A Novel Machine Learning-Assisted Optimization Approach for Flexible Wearable Antenna

With the increasing popularity of wearable devices, maintaining the performance stability of flexible antennas under complex deformations has become a pressing issue. To address this challenge, we propose a machine learning-assisted optimization (MLAO) framework that integrates advanced neural network models with evolutionary algorithms. First, we redesign a multislot antenna with flexible materials and analyze its performance under various deformation conditions, including bending and crumpling. Then, we propose an attention- and residual connection-based convolutional neural network that predicts electromagnetic responses from antenna images with over 95% accuracy, together with an improved multilayer perceptron compensation model that reduces training data generation cost by 45.1%. Finally, we formulate the average return loss as the optimization objective of the genetic algorithm within the proposed framework. Simulation results demonstrate that the optimized antenna improved frequency resonance under various deformations and achieved a 43.02% increase in bandwidth, significantly enhancing performance stability. The optimized antenna is fabricated and tested, thereby validating the effectiveness of the proposed MLAO approach. Moreover, the framework can be readily extended to other antenna types, offering a general solution for the robust design of flexible antennas.