Ultra-Wideband, Polarization-Independent, Wide-Angle Multilayer Swastika-Shaped Metamaterial Solar Energy Absorber with Absorption Prediction using Machine Learning

This paper proposes a double layer of gold multipattern swastika (DLMP) resonator based on SiO2 substrate. The average absorption of 95% is achieved for the DLMP metasurface-based solar absorber in the spectrum (0.1-3 mu m) covering the ultraviolet, visible, near-infrared (NIR), and some range of mid-infrared regions which makes proposed solar energy absorber ultra-wideband. The absorptance rate of more than 90% is achieved for the bandwidth of 2516 nm, in absorptance spectrum of 0.314 to 2.830 mu m. Shape analysis is also carried out for proposed structure with simulations of five variations and comparative analysis in terms of absorptance response under solar radiation is also presented to check the effect of shape variation on absorption. Furthermore, the influence of several structural parameters on absorptance spectra is also investigated. It is also observed that the absorptance spectrum of proposed solar absorber is angle insensitive for the range of 0 degrees to 70 degrees and is also polarization insensitive. General regression neural network is used to build regression models which can learn and predict the behavior of absorbers in assorted conditions. Experimental results prove that these models can predict the absorber behavior with high accuracy and can reduce the simulation time, resource requirements by 80%.

Automated summary

A double layer of gold multipattern swastika (DLMP) resonator is proposed for solar energy absorption, achieving an average absorption of over 95% across a wide spectrum in the ultraviolet, visible, near-infrared, and mid-infrared regions. Shape analysis, simulations, and comparative studies are conducted to investigate the impact of structural parameters and shape variations on absorption efficiency, with results showing angle insensitivity and polarization insensitivity. Regression models built using general regression neural networks demonstrate high accuracy in predicting absorber behavior, reducing simulation time and resource requirements by 80%.