AI-driven optimization of photocatalytic hydrogen production: Integrating techno-economic analysis and regional environmental constraints

Photocatalytic (PC) hydrogen production offers a promising pathway for sustainable hydrogen generation. However, most existing studies mainly aim to increase the hydrogen evolution rate (HER), overlooking that high-performing catalysts are not necessarily economically viable. Real-world deployment requires simultaneous optimization of catalyst performance and system-level economics under varying environmental conditions. To bridge this gap, this study introduces an AI-integrated framework combining surrogate modeling of HER with techno-economic optimization and regional constraint analysis. Thirty-five machine learning models were benchmarked, and the Voting CGF ensemble achieved the best predictive performance (R2 = 0.9427, MSE = 0.0009), with conduction band position (Ec) identified as the most influential feature. This framework was used to evaluate 1320 catalyst designs comprising four photocatalysts, 30 cocatalysts, and 11 sacrificial agents. Simultaneous optimization of LCOH and HER uncovers distinct catalyst design strategies often overlooked by conventional HER-focused approaches. Notably, [g-C3N4, Group A, None] combination achieved a levelized cost of hydrogen (LCOH) as low as $0.487/kg, challenging the idea that sacrificial agents are always needed for economic viability. Furthermore, regional analysis also provided a critical insight: locational constraints can be more impactful than catalyst choice. Freshwater availability was the most critical constraint, with LCOH varying more than 13-fold under identical catalyst designs. This novel framework enables rapid identification of catalyst designs satisfying simultaneous HER, LCOH, and regional requirements without repeated experiments, paving the way for scalable and cost-effective PC hydrogen production.