High performance microstrip array design by machine learning-assisted multi-objective optimization approach

In this work, we design a high-performance microstrip patch antenna using an efficient machine learning (ML)-assisted framework. A Light Gradient Boosting (LightGBM) model is employed to predict the electromagnetic behavior of both the microstrip array and its wide-angle impedance matching (WAIM) structure. The ML models are trained on a dataset generated using full-wave simulations, significantly reducing evaluation time while maintaining high accuracy. To optimize the loaded array's structural parameters, a multi-objective particle swarm optimization (PSO) algorithm is employed to ensure a well-balanced trade-off among impedance matching, peak gain, and side lobe level (SLL). The proposed approach achieves substantial improvements in these key parameters while significantly reducing evaluation time, with each structure taking approximately 10.62 seconds to assess - far less than the time required for conventional design approach. The results confirm the effectiveness of the ML-assisted optimization in designing high-performance microstrip arrays with enhanced scanning capabilities.