Explainable Nonparametric Importance Sampling in Stochastic Simulation

We propose a new explainable multivariate importance sampling method in stochastic simulation. Importance sampling has the potential to significantly reduce the estimation variance when its instrumental density is properly designed. Prior studies have demonstrated the advantages of using a nonparametric method when the relationship between input variables and the output variable is unknown. However, importance sampling easily faces the curse of dimensionality, as the input dimension grows. This study devises a new method that identifies crucial input variables to effectively formulate a nonparametric instrumental density and explains the importance of each variable. The wind turbine reliability case study demonstrates the method’s efficiency.