Sensitivity analysis of rare-event splitting applied to cascading blackout models

Splitting is a technique that can be used to improve the efficiency in simulating rare events. The basic idea is to create separate copies (splits) of the simulation whenever it gets close to the rare event. To implement splitting, several decisions must be made - for example, choosing a level function, choosing the number of simulation runs for each level, etc. This paper analyzes the sensitivity of the variance of the rare-event estimator to several parameters used within the splitting framework. We specifically consider a two-level fixed-effort variation of splitting for which analytic results can be derived. Results are applied to a simple model of cascading blackouts. The results illustrate that a good choice for the locations of levels may be more important than a good choice for the importance function for these types of problems.