Multifidelity strategies for forward and inverse uncertainty quantification of wind energy applications

Abstract

The complexity of analyzing and designing wind power plants is significantly amplified when high-fidelity numerical simulations are integrated for evaluating performance. To accurately predict the performance of these systems, one requires both computational fluid dynamics simulations, to predict the flow evolution around each turbine and wake interactions among turbines, and structural simulations, to evaluate the mechanical loads. Moreover, wind power plants operate in an intrinsically stochastic environment and therefore the characterization of their performance also needs to include uncertainty quantification (UQ). UQ is the ensemble of tasks that deals with the characterization of the sources of uncertainty and their propagation through numerical codes. The characterization process is also known as inverse UQ, in that one must invert for the input distributions that are consistent with observed data, and the propagation task is also known as forward UQ, in that the characterized distributions are pushed forward through the numerical simulations to compute corresponding output distributions on quantities of interest. Both inverse and forward UQ algorithms, despite rapid advancements in recent years, still require a large number of numerical simulations to be performed to provide an acceptable accuracy in their solutions. In this work, we demonstrate how multifidelity surrogate-based strategies can be successfully coordinated and leveraged to reduce the overall computational burden of both inverse and forward UQ workflows applied to wind energy production.