Uncertainty quantification in response surface models in parametric analyses
GCMS '09 Proceedings of the 2009 Grand Challenges in Modeling & Simulation Conference
| Hi-index | 0.00 |
In this paper, the response surface modeling approach is applied to study the effectiveness of different grounding schemes implemented for the mitigation of single-line-to-ground faults in a simulation of a notional Medium Voltage DC shipboard power system. For preliminary exploration multivariate adaptive regression splines (MARS) models are used to find the functional relationships between the input parameters and response variables. Using the constructed MARS models, Sobol' total sensitivity indices are computed for the models to study possible differences in the sensitivities of the response variables for the different grounding schemes. For final analyses, and for response surface models in which the MARS models show significant prediction error, Gaussian process models are used to account for prediction variance. The work presented herein builds upon analysis previously presented in [1], discussing corrections made to the simulation model, and incorporating additional response surface models.