Surrogate-Assisted Multi-State Tuning-Driven EM Optimization for Tunable Cavity Filter

This paper provides a surrogate-assisted multi-state tuning-driven electromagnetic (EM) optimization technique for the tunable evanescent mode cavity filter. The design of the tunable filters needs to be tuned to satisfy multiple tuning states (i.e., tuning states with different center frequencies and bands) simultaneously. A novel surrogate-assisted multi-state tuning-driven EM optimization technique that employs artificial neural networks (ANN) is used to address the challenge. This surrogate model is composed of several sub-surrogate models. Each sub-surrogate model is developed to perform the optimization for each tuning state. All the tuning states are optimized simultaneously. Thereby, there is a higher probability of finding a solution that satisfies all tuning states. Furthermore, parallel computation scheme is exploited to generate the training samples to improve the time cost. A new trust-region updating formulation specifically for multi-state tuning is proposed to improve the convergence of the optimization algorithm.