数理モデルにおける大域的感度解析手法の比較と評価

抄録

This study investigates global sensitivity analysis methods on mathematical models in terms of a tradeoff between computational cost and accuracy. Sobol’ method based on Polynomial Chaos Expansion (PCE-Sobol’), Morris method, and variance analysis with orthogonal tables are applied to a mathematical model of an experimental facility for a space propulsion system. It was found that variance analysis with an orthogonal table is the most efficient for identifying the most sensitive factor with fewer evaluations. Conversely, the PCE-Sobol’ and Morris methods are more suitable for screening sensitive factors with hundreds of evaluations.