Bayesian optimization considering constrained boundary exploration
Fan Jia, Mei Han · 2023
Bayesian optimization has become a popular solution for solving black-box or expensive optimization problems. Optimization problems accompanied with constraints are more common in practical applications. Existing methods usually focus on finding the optimal solution in the feasible region, and constrained boundaries sometimes lead to inefficiencies and even limit their applicability. Therefore, this paper proposes a new strategy that uses the classical Kriging surrogate model and presents a two-stage acquisition function combining the search for a feasible optimal solution with constraint boundary exploration. The aim is to find the optimal solution satisfying the constraints quickly and precisely while minimizing the calls to the objective function and constraints. Three numerical examples demonstrate the success of our proposed method.