Metamaterial waveguide modeling by a genetic algorithm-based artificial neural network

Abstract

The need to predict data with less computational effort lead to studies to improve the training algorithms and Artificial Neural Networks (ANN) architectures. We demonstrated how to improve the performance and to increase the efficiency of the ANN by implementing a hybrid model where the parameterization is made by Genetic Algorithm (GA). In this work we applied the hybrid GA-ANN model for the analysis and synthesis of metamaterial based waveguides, composed of metallic and dielectric thin layers claddings surrounding a dielectric core. The parameter to be computed is the length propagation as a function of wavelength, metal filling ratio, refractive indexes of the metal and dielectric layers, waveguide core width and core refractive index. For the synthesis, one of the geometrical parameters becomes the ANN output. The proposed GA ANN was capable of predicting the propagation characteristics of new unseen configurations of metamaterial waveguides with low relative error.