Hybrid Modeling and Multi-Fidelity Approaches for Data-Driven Branch-and-Bound Optimization

Computer Aided Chemical Engineering 2023

High-Fidelity (HF) simulations are essential in quantitative analysis and decision making in engineering. In cases where explicit equations and/or derivatives are unavailable, or in the form of intractable nonlinear formulations, simulation-based optimization methods are used. We recently proposed a data-driven equivalent of spatial branch-and-bound that constructs underestimators of high-fidelity simulation data. Within this framework, low-fidelity surrogate data can also be used to inform underestimators. In this work, we utilize the recent advances in hybrid multifidelity surrogate modeling techniques to improve the validity of our underestimators, which leads to better bounds and incumbent optima with lower sampling requirements. Specifically, we show that by modeling the error between the high-fidelity and low-fidelity data, the surrogates learn more about the underlying function with less sampling requirements.

Link to Publication