Machine-learning-empowered FDTD/FEM simulations for predictive solar energy absorption in plasmonic metamaterial...

Metamaterials (MTMs) exhibit significant potential for solar energy harvesting; however, achieving optimal performance remains challenging with numerical simulations alone. The novelty of this work lies in the integration of high-accuracy electromagnetic simulations with machine learning (ML) techniques to design and predict the optical behavior of a plasmonic MTM solar absorber. The proposed nanostructure, consisting of TiN plasmonic nanocavity arrays topped with a TiO₂ thin layer, demonstrates an impressive 95.44% absorption efficiency across the 200-1300 nm wavelength range with a minimal thickness of 330 nm. This high efficiency is attributed to strong coupling among surface plasmon polaritons (SPPs), localized surface plasmon resonances (LSPRs), and nanocavity modes, as elucidated through finite-difference time-domain (FDTD) and finite element method (FEM) simulations. The simulation results are corroborated with experimental data from the literature, confirming the validity of the model. The proposed MTM is polarization-insensitive, scalable, and cost-effective, maintaining stable absorption performance at oblique incident angles up to 50°. To predict optical absorption characteristics, we implement an ML approach of Particle Swarm Optimization-Adaptive Neuro-Fuzzy Inference System (PSO-ANFIS) and Genetic Programming (GP). The PSO-ANFIS model achieves high predictive accuracy, with an R² value of 0.92, an average absolute relative deviation of 2.37%, and a percent bias of 0.14% for test data. GP introduces an innovative yet computationally simple mathematical equation for the prediction of optical absorption. This integration of ML with electromagnetic simulations not only streamlines the responses and design optimization process but also provides new insights into the mechanisms of light absorption, paving the way for next-generation solar energy harvesting technologies.