Autonomous agents and cooperation for the control of cascading failures in electric grids

Abstract

A power system can be thought of as a stochastic hybrid system: a finite state machine whose states involve continuous variables with uncertain dynamics. Transitions in this machine correspond to outages of generation and transmission equipment. A cascading failure corresponds to a series of such transitions whose net effect is a blackout. We present evidence that the probability of cascading failures is subject to phase transitions - large and abrupt changes that result from only small changes in system stress. We suggest a network of distributed, autonomous agents to reduce the ill effects of cascading failures. These agents improve their decisions by cooperating (sharing goals and exchanging information with their neighbors). Results from experiments on the IEEE 118 bus test case are included.