Learning‐Enhanced Model‐Based Control for Power Electronic Converter Systems: A Review

ABSTRACT

Model‐based control remains the engineering foundation for power electronic converter control because it keeps converter models, control objectives, constraints and modulation interfaces explicit. Learning methods are increasingly introduced to address limits of hand‐designed controllers under uncertain parameters, nonlinear operation, high‐dimensional switching decisions, modular coordination and tight real‐time budgets. Reviews organised by learning algorithms, however, do not show where a learnt component enters the control architecture, how directly its output is exposed to the switching stage or what evidence is required before deployment. This review therefore examines learning‐enhanced model‐based converter control from an architectural viewpoint. Starting from a basic converter‐control formulation, it organises the field into five bounded insertion roles: parameter tuning, model/surrogate learning, control‐law learning, safety guarding and supervisory control. Each role is defined by its input/output object, control‐layer exposure and retained model‐based or physical anchor. The roles are then traced across DC–DC converter regulation under constant‐power loads, grid‐connected voltage‐source converters, predictive and multilevel converters and dual‐active‐bridge, triple‐active‐bridge and solid‐state‐transformer modular systems as an application evidence base rather than as a topology taxonomy. This role‐based reading links each insertion point to the evidence it must report, including stability and feasibility, safety and protection‐envelope coverage, embedded timing, operating‐envelope generalisation, reproducibility and reliability‐aware operation. The review finds that credible learning‐enhanced converter control is strongest when learning remains bounded by explicit models, modulation laws, constraints or protection logic and when these evidence dimensions are evaluated together rather than reported in isolation.