Data-Driven Approximations of Chance Constrained Programs in Nonstationary Environments

We study sample average approximations (SAA) of chance constrained programs. SAA methods typically approximate the actual distribution in constraint using an empirical constructed from random samples assumed to be independent and identically distributed according distribution. In this paper, we consider a nonstationary variant problem, where are independently drawn sequential fashion unknown possibly time-varying This nonstationarity may driven by changing environmental conditions present many real-world applications. To account for potential data generation process, propose novel robust method exploiting information about Wasserstein distance between sequence data-generating distributions As key result, obtain distribution-free estimates size required ensure that will yield solutions feasible under with high confidence.