Machine Learning-Enabled Inverse Design of Radiative Cooling Film with On-Demand Transmissive Color

Abstract

Passive radiative cooling is a cost-efficient and eco-friendly approach to cool terrestrial objects by dissipating heat to the outer space. Colored radiative cooling (CRC) has many advantages over conventional passive radiative cooling and has garnered growing interest recently. However, existing CRC films are normally opaque, where the incident sunlight is either reflected for rendering color or absorbed to generate waste heat. In this work, we design a transmissive CRC film that allows a specific portion of light to pass through and provides more vivid colors. Such a transmissive film achieves the coloration and cooling dual-function by stacking a solar transparent selective emitter on top of a nanocavity-based color filter. The top emitter is first designed by using a mixed-integer memetic algorithm, where the layer materials, their number sequence, and thicknesses are simultaneously optimized. The variability in both material composition and layer thickness enables the emitter with a near-ideal emissivity in the atmospheric windows for subambient cooling, and an ultrahigh transmissivity in the solar range for sunlight penetration. Then, the structures of bottom nanocavity are determined by using a tandem neural network for on-demand color generation. This machine learning-assisted inverse design approach provides real-time structure prediction for on-demand colors and offers great flexibility in balancing the cooling and coloring functionalities. The proposed methodology can have special significance in broadening the application of passive radiative coolers in energy-efficient buildings, power-generating windows, and sustainable greenhouses.