AI-based orchestration of photochemical reactions for solar fuel production

Since the Industrial Revolution, and particularly in recent decades, growing attention has been devoted to environmental issues and global warming. Scientific and technological progress has therefore increasingly focused on developing new processes and technologies capable of mitigating the rise in global temperature by reducing the concentration of Greenhouse Gases (GHG) in the atmosphere. Among the major GHGs, carbon dioxide (CO2) represents the gas with the greatest overall climatic impact, due to both its high atmospheric concentration and its long residence time. For this reason, recent research has concentrated on the development of Carbon Capture, Utilization, and Storage (CCUS) technologies, which have shown promising results at the experimental level. This work focuses on the optimization of a photocatalytic process for the reduction of CO2 to CH4, employing three newly synthesized porphyrins as catalysts. The main goal is to maximize the reaction yield and increase the amount of useful product obtained from CO2 reduction. In complex chemical systems, the reaction yield depends on numerous experimental variables, and a manual optimization of such parameters would require a large number of experiments, leading to significant time and economic costs. In this thesis, an Artificial Intelligence (AI) based optimization algorithm is implemented to drastically reduce the number of experiments required to identify the optimal reaction conditions. The proposed approach relies on Bayesian Optimization (BO) using Gaussian Process (GP) as the regression model. This methodology has proven to be highly efficient in identifying an optimal balance between turnover number (TON) and selectivity, achieving high values of the objective function with a remarkably small number of iterations. Furthermore, the model successfully identified the variables exerting a positive influence on the process, progressively focusing the search on the most promising regions of the parameter space.