Particle-Swarm-Optimization-Based Selective Neural Network Ensemble and Its Application to Modeling Resonant Frequency of Microstrip Antenna

From communication systems to biomedical systems, microstrip antennas (MSAs) are used in a broad range of applications, and this primarily due to their simplicity, conformability, low manufacturing cost, light weight, low profile, reproducibility, reliability, and ease in fabrication and integration with solid-state devices [1][2]. Recently, these attractive features have increased the applications of MSAs and stimulated greater effort to investigate their performance. In designing MSA, it is very important to determine its resonant frequencies accurately, because MSA has narrow bandwidths and can only operate effectively in the vicinity of the resonant frequency. So, a model to determine the resonant frequency is helpful in antenna designs. Several methods, varying in accuracy and computational effort, have been proposed and used to calculate the resonant frequency of rectangular MSA [3]-[13]. These methods can be broadly classified into two categories: analytical and numerical methods. Based on some fundamental simplifying physical assumptions regarding the radiation mechanism of antennas, the analytical methods are the most useful for practical design as well as providing a good intuitive explanation of the operation of MSAs. However, these methods are not suitable for many structures, in particular, if the thickness of the substrate is not very thin. The numerical methods provide accurate results but usually require tremendous computational effort and numerical procedures, resulting in roundoff errors, and may also need final experimental adjustment to the theoretical results. They suffer from a lack of computational efficiency, which in practice can restrict their usefulness due to high computational time and costs. The numerical methods also suffer from the fact that any change in the geometry, including patch shape, feeding method, addition of a cover layer, etc., requires the development of a new solution. During the last decade, artificial neural network (NN) models have been increasingly used in the design of antennas, microwave devices, and circuits due to their ability and adaptability to learn, generalization, smaller information requirement, fast real-time operation, and ease of implementation features [14][15]. Through training process, a NN model can be developed by learning from measured/simulated data. The aim of the training