Machine learning-enabled design of metasurface based near-perfect daytime radiative cooler

Abstract

Global warming and the energy crisis are major challenges facing the world today. Traditional structural design methods for radiative coolers are no longer sufficient to meet the increasingly demanding realities of these challenges. Simplifying the design process and optimizing structural parameters have become pressing issues. In this paper, a new optimization method for metasurface based daytime radiative cooler is proposed, which utilizes the K-nearest neighbor (KNN) algorithm in machine learning (ML) to construct regression models and optimize designs for two structures of quadrangular prismatic metasurface (QPM) and circular truncated cone metasurface (CTCM). The results show that the atmospheric window emissivity (mainly in the 8–13 μm range) of QPM and CTCM reaches 96.98% and 97.81%, respectively, while the absorptivity/emissivity in solar spectrum is only 9.34% and 7.21%. The mean absolute percentage error (MAPE) of the KNN regression model is only 0.74% and 0.29%, which is significantly better than other classical ML algorithms. Meanwhile, the designed structures achieve net cooling power of 63.48 W/m2 and 90.13 W/m2 at ambient temperature for both QPM and CTCM. This systematic study provides a novel approach for the design of daytime radiative coolers.