Optimal modulation methods, such as optimized pulse patterns (OPPs), are traditionally computed offline using specific design criteria and system considerations. In doing so, optimal performance can be achieved, as quantified, for example, by the load current total harmonic distortion. However, because these methods are tailored to the system of interest at the time, their offline nature limits adaptability, while real-world nonidealities can result in deviations from the optimal performance.

This project aims to develop real-time, learning-based optimal modulation methods. By learning from the system characteristics in real time, these methods will enable optimal performance across a wide range of operating conditions while ensuring the full utilization of the available hardware.