(671f) Accelerating Optimization of Unsteady Chemical Reactor Operation through Multi-Fidelity Modeling

High fidelity models are often required to optimize operations for chemical reaction systems characterized by complex dynamics. However, such models can be computationally expensive, and is often a bottleneck in their use for optimization. In this work, we propose a multi-fidelity [1] approach which alternates between a low-fidelity and the original high-fidelity model to accelerate the optimization of static as well as periodic operations for such reacting systems.

While active learning techniques like Bayesian Optimization [2] intelligently sample design spaces for the high-fidelity model, we can also leverage low-fidelity data to gain useful information about the high-fidelity space so to further accelerate the optimization scheme. In this sense, by combining multi-fidelity modeling with optimal sampling techniques, we aim to improve conventional Bayesian Optimization [3].

We demonstrate our framework on a simple example (an elementary reaction scheme with slow and fast reactions) [4]. By applying the Quasi Steady State Approximation, we derive a low-fidelity model that reduces the dimensionality of the reaction system, providing a computationally efficient model - which, however, is only asymptotically accurate at particular regions of parameter space. This constructive interaction between the two models, facilitated by multi-fidelity Gaussian Process Regression, intelligently samples the parameter space and allows for efficient optimization of the yield and selectivity under both static and periodic operation of such reacting systems. Our results highlight the potential of multi-fidelity algorithms in optimizing expensive functions and advancing the study (and improve the performance) of chemical kinetics.

[1] S. Guth, et al, Application of Gaussian process multi-fidelity optimal sampling to ship structural modeling. 2022.

[2] B. Shahriari, et al., Taking the human out of the loop: A review of Bayesian optimization, 2016.

[3] K. Kandasamy, et al., Multi-fidelity Gaussian Process Bandit Optimisation, 2019.

[4] Rawlings, et al., Chemical Reactor Analysis and Design Fundamentals, 2002.