Adaptive AI-based Voltage Regulation in DC Microgrids Using Novel Power Optimization with Learning for Load Operations Algorithm

Abstract This paper presents a novel AI-based Adaptive Power Optimization with Learning for Load Operations (APOLLO) Algorithm for improving power quality in DC microgrid systems, which significantly outperforms conventional control methods. Our hybrid framework integrates Convolutional Neural Networks (CNNs) for feature extraction, Long Short-Term Memory (LSTM) networks for temporal pattern recognition, and Deep Reinforcement Learning (DRL) for adaptive control optimization. The proposed system demonstrates superior voltage regulation, achieving a 94.7% reduction in steady-state error and a 78.3% faster transient response compared to traditional PI controllers. Computational efficiency tests reveal an average execution time of 2.5 ms on standard embedded hardware platforms, making real-time implementation feasible for practical applications. The algorithm performs robustly under various disturbances, including load variations (± 50%), source fluctuations, and fault scenarios. Extensive validation was conducted through simulation studies using MATLAB/Simulink and hardware-in-the-loop testing on a laboratory-scale 380V DC microgrid prototype with distributed renewable energy sources, validating the approach's effectiveness across diverse operating conditions. This work addresses critical challenges in DC microgrid stability and demonstrates the potential of AI techniques to enhance power quality in next-generation distributed energy systems.