Tuning phonon transport via complex nanostructure design using an X+EA hybrid optimization strategy

With the widespread application of nanomaterials in micro-nano devices, phonon transport control has become an indispensable part of thermal management. In this work, we propose an X+EA hybrid optimization strategy based on evolutionary algorithms (EAs) to regulate phonon transport across nanostructures. Due to the inconvenient hyperparameters tuning associated with EAs, we conduct grid searches on the hyperparameter configurations of three typical EAs: genetic algorithm (GA), particle swarm optimization (PSO), and simulated annealing (SA), resulting in more suitable parameter ranges. Compared to previous nanostructure design work using Bayesian optimization (BO) and Monte Carlo tree search (MCTS), we find that EAs exhibit robust structural design capabilities and extremely low optimization costs, particularly the GA with the fewest hyperparameters. Furthermore, we develop a highly compatible X+EA hybrid optimization framework, where X can be substituted with the optimization results of any algorithm and used as the initial structures in the subsequent running of EAs. Among various hybrid optimizations with nonstrict hyperparameter settings, the global search capability of BO+GA even rivals that of BO, while reducing optimization time by over 95%. And MCTS+GA effectively resolves the issue of MCTS alone yielding only a few local optima. The proposed X+EA framework can adapt to the nanostructure design with high degrees of freedom or large candidates.