A "Random Chemistry" Algorithm for Identifying Collections of Multiple Contingencies That Initiate Cascading Failure

This paper describes a stochastic "Random Chemistry" (RC) algorithm to identify large collections of multiple (n-k) contingencies that initiate large cascading failures in a simulated power system. The method requires only O(log (n)) simulations per contingency identified, which is orders of magnitude faster than random search of this combinatorial space. We applied the method to a model of cascading failure in a power network with n=2896 branches and identify 148243 unique, minimal n-k branch contingencies (2 ≤ k ≤ 5) that cause large cascades, many of which would be missed by using pre-contingency flows, linearized line outage distribution factors, or performance indices as screening factors. Within each n-k collection, the frequency with which individual branches appear follows a power-law (or nearly so) distribution, indicating that a relatively small number of components contribute disproportionately to system vulnerability. The paper discusses various ways that RC generated collections of dangerous contingencies could be used in power systems planning and operations.