Comparison of Selected Support Vector Machine Approaches for Stochastic Power Electronic Circuit Simulation with Parasitics

This paper provides a comparison between op-timization methods used for tuning the hyperparameters of Support Vector Machine model in a stochastic circuit simulation for conducted interference. The methodology is used to create a surrogate model of the frequency and amplitude of the dominant mode of the interference, which is a result of presence of parasitics in the considered switching circuit. Optimization algorithms are compared by obtaining the computational time and by computing a posteriori error of their predictions. The best optimization algorithm in the example provided here is found to be the quasi-Newton Broyden–Fletcher–Goldfarb–Shanno algorithm.