Metasurface-based solar absorber with absorption prediction using machine learning.

We propose a metasurface-based solar absorber design using machine learning. The absorption analysis for two complementary metasurface-based designs is presented. The complementary circular array metasurface design and circular array metasurface design are investigated for visible, infrared, and ultraviolet regions with a wavelength range of 0.2 μm–0.8 μm. The highest average absorption is observed for complementary circular array metasurface design. This design is further investigated for different parameter variations like resonator thickness, SiO 2 height, SiO 2 width, and graphene chemical potential. All these parameter variations and their absorption results are also predicted with machine learning algorithms. Experiments are performed using different degrees of polynomial regression models for predicting the absorption values for missing/intermediate wavelengths. Two test cases R-30 and R-50 are designed to test the regressor models and R2 Score metric is used to test the prediction efficiency of the regressor model. It is observed from the experimental results that, prediction efficiency of more than 0.99 (R2) can be achieved using higher degree polynomial regression models. The proposed metasurface-based solar absorber design can be used for solar energy harvesting applications. • Highly efficient metasurface based solar absorber is designed. • Machine learning approach is used for prediction of absorption for variation in different parameters. • Polynomial regressor is used for predicting the values. • Highest average absorption of 89% is achieved for visible region. • The highest prediction efficiency of 0.99 (R2) is achieved. [ABSTRACT FROM AUTHOR]