A Review: Deep Learning-Based Intelligent Control for DC–AC Power Converters: Loss Minimization and Power Quality Enhancement

Abstract

The rapid expansion of renewable energy and electric vehicles has increased the need for intelligent, high-performance DC–AC converters. Conventional control methods such as PI, PID, and MPC struggle under nonlinear and grid-interactive conditions, leading to higher THD and reduced stability. This paper reviews deep learning-based control strategies for DC–AC converters, emphasizing loss reduction, power quality enhancement, and real-time adaptability. Real neural networks can be  classified into specific architectures such as convolutional neural networks (CNNs,) Long Short-Term Memory (LSTMs), and Deep Reinforcement Learning (DRLs), and Physics-Informed Neural Networks (PINNs). are treated in detail. The results demonstrate that DL-based inverter controllers outperform classical control algorithms not only for THD attenuation and dynamic stability, but also for efficiency, under a digital control platform configuration, particularly when multi-point-injected energy grid-forming inverter topologies are considered. The paper concludes with directions for future research on real-time application feasibility, computational burden, and hybrid intelligent control systems. Finally, we present future research directions towards self-adaptive, sustainable, and cyber-resilient inverter control strategies for smart grids.