Single-inductor multiple-output converter for hydrogen electrolyzer arrays in PV-HESS-PEM microgrid using deep neural network based model predictive control

Abstract

This study investigates a microgrid including photovoltaics (PV), a hybrid energy storage system (HESS), and proton exchange membrane (PEM) hydrogen electrolyzer arrays. HESS mitigates transient power mismatches to regulate bus voltage, while PEM electrolyzers absorb excess PV power. The PEM electrolyzer arrays are integrated into the microgrid via a single-inductor multiple-output (SIMO) DC-DC converter, which reduces the number of inductors and switches compared to conventional solutions, demonstrating excellent scalability for more complex system topologies. To address the cross-regulation issue of the SIMO converter and meet the specific requirements of PEM arrays, a deep neural network-based model predictive control (DNN-MPC) method is proposed. This method overcomes the computational bottlenecks of classical MPC, by leveraging offline training, enabling real-time control. Simulations on a Matlab/Simulink platform verify that DNN-MPC effectively suppresses cross-regulation and drives PEM arrays under various operating scenarios. Comprehensive comparative analysis demonstrates that DNN-MPC achieves superior performance compared to classical MPC, with RMSE reductions of 92.3 %, 92.8 %, and 89.9 % across photovoltaic variation, load change, and system reconfiguration scenarios, respectively. Additionally, the proposed method reduces computational time by 7.0–22.5 % while maintaining excellent voltage tracking accuracy with overall RMSE values below 0.47 V across all tested conditions.