GP-RTO-EI

We present an RTO algorithm that relies on trust-region ideas in order to expedite and robustify convergence, and uses Bayesian optimization through Gaussian processes as a workhorse. We explore Expected Improvement and Upper Confidece Bound as adquisition functions, and adjust the size of the trust region based on the Gaussian processes’ ability to capture the plant-model mismatch in the cost and constraints. We draw parallels to expensive black-box optimization, hybrid modelling, reinforcement learning, and dual control which are all present in the proposed approach. Finally, we illustrate this new modifier-adaptation scheme on benchmark problems.