Machine learning-optimized graphene-integrated refractory metasurface based absorber for broadband solar thermal energy conversion

A hierarchically nested graphene-integrated refractory metasurface absorber is designed and optimized for broadband solar-thermal conversion over 0.20–3.0 μm. The structure comprises five concentric resonator tiers (strontium, caesium, GaAs/InAs, tungsten, and Fe₃O₄) placed above a gold ground plane, with a monolayer graphene interface (0.34 nm) enabling tunable conductivity and enhanced inter-resonator coupling.Full-wave simulations in COMSOL Multiphysics v6.2 show near-unity absorption, with peak values > 99.99% at λ₁ ≈ 240 nm, λ₂ ≈ 1267 nm, and λ₃ ≈ 2897 nm, and an average absorptance of 98.58% across the spectrum. Angular analysis indicates TM absorption > 96.99% at 50° and > 89.41% at 60°, while TE polarization remains > 99.48% up to 30°.Impedance retrieval via the Nicholson–Ross–Weir method confirms near-unity normalized impedance, indicating impedance matching as the dominant absorption mechanism. A Random Forest surrogate model trained on 680 simulations (80/20 split, 5-fold cross-validation) predicts full spectra (512 points) from graphene thickness and incidence angle with R² > 0.95, outperforming Gradient Boosting (0.961), ANN (0.943), SVR (0.912), and Linear Regression (0.731). The design is suitable for concentrated solar power and high-temperature thermal applications.