Integrating hydrogen systems and electric vehicles into multi-energy microgrids creates complex operational challenges requiring accurate demand forecasting and robust decision-making under uncertainty. A new research framework tackles this by combining machine learning forecasting with decision theory for enhanced microgrid resilience.
The proposed system manages a 14-bus test network coupled with district heating and gas distribution infrastructure. It coordinates renewable generation, combined heat and power units, power-to-hydrogen conversion technology, and plug-in electric vehicles to meet three simultaneous demand streams. The innovation lies in its two-stage architecture: first, a hybrid CNN-LSTM neural network predicts electricity, heat, and gas demands with sub-15% forecast error across all carriers. These predictions then feed into an Information Gap Decision Theory optimization engine rather than using fixed nominal profiles.
IGDT enables operators to adopt different risk postures. Testing shows meaningful trade-offs: risk-averse scheduling increases operating costs 28% and electricity transaction costs 51% but delivers stronger protection against forecast errors and demand volatility. Conversely, risk-seeking strategies cut total costs and emissions by 40% and 85% respectively by maximizing exports during favorable conditions, though exposure to adverse scenarios increases.
The modeling combines Mixed-Integer Nonlinear Programming optimization with Python-based forecasting, demonstrating practical implementation feasibility. Results highlight how incorporating real forecast uncertainty into planning—rather than assuming perfect predictions—fundamentally changes optimal dispatch patterns and cost structures. This is particularly relevant for hydrogen microgrids, where electrolyzer flexibility and thermal storage provide multiple optimization levers.
The framework addresses industry needs for decision support under uncertainty, enabling microgrid operators and planners to explicitly choose between economic performance and operational safety based on their risk appetite and local conditions.
