| Abstract: |
Power electronic converters are the principal actuating and interfacing hardware of modern smart grids, coupling renewable generators, energy storage, electric vehicles, microgrids, and the bulk network through controlled dc-dc, dc-ac, and ac-dc stages. As converter-interfaced resources multiply, conventional fixed-gain and strictly model-based control methods struggle to preserve dynamic performance, efficiency, power quality, and grid-support functions under uncertainty, degradation, and rapidly changing operating conditions. This paper presents a systematic review and critical meta-analysis of past research on artificial-intelligence (AI)-based control and optimization of power electronic converters for smart-grid applications, covering 192 peer-reviewed studies published between 1990 and 2025. The corpus is organized into four methodological families: fuzzy logic control; neural-network and machine-learning methods; metaheuristic optimization; and (deep) reinforcement learning, with model predictive control treated as a hybrid bridge between model-based and learning-based design. Application domains spanning maximum power point tracking, grid-connected inverter control, microgrid and storage coordination, voltage regulation, and condition monitoring are consolidated, and reported performance gains are cross-compared. Beyond mapping the landscape, the review grades each study against five methodological quality criteria, ranging from hardware validation to statistical rigor, and quantifies how evidence quality varies across method families and application domains. Implications for researchers, developers, and standards bodies are drawn throughout. The critical meta-analysis exposes recurring weaknesses in past work: limited hardware validation, non-standardized benchmarks, weak reproducibility, insufficient statistical rigor, and inadequate safety guarantees. Prioritized research directions are identified to steer future efforts toward verifiable, certifiable, a |
