Machine Learning Enabled Smart Design of High-Performance Solar Absorbers and Colored Passive Radiative Coolers

Student thesis: Doctoral Thesis


With the rapid growth of world economy and population, the global demand for heating and cooling is rising day by day, imposing tremendous stress on electricity systems and causing an increasing amount of greenhouse gas emission. To tackle this challenge and move toward carbon neutrality, the development of high-performance solar absorbers and passive radiative coolers have drawn growing attention. However, design of these devices is challenging using expensive, laborious, and time-consuming trial-and-error methods. Herein, machine learning techniques were applied to accelerate the design of nanocomposite solar absorbers and colored passive radiative coolers. Firstly, a Bayesian optimization-based geometrical design approach was proposed for different photonic structures to achieve target spectral characteristics. The comparison with other global optimization methods showed the efficiency and robustness of Bayesian optimization especially when 3D fullwave simulations were involved. Secondly, a mixed-integer memetic algorithm was employed to design solar transparent selective emitter, where the number of layers and the material and thickness of each layer were simultaneously optimized. The optimal selective emitter design achieved a near-ideal selectivity in its thermal emissivity, and thus promised a theoretical net cooling flux of 127 W/m2, corresponding to a sub-ambient cooling temperature of 15 ℃. Thirdly, by combining the optimal selective emitter with a bottom Ag-TiO2-Al color module, sub-ambient daytime radiative coolers with reflective colors were developed, showing vivid appearance with good angular performance. Finally, the monolithic integration of passive radiative coolers with a stepwise nanocavity color module was demonstrated to enable spatial transmissive colors. A tandem neural network was applied to inversely retrieve the geometrical parameters of each pixelated structure for on-demand coloration. This work sheds light on the accelerated design of novel solar absorbers and passive radiative coolers for green energy and sustainability applications.
Date of AwardJul 2022
Original languageAmerican English


  • machine learning; smart design; solar absorbers; sub-ambient radiative coolers; angle-robust reflective color; spatial transmissive color.

Cite this