| Title |
Design of a Uniformly Distributed Reflector Based on AI Reinforcement Learning |
| Authors |
Jae-Wook Ahn ; Uh-Chan Ryu |
| DOI |
http://doi.org/10.5207/JIEIE.2026.40.1.21 |
| Keywords |
AI reinforcement learning; Concentrated photovoltaic (CPV); Contamination-prone air filters; Inverse reflector design; Uniform illumination distribution |
| Abstract |
In this study, we propose an inverse design method based on reinforcement learning for reflectors, aiming to achieve uniform light distribution on targeted surfaces, including contamination-prone air filters and concentrated photovoltaic solar panels. Conventional inverse reflector design approaches-such as differential equations, 3D vector formulas, and dedicated software-are limited by high computational complexity and the inability to implement beyond constrained structures. To overcome these limitations, we developed a reinforcement learning algorithm that rewards illumination uniformity, enabling the agent to autonomously derive the optimal reflector shape using only a few initial conditions, such as the reflector’s coordinates and distance. Through 5,000 training episodes, we obtained a reflector shape that achieved approximately 53% uniformity. Although this is lower than the 69% uniformity reported in previous studies employing 3D vector equations, it is sufficient for practical applications to air filters and solar panels, while offering the advantage of being implemented through very simple methods. |