Designing radiative cooling metamaterials for passive thermal management by particle swarm optimization

Abstract

The passive radiative cooling technology shows a great potential application on reducing the enormous global energy consumption. The multilayer metamaterials could enhance the radiative cooling performance. However, it is a challenge to design the radiative cooler. In this work, based on the particle swarm optimization (PSO) evolutionary algorithm, we develop an intelligent workflow in designing photonic radiative cooling metamaterials. Specifically, we design two 10-layer SiO2 radiative coolers doped by cylindrical MgF2 or Air impurities, possessing high emissivity within the selective (8-13 µm) and broadband (8-25 µm) atmospheric transparency window, respectively. Our two kinds of coolers demonstrate power density as high as 119 W/m2 and 132 W/m2 at the room temperature (300 K). Our scheme does not rely on the usage of special materials, forming high-performing metamaterials with conventional poor-performing components. This significant improvement of the emission spectra proves the effectiveness of our inverse design algorithm in boosting the discovery of high-performing functional metamaterials.

Machine Learning Aided Design and Optimization of Thermal Metamaterials

“…PSO offers the advantage of easy implementation, few hyperparameters to be tuned, and faster convergence. It has been successfully applied in multiobjective optimization scenarios. − As the algorithm operates at the particle level, careful attention should be given to tuning both behaviors of individual particle and the entire swarm. A complete iteration corresponds to the position update of each individual particle.…”

Inverse Design of Phononic Crystal with Desired Transmission via a Gradient-Descent Approach

“…With the advancement of various machine learning and intelligent algorithms, researchers can more conveniently use computer-aided methods to predict the thermal transport properties of materials [20] or develop new thermal metamaterials. [21,22] For example, multi-objective optimization, [23] active machine learning, [24] genetic algorithm optimization, [25] and particle swarm optimization [26] have been applied in this field. However, machine learning often necessitates a substantial training dataset, resulting in time-consuming learning processes.…”