A Hybrid Self-Adaptive Differential Evolution Algorithm With Simplified Bayesian Local Optimizer for Efficient Design of Antennas

• Published in: IEEE transactions on antennas and propagation Vol. 73; no. 1; pp. 391 - 404
• Main Authors: Gao, Tian-Ye, Jiao, Yong-Chang, Zhang, Yi-Xuan, Zhang, Li
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.01.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Adaptation models; Adaptive algorithms; Algorithms; Antenna design; Antenna optimization; Antennas; Bayes methods; Bayesian analysis; Bayesian optimization (BO); Computational modeling; Design optimization; differential evolution (DE) algorithm; Evolutionary algorithms; Evolutionary computation; Gaussian process (GP); local optimization; Optimization; Prediction algorithms; Predictive models; Simulation; Switches; Training
• ISSN: 0018-926X; 1558-2221
• DOI: 10.1109/TAP.2024.3501406

For antenna optimization, computationally expensive full-wave EM simulations are necessary, making efficient design of antennas a challenge. Since there are only a few local minimums, some existing algorithms without considering this feature need a lot of useless EM simulations, leading to poor optimization efficiencies. In this article, a hybrid self-adaptive differential evolution (SADE) algorithm with a simplified Bayesian local optimizer (SBLO) (SADE-SBLO) is proposed for improving antenna optimization efficiencies, in which the SADE is used to generate the offspring population. The algorithm also consists of the following four modification strategies: 1) an individual parallel prediction method for reducing surrogate model training (SMT) and prediction times; 2) an offspring quality pre-assessment method for improving offspring quality and further reducing the number of EM simulations; 3) a self-adaptive database increment method for adapting the algorithm to different optimization stages and also serving as a start-up switch for the local optimizer; and 4) an SBLO for improving optimization efficiency in the later stage. These strategies are closely integrated to make the algorithm better balance exploration and exploitation, reduce useless EM simulations, and converge faster. Four representative antenna cases are optimized. Compared with some existing algorithms such as DE and the surrogate model-assisted differential evolution algorithm (SADEA), the proposed algorithm is efficient.