Design optimization and ML-based performance prediction of microgrid-hydrogen refueling systems using Gaussian process regression

This paper investigates the synergies of microgrids (MGs) and hydrogen refueling stations (H2RS) for reliable and cost-effective energy supply in arid regions. An optimization model is proposed to determine the optimal capacities of MG-H2RS components, considering real weather conditions and a net-energy metering strategy. The objective is to minimize lifecycle costs while meeting electrical and hydrogen demands. A Gaussian process regression model (GPRM) is developed to create a machine learning-based predictive method that captures the complex, non-linear relationships between input variables (weather conditions, energy consumption) and output variables (energy generation and hydrogen production). Results show that the optimal design reduces lifecycle costs by 15.8 %, cuts fossil fuel consumption by 62.2 %, and reduces carbon emissions by 65.3 %. The GPRM achieves R² values above 0.99 for key outputs, and the Q-Q regression plots and error distributions show minimal deviation in most performance outputs, confirming the model's stable convergence and strong predictive accuracy.