Robust Inverse Design of Microwave Limiter Based on Feature Extraction and Noise Injection

Inverse design based on response data to infer limiter design parameters is crucial for the electromagnetic pulse (EMP) protection of circuit boards. However, inverse design is challenging due to it being a non-convex optimization problem in a high-dimensional space. This study addresses the limiter inverse design problem using feature extraction and data augmentation based on noise injection. The proposed method enhances the model's training by introducing noise, significantly boosting the robustness and performance of the model. Differing from traditional methods requiring multiple characteristic curves, the proposed method can provide accurate prediction by only a single burnout time and temperature curve. The comparison with other machine learning methods proves the significance of the proposed method.