Multidomain Hybrid Neuroevolutionary Algorithm for the Optimization of Instant Ultrahigh Hydrogen Evolution of Polaron Nanomaterial Water-Splitting Devices

• Published in: ACS sustainable chemistry & engineering Vol. 13; no. 37; pp. 15601 - 15614
• Main Authors: Lai, Yi-Sheng, Tseng, Yung-Lin, Chen, Min-Chi
• Format: Journal Article
• Language: English
• Published: American Chemical Society 22.09.2025
• Subjects: genetic algorithm; cuprous oxide; photoelectrochemical water splitting; efficiency control; neural network; artificial intelligence
• ISSN: 2168-0485; 2168-0485
• DOI: 10.1021/acssuschemeng.5c06201

An intelligent self-adaptive system for water splitting energy regulation to fit the global environment is established by our neuroevolutionary model and over 1,60,000 research data points. Via ultralow external compensatory bias applied (0.3 V), the performance of the octahedron Cu2O nanoparticles photocathode water splitting is up to 8.32%, and the highest hydrogen evolution volume is as high as 13.22 L/m2·h by truncated octahedron Cu2O nanoparticles. Besides, the CO2 emission is as low as 2.54 × 10–8 kg per generated liter of H2. However, since the characteristic morphology induced a polaron surface state difference between the Cu2O nanoparticles, the various environmental parameters affected the performance of Cu2O photocathode water splitting devices. An artificial intelligence system and a prediction database are established in this study for the efficiency control system of Cu2O water splitting devices for a timely, high, and suitable performance in hydrogen evolution under drastic and fast environmental changes. Multiobject genetic algorithms optimize the mathematical function of the artificial neural network as a neuroevolutionary model. The model performs a 2.62 × 10–10 loss value and aims to precisely predict the performance of Cu2O nanoparticles photocathode water splitting devices. Via this artificial intelligent system, the actual efficiency of our Cu2O water splitting devices deviates from the theoretical prediction efficiency by nearly 0.2%.