This study presents a novel Adaptive Neuro-Fuzzy Inference System (ANFIS)-based modeling framework for printed folded dipole antennas, addressing a significant gap in computational intelligence applications for antenna design. By generating a comprehensive dataset of 1000 antenna configurations through full-wave electromagnetic simulations and systematically varying substrate dimensions, a surrogate model capable of accurately predicting both resonance frequency and minimum return loss (S11) is developed by ANFIS models. Recognizing that derivative-based algorithms such as Backpropagation (BP) and Hybrid Algorithm (HB) often face challenges such as local minima and unstable gradient computations when training ANFIS models, a hybrid ANFIS-PSO approach integrated with Particle Swarm Optimization (PSO) is proposed, yielding more accurate and reliable estimations of folded dipole antenna (FDA) parameters than traditional methods to address these issues. Within the scope of simulation studies, four distinct optimization algorithms, namely BP, HB, Genetic Algorithm (GA), and PSO, are rigorously compared for ANFIS training, with population-based methods significantly outperforming gradient-based approaches. The proposed ANFIS-PSO model demonstrates superior performance across all evaluation metrics, achieving Root Mean Square Error (RMSE) values of 0.0082 GHz for resonance frequency prediction and 0.0047 dB for S11 prediction, with Coefficient of Determination (R²) values exceeding 0.96 for both parameters. The proposed framework offers several key advantages over existing ANFIS-based antenna modeling studies, including a novel application domain (folded dipole antennas), multi-output modeling capability, and superior predictive performance, positioning it as a valuable tool for rapid design exploration and optimization of folded dipole antennas for 2.4 GHz applications.
Haznedar et al. (Thu,) studied this question.