Here we develop an approach to predict power grid weak points, and
specifically to efficiently identify the most probable failure modes in load
distribution for a given power network. This approach is applied to two
examples: Guam's power system and also the IEEE RTS-96 system, both modeled
within the static Direct Current power flow model. Our algorithm is a power
network adaption of the worst configuration heuristics, originally developed to
study low probability events in physics and failures in error-correction. One
finding is that, if the normal operational mode of the grid is sufficiently
healthy, the failure modes, also called instantons, are sufficiently sparse,
i.e. the failures are caused by load fluctuations at only a few buses. The
technique is useful for discovering weak links which are saturated at the
instantons. It can also identify overutilized and underutilized generators,
thus providing predictive capability for improving the reliability of any power
network.