Cost-Aware Bayesian Optimization for Self-Driven Condition Screening of Flow Electrosynthesis
Runzhe Liang, Siyuan Zheng, Kai Wang, Zhihong Yuan
Abstract
Flow electrosynthesis has attracted increasing attention as a green and sustainable manufacturing method. However, it is still a challenging undertaking to determine the appropriate experimental conditions for different electrosynthesis systems. As an AI-guided decision-making method, Bayesian optimization (BO) suffers a lot from neglecting the characteristics of chemical reactions and one-by-one low-speed screening processes when it is used to identify the optimal conditions. Herein, we design a cost-aware multi-objective Bayesian optimization framework to seek the highest current density for the electrosynthesis of tetrabenzylthiuram disulfide (TBzTD), while keeping the yield and Faraday efficiency at high levels. The optimization was finished within only 4 h including 11 experiments, and the optimized current density became 200% as high as the maximum of trial-and-error experiments, which efficiently reduced the experimental cost compared to the standard BO that requires a few days. The approach shows the possibility of utilizing the least experimental budgets for fully optimal condition screening, significantly reducing the burden on trial-and-error experimental expenses.