Efficient simulation of blackout probabilities using splitting

Standard Monte-Carlo simulation may be computationally intractable when the events of interest are extremely rare. This paper applies the rare-event simulation technique of splitting to the problem of estimating large-scale blackout probabilities. First, a stochastic model of cascading line failures is developed. Then, a simple network is presented and an analytical solution is derived for the simple network. Exploration of the analytical solution provides some guidance for setting splitting parameters in more complicated networks. In particular, geometrically increasing levels typically give an improvement over equally-spaced levels, due to the cascading nature of blackouts. The splitting methodology is applied to several different network topologies of varying complexity – a mesh network, a grid network, and the IEEE 118-bus network. Numerical results indicate that splitting has the potential to be effective on problems for which standard simulation may be infeasible.

► We present a new stochastic model for rare-event cascading line failures.
► Application of a splitting methodology greatly improves the simulation efficiency.
► Analytical results provide insight on how to select splitting parameters.
► Numerical results show speed benefits where standard simulation is intractable.