Application of a graph convolutional network for predicting the feasible operating regions of power to hydrogen facilities in integrated electricity and hydrogen-gas systems

The integrated operation of electricity systems and power to hydrogen (PtH) facilities with direct hydrogen injection into natural gas pipeline systems provides important flexibility to accommodate the intermittent outputs of renewable energy sources. However, no efficient approach has yet been developed to quantify the feasible operating region of PtH facilities to ensure the stable and safe operations of both electric power and natural gas systems. The present work addresses this issue by analyzing the feasible operating regions of PtH facilities under constraints considered for both electric power and natural gas systems. A graph convolutional network (GCN) is trained to predict the nonlinear flows of electricity, hydrogen, and natural gas under a wide range of circumstances, and these predictions are employed to generate numerous feasible operating points efficiently. The GCN approach collects PtH operational data associated with different natural gas network topologies . The convex hull approach is then employed to construct the feasible operating region representing the smallest convex set that contains all operating points. The proposed models and methods are validated based on the numerical results obtained for an integrated IEEE 118-node power system and a 25-node natural gas system. • We propose a convex-hull approach to construct operating region of electrolyzers. • Graph convolutional network is employed to efficiently identify operating points. • Our approach outperforms traditional ones in terms of solution accuracy. • Our model enables high-dimensional visualization for flexible electrolyzer operation.