Variable-Fidelity Surrogate Model-Based Machine Learning-Assisted Optimization and Its Application to Worst-Case Performance Searching of Antennas

摘要

An efficient worst-case performance (WCP) searching method is proposed using variable-fidelity surrogate model-based machine learning-assisted optimization (MLAO). First, variable-fidelity responses of a microwave device or antenna are achieved using full-wave electromagnetic simulations under different configurations. Next, these responses together with input parameters are used to train a surrogate model. Then, the WCPs of responses are searched in the input tolerance regions of parameters using an evolutionary algorithm. Compared with the conventional method, the total computational time of the proposed method is dramatically reduced. Finally, a mobile phone antenna is simulated to validate the proposed method.