Risk-based data-driven energy management for integrated electrical and hydrogen microgrids with improved hydrogen vehicle charging prediction

The increasing integration of renewable energy sources (RESs) and hydrogen-powered vehicles (HVs) into integrated power and hydrogen microgrids (IPHMs) poses significant operational challenges due to uncertainties in RES generation and dynamic HV fueling demands. Current methods, such as gated recurrent unit (GRU) networks for predicting HV fueling demands, often fail to effectively prioritize and combine the full range of influencing factors. Moreover, standard approaches to RES output uncertainty typically use static, predefined bounds for uncertainty sets, which can introduce subjectivity, reduce adaptability, and lead to suboptimal energy management solutions. This paper addresses these deficiencies by proposing a novel risk-based, data-driven robust energy management framework for IPHMs. The primary goals are to enhance HV fueling prediction accuracy and to optimize IPHM operation under uncertainty. First, this paper develops a multi-head attention-based GRU (MHA-GRU) network, further enhanced with copula functions (MHA-GRU-Copula), to more accurately predict HV fueling demands by embedding a comprehensive suite of features including starting location, destination, hydrogen station selection, transportation system structure, and the correlation between travel time and hydrogen consumption. Second, a risk-based data-driven robust energy management model is formulated to dynamically optimize the bounds of RES uncertainty sets, achieving a better trade-off between robust operation costs and potential risk costs. Case studies on a realistic multiple-IPHM system demonstrate that the MHA-GRU-Copula network achieves significantly improved prediction accuracy, reducing mean absolute error by 18.6 % and mean squared error by 14.4 % compared to standard GRU models. Furthermore, the proposed risk-based optimization approach lowers total operational costs by 7.2 % and risk costs by 24.5 %, outperforming conventional methods with fixed uncertainty bounds. An optimal trade-off was found at an uncertainty set bound of 56 %. The proposed framework ensures more economic and reliable operation of IPHMs by effectively addressing inherent uncertainties in both transportation and energy systems, offering significant applications for the planning and management of advanced, integrated energy infrastructures.