| Abstract [eng] |
The rapid expansion of renewable energy generation and energy-storage deployment has increased the demand for power conversion systems that combine high efficiency, reduced thermal stress, extended storage lifetime, reliable grid interaction, and adaptive real-time operation. Multilevel converter topologies, including neutral-point-clamped, active-neutral-point-clamped, flying-capacitor, cascaded H-bridge, T-type, and modular multilevel converters, provide improved waveform quality, lower semiconductor voltage stress, reduced filtering requirements, and enhanced scalability. Wide-bandgap devices, particularly gallium nitride and silicon carbide, further enable higher switching frequencies, lower losses, increased power density, and more compact converter designs. However, converter topology, semiconductor characteristics, storage dynamics, thermal behavior, and control implementation are strongly interdependent and therefore require integrated optimization. This review critically examines energy and exergy optimization in energy-storage-integrated multilevel converters from a converter-control co-design perspective. The literature is systematically classified according to converter topology, semiconductor technology, storage characteristics, control method, artificial-intelligence technique, validation approach, and reported performance metrics. An exergy-based framework is introduced to distinguish conventional energy losses from degradation of useful energy potential, including contributions from semiconductor losses, passive components, storage irreversibility, thermal gradients, circulating currents, capacitor-voltage imbalance, auxiliary consumption, and control-dependent switching activity. Conventional, nonlinear, predictive, and AI-assisted control strategies are compared in terms of efficiency, harmonic distortion, dynamic response, robustness, computational burden, thermal-stress mitigation, storage degradation, and real-time feasibility. The analysis shows that converter topology and control strategy should be optimized concurrently. Redundant switching states in active-neutral-point-clamped and modular multilevel converters can improve capacitor-voltage regulation and thermal distribution, but at the cost of greater sensing, computation, and implementation complexity. GaN devices support high-frequency and high-power-density operation but impose stricter requirements on control latency, parasitic-aware design, electromagnetic compatibility, gate driving, and thermal management. AI-assisted model predictive control and reinforcement learning offer adaptive and multi-objective capabilities, although their deployment remains constrained by limited training data, poor generalization, stability verification, computational demand, and embedded hardware limitations. The review identifies persistent gaps in standardized datasets, benchmark conditions, experimentally validated exergy assessment, and reproducible converter-control-storage co-design. Future research should emphasize exergy-aware predictive control, degradation-conscious dispatch, Pareto-based co-optimization, digital-twin-assisted monitoring, and standardized hardware-in-the-loop and experimental validation. |
