Enhancing overall performance of thermophotovoltaics via deep reinforcement learning-based optimization

• DOI: 10.1063/5.0213211
• Publication Date: 2024-07-08T09:58:35Z
• Authors: Shilv Yu; Zihe Chen; Wentao Liao; Cheng Yuan; Bofeng Shang; Run Hu
• Keywords: Thermophotovoltaic

Thermophotovoltaic (TPV) systems can be used to harvest thermal energy for thermoelectric conversion with much improved efficiency and power density compared traditional photovoltaic systems. As the key component, selective emitters (SEs) re-emit tailored radiation better matching absorption band of TPV cells. However, current designs SEs heavily rely on empirical design templates, particularly metal-insulator-metal (MIM) structure, lack considering overall performance optimization efficiency. Here, we utilized a deep reinforcement learning (DRL) method perform comprehensive 2D square-pattern metamaterial SE, simultaneous material selections structural parameters. In DRL method, only database refractory materials gradient refraction indexes needs prepared in advance, whole roadmap will automatically output SE optimal Figure-of-Merit (FoM) efficiently. The is composed novel combination TiO2, Si, W substrate, its thickness structure precisely optimized. Its emissivity spectra match well external quantum curve GaSb cell. Consequently, significantly enhanced an 5.78 W/cm2, 38.26%, corresponding FoM 2.21, surpassing most existing designs. underlying physics explained by coupling effect multiple resonance modes. This work advances practical application potential paves way addressing other multi-physics problems