Reinforcement Learning Approach to High-Efficiency Thermophotovoltaic Filter Design

Area of Science:

• Energy Conversion
• Materials Science
• Optical Engineering

Background:

• Thermophotovoltaic (TPV) systems efficiently convert thermal radiation to electricity, especially from waste heat.
• Optimizing TPV efficiency requires precise spectral matching between the thermal source and photovoltaic cell.
• Designing multilayer optical filters for spectral control presents complex optimization challenges.

Purpose of the Study:

• To develop a deep reinforcement learning (DRL) framework for designing high-performance multilayer optical filters for TPV systems.
• To enable selective transmission of photons with energies above the photovoltaic bandgap.
• To maximize TPV system power conversion efficiency by reflecting unwanted photons.

Main Methods:

• Integration of transfer matrix method simulations with a DRL framework.
• Utilization of a customized reward function to guide filter design.
• Incorporation of the detailed balance model to predict system performance.

Main Results:

• Demonstration of DRL-designed filters approximating ideal spectral profiles.
• Prediction of TPV efficiencies exceeding 50% for silicon PV cells.
• Achieved high efficiencies at emitter temperatures below 1500 °C.

Conclusions:

• DRL offers a scalable, data-driven approach for designing advanced optical components in TPV systems.
• The developed framework effectively addresses the spectral matching challenge in TPV design.
• This method paves the way for next-generation, high-efficiency energy conversion systems.