Composition-aware reduced-order modelling of real-time hydrogen blending in natural gas pipeline networks

Blending green hydrogen turns a single-composition gas network into one that must deliver a variable hydrogen fraction safely to every customer. A hydrogen-scheduled reduced-order model of the pressure–flow dynamics is coupled to a transport model, verified against a converged plug-flow reference, that carries injected hydrogen to every node with delays up to 81 h. For a frozen flow field the transport is linear, and prediction reduces to an impulse response without online states. On a real 22-node SciGRID_gas German sub-network the 44-state model matches a 275-state transient reference to 0.11% ( 𝑅 2 = 0 . 9 9 9 ) at 460 times its speed, and a steady-state solver to 0.21%. Against a mismatched plant on a time-varying flow field, a two-point scheduler holds every node below 20 vol% with a 2.0-volume-point back-off covering a 0.84-volume-point uncertainty budget, absorbing 36% by mass of available green hydrogen; a composition-blind baseline violates that limit an eighth of the time.