Parameterized Surrogate Models of Electromagnetic Systems Through Decision Tree and Random Forest Models

Simulations are an integral part of the electromagnetic systems' design and optimization process. Mathematical models associated with them are usually very large, time and memory-consuming. It is useful to have faster surrogate models with an accuracy comparable to full-order models, especially in an optimization process. For this purpose, we propose to use Decision Trees and Random Forests as parametrized surrogate models of electromagnetic systems. We focus on approximating the parameter-dependent transfer functions using Decision Trees and Random Forests. We used limited data of the module of transfer functions obtained from PEEC simulations to train models appropriately sampled in the parameter space with Latin Hypercube sampling. Trained models predict Transfer Function Modules at any parameter sample in the design space domain with good accuracy.