Machine Learning Inverse Design Reveals a Double Narrow-Band Absorption Approach for Effective Colored Radiative Cooling Paints.

Colored radiative cooling paints offer immense potential for aesthetic thermal management. However, previous studies have been limited to either intuition-driven forward designs that are too expensive to yield non-intuitive optimal designs or to optimizations only at the level of idealized optical spectra. We present a machine-learning-enabled inverse design framework for generating paint formulations with optimal cooling performance and desired color. By integration of photon Monte Carlo simulations with surrogate modeling, this approach bridges the gap between theoretical spectra and physical realization. Our approach uncovers a non-intuitive "double narrow-band absorption" strategy that reduces solar heating power by up to 193 and 37 W/m2 compared to conventional additive and single-band subtractive strategies, respectively. A systematic exploration of the HSL color space reveals that 24% of colors can achieve daytime subambient cooling when realistic material limitations are considered. This automated pipeline provides practical guidelines for designing high-performance, aesthetically tailored radiative cooling coatings.