such that locally optimum solutions can approach globally optimum solutions. That is, autonomous agents working with information from their local neighborhoods can find solutions that are nearly globally optimal.