Machine Learning Prediction Framework for Tailoring the Optical Response of Particulate Media

Abstract

Accurate and efficient prediction of the reflectance of particulate media is crucial for advancing optical technologies. However, traditional reflectance prediction methods often struggle to balance precision with computational efficiency, limiting material design and optimization, especially for large-scale systems. Here, we developed a novel reflectance prediction framework based on the Monte Carlo method (MCM) using a machine learning (ML) strategy. This framework addresses the challenges of low computational accuracy at high particle concentrations and inefficiency in predicting high reflectance in conventional MCMs, achieving simultaneous improvements in both accuracy and efficiency. This realization comes from the mapping of the relationship between input optical features and output reflectance in MCMs by ML and the development of a new experimentally dependent scattering correction model based on this mapping. Rigorous experimental validation and numerical simulations demonstrate the framework’s accuracy, reliability, and versatility across a variety of particulate systems. Furthermore, we applied this framework to create a high-throughput optimization algorithm tailored for radiative cooling applications, effectively guiding the optimization of representative ZrO₂/PDMS films and showcasing the framework’s practical potential. Overall, our approach significantly accelerates the optimization of particulate media, paving the way for the development of innovative materials with tailored optical properties.