Modeling and Prediction of Additional High Frequency Harmonic Losses Created by PWM Supply Utilizing Machine Learning

This paper presents an innovative methodology for modeling and predicting high-frequency harmonic losses in the rotating field of an electrical machine powered by a frequency converter. Conventional approaches are often unreliable because these losses depend heavily on switching frequency, modulation index, load conditions, and transient inductance, all of which can vary significantly. To address this challenge, a data-driven surrogate modeling approach based on Gaussian process regression is proposed. The training data were obtained from measurements on three induction machines with rated powers of 15 kW, 37 kW, and 75 kW, all tested under the same relative load conditions. The frequency converter was configured with varying reference switching frequencies, and under each condition, the additional high-frequency harmonic losses induced by the pulse-width modulation supply were quantified through waveform analysis. Model performance was validated using a portion of the measurement data reserved for testing. The results demonstrate that the proposed methodology predicts additional harmonic losses with high accuracy and provides valuable insights for the design of more efficient inverter-fed drive systems.