Model‐Based Bayesian Optimization for Organic Photovoltaics: Combining Bayesian Optimization With Physical Domain Knowledge

ABSTRACT Efficient process optimization is of crucial importance in materials science and industrial manufacturing. Bayesian optimization has proven to be an effective method for optimizing time‐consuming experiments; however, it is typically carried out as a black‐box approach without utilizing prior physical knowledge. In this work, we apply a model‐based Bayesian optimization approach in which a physical model of the solar cell is integrated into the optimization process. Simultaneous exploration and exploitation of the parameter space is enabled by the acquisition function, as is customary for Bayesian Optimization. We show that the so‐called Knowledge Gradient fits particularly well for the model‐based approach. The method was validated both experimentally, using the PTQ10:BTP‐eC9 material system, and statistically, using a benchmark function specifically developed for organic photovoltaics. The results show that the model‐based approach is superior to the conventional black‐box approach and that the global optimum can be identified more reliably. This work demonstrates the potential of integrating domain knowledge into Bayesian optimization algorithms and opens up new perspectives for accelerated process optimization in photovoltaic research.