Project overview
Monte Carlo simulation is a standard tool for quantifying uncertainty in electrical circuits, but it can become expensive as the number of uncertain parameters grows. This project investigated support vector machine and least-squares SVM regression as surrogate models for approximating the output distribution of an RLC circuit. I implemented the models in MATLAB, generated training inputs with Latin hypercube sampling, and studied sensitivity to training-set size, kernel choice, and model hyperparameters. An iterative training-data reduction procedure produced an output distribution close to a 5,000-point Monte Carlo reference using 120 training samples at a validation-error tolerance of 10−5.