Application of machine learning to optimized design of layer structured particles

Abstract

Direct absorption solar collectors are a solution to the problem of energy depletion: they work by dispersing plasmonic nanoparticles in a liquid and exposing them directly to sunlight. Improving the sunlight absorption performance of plasmonic nanoparticles is an important issue to increase their utilization efficiency and reduce their production cost. To solve this problem, we have proposed metal-insulator-magnet plasmonic nanoparticles with a layered structure consisting of a spherical insulator sandwiched between thin films. These particles can be easily fabricated by sputtering a thin film on a spherical insulator, which significantly reduces the amount of material used and the process is inexpensive. However, the combination of factors that determine the radiative properties of the particles is enormous. Therefore, the goal of this study is to find the optimal particle design using machine learning. Three types of machine learning were used: neural networks, support vector machines, and light gradient boosting machines. Learning is done by performing an electromagnetic field analysis based on the finite element method and using the calculated radiative properties as the correct values. The accuracy of the machine learning was evaluated by predicting the absorption property from the particle parameters. The constructed machine learning code was then used to optimize the particle parameters. It was shown that machine learning is effective for optimization design of objects with a large number of parameters.