Adaptive Neural Network-Based PI (ANN-PI) Control for DC Microgrids in Renewable Hydrogen Production Systems

Abstract

This research presents a neural-adaptive control technique for DC microgrids in renewable hydrogen production systems. The proposed approach tackles voltage stability issues arising from variable solar production and fluctuating electrolyzer loads with an adaptive neural network-based proportional-integral (ANN-PI) controller with online system identification. The control architecture utilizes two concurrent multilayer perceptron (MLP) networks: one for real-time system identification to estimate the Jacobian matrix, and another for adaptive proportional-integral (PI) parameter adjustment. The decentralized architecture removes communication dependencies among converters, hence improving reliability and scalability. Simulation results indicate a better dynamic response with a 50% decrease in settling time, increased voltage stability retaining the DC bus voltage within ±2% of the nominal 400 V, and resilient performance under diverse situations, including load transitions and changes in solar irradiation. The neural-adaptive method effectively facilitates intelligent, model-free regulation for electric-hydrogen DC microgrids.