AI-Assisted Antenna Optimization: Integrating Evolutionary and Inverse Cascade Neural Networks With Differential Evolution

Combining artificial intelligence (AI) technology with antenna optimization has long been a research direction with significant potential. However, most contributions in this area have focused on constructing surrogate models for antenna optimization using supervised machine learning. These surrogate models can quickly evaluate antenna design solutions but cannot directly modify the design variables to drive optimization. To address this limitation, this article designs two neural networks to assist the differential evolution (DE) algorithm in optimizing antennas. The first is an evolutionary neural network (ENN), which learns the modification patterns of DE on antenna design variables, thereby enhancing the probability of producing more optimal solutions. The second is an inverse cascade network, constructed by first creating multiple inverse subnetworks for each frequency point and then synthesizing the outputs of these subnetworks using a cascade network. This type of neural network can directly output high-quality solutions based on specific requirements. Finally, by integrating these two networks with DE, an AI-assisted antenna optimization method is realized. Experimental results show that with the assistance of these two neural networks, DE can achieve faster optimization of antennas.