(461b) Optimization Under Uncertainty Using Surrogate Models for Confidence Evaluation

This work presents a novel approach to solving optimization under uncertainty problems. These problems are important because many real-life engineering processes have significant inherent uncertainty. The uncertainty in these problems make it difficult to optimize using common optimization methods. The proposed method determines the range for which a possible solution may lay in by estimating a confidence value to that range. The confidence value is determined by testing the percentage of surrogate quadratic models that would have solutions inside the range of interest. Uncertainty in the objective function is accommodated using randomly selected quadratic models and soft constraint penalties. The size of the range of interest and the overall objective function are minimized to yield a range for which the solution to the objective function lies with a specified confidence value. The paper outlines case studies conducted using the novel optimization method as well as a more relevant example involving the optimization of a closed-loop dynamic diabetic patient model system.