The last decade has seen extensive effort devoted to analysis of power grid cascading failure, in which initial outage of a small number of network elements subsequently overloads other components, forcing their removal from service, in turn overloading additional components, potentially expanding to a large large-scale blackout. These methods have been largely restricted to steady state models, while experience in outage events indicates that transient power swings often dominate the final stages of cascading failure, suggesting the
need to represent dynamics.
Work here develops a dynamic model for cascading grid failure, augmenting "swing dynamics" with specially structured, smoothed representations of protective relays that disconnect transmission lines, generators, and loads. This construction offers a nearly Hamiltonian structure, with the gradient of a scalar "potential-like" function playing a key role. Any partially degraded network configuration yields an equilibrium point, locally stable if that configuration is operable. Cascading failure is then a sequence of transitions between these
degraded network equilibria, with vulnerability to transition characterized by the potential barrier to be overcome along a transition path. Exploiting analogous problems in computational chemistry, this talk will describe adaptation of Nudged Elastic Band and String methods to transition path calculation in the power context.