A Sensitive Data Labeling Strategy for Optimizing a Broadband Vertical Transition in W Band

This letter proposes a sensitive data labeling method for the automated design and optimization of a W-band via-hole vertical transition structure. First, the dynamic thresholds are introduced to identify sensitive regions of return loss (RL). A surrogate model based on artificial neural networks (ANNs) is then developed to establish the mapping between geometric parameters and $\vert S_{11}\vert $ , with its validity demonstrated through the presentation of cases. Finally, optimization results, which maintain high prediction accuracy while reducing optimization time by 38.29% and improving the 30-dB RL bandwidth by 8.6 GHz compared with conventional methods, are obtained using genetic algorithm (GA), thereby demonstrating the effectiveness of the proposed data labeling method for modeling and optimization.