GENETIC ALGORITHM WITH DEEP NEURAL NETWORK SURROGATE FOR THE OPTIMIZATION OF ELECTROMAGNETIC STRUCTURE

In this current technological era, engineering design process encounters with
multiple complexity as it depends on computer simulation to ensure accurate model
before the model is actually constructed. This leads to the problems where the
engineers have to feed their simulation with new inputs and parameters to get the best
solution. Specifically, in electromagnetic field, bidirectional scattering distribution
function of a diffraction grating is computed using MEEP simulation and requires
numerous numbers of parameters. This paper will report on an initial study of the usage
of Genetic Algorithm (GA) merged with Deep Neural Network based surrogate model
to optimize simulation for electromagnetic structure. The behavior of Genetic
Algorithm (GA) where it generates and evolves the parameters towards a high-quality
solution gives an advantage in obtaining ideal combination of parameters to fit in with
the simulation. The GA will be assisted with the convergence of deep neural network
to increase the performance by reducing the computational time.