A Hierarchical Framework for Large-Scale Generation of Multidimensional Optimized Pulse Patterns for SHC-PWM

Abstract

The practical adoption of multidimensional optimized pulse patterns is hindered by the massive computational cost of generating their offline lookup tables (LUTs). This article introduces a hierarchical framework that systematically dismantles this computational bottleneck. The method leverages a data-driven sensitivity analysis to guide an adaptive solver, which strategically applies robust global solvers only to the few highly sensitive regions of the solution space while rapidly solving the vast majority of points with a fast local solver. This adaptive policy, which includes robust fallback mechanisms, is shown to be both fast and highly reliable. Comprehensive benchmarks, through experimental validation, show the framework can generate LUTs with over 11 million points, achieving a success rate greater than 99.5 %. For a baseline case, it is approximately 24 times faster and more robust than a state-of-the-art heuristic method. The proposed method provides an efficient and systematic pathway for the practical implementation of advanced programmed PWM techniques in high-power converters. The framework is presented and experimentally validated for the challenging case of multidimensional selective harmonic control, which requires precise tracking of harmonic magnitude and phase references.