BayesianKAN: A reaction condition optimization framework integrating Kolmogorov-Arnold network and Bayesian...

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Machine Learning

Background:

• Traditional chemical reaction optimization methods are inefficient and lack predictive accuracy.
• Interpreting complex reaction condition effects remains a challenge.

Purpose of the Study:

• To develop a novel framework for optimizing chemical reaction conditions using Kolmogorov-Arnold Networks (KAN) and Bayesian Optimization (BO).
• To enhance reaction yield and selectivity while improving model interpretability and predictive accuracy.

Main Methods:

• Integration of the Kolmogorov-Arnold Network (KAN) for accurate condition-outcome mapping.
• Application of Bayesian Optimization (BO) for iterative refinement of reaction conditions guided by the KAN model.
• Implementation of the framework as the BayesianKAN software.

Main Results:

• KAN demonstrated superior fitting accuracy and generalization compared to response surface methodology and seven other machine learning models.
• The BayesianKAN framework was validated in H2O2 synthesis and flavone production.
• Significant yield increases were achieved: 60.7% for H2O2 and 5.2% for flavones.

Conclusions:

• The proposed BayesianKAN framework offers an efficient and accurate approach to chemical reaction optimization.
• This method effectively handles sparse data and provides interpretable models.
• BayesianKAN shows significant potential for advancing chemical synthesis and process development.