Efficient Optimization of High‐Quality Microwave Metasurface Absorbers Using a Random Forest‐Assisted Improved Estimation of Distribution Algorithm

Abstract The design of microwave metasurface absorbers (MMAs) is complicated because numerous design parameters require high‐performance optimization algorithms. In this article, a random forest‐assisted improved estimation of distribution algorithm (RFIEDA) is proposed. RFIEDA aims to obtain a high‐quality MMA design with a limited number of exact expensive evaluations, where the geometric parameters and MMA unit patterns are both optimized. To address the mixed variable nature of geometric parameters and MMA unit patterns, a coding method is presented to transform geometric parameters into binary sequences, effectively making them discrete like MMA unit patterns. A random forest (RF) model is employed to establish a mapping between design variables and the objective function. Improved estimation of distribution algorithm (IEDA) is used to globally search for the combination of MMA unit patterns and geometric parameters, which cooperated with RF to accelerate the convergence speed. Moreover, a model management strategy is introduced to identify candidate solutions from the individuals generated by the IEDA for the exact expensive evaluations. The performance of RFIEDA is demonstrated by a broadband MMA and a triple‐band MMA.