System Order Reduction for Gas and Energy Networks

No matter if natural gas, biogas or hydrogen, gas transport needs to be simulated ahead of dispatch to account for volatilities in demand and supply, so denominations are delivered reliably. The emancipation from producing countries alongside the renewable energy transition increases the number of scenarios to be simulated manifold, which in turn requires the acceleration of computational models to ensure completion of computer simulations before deadlines. Gas is transported through a network of pipelines which can be mathematically modeled as large‐scale nonlinear port‐Hamiltonian input‐output systems. To reduce computational complexity we propose unsupervised learning via synthetic data of the model's system‐theoretic properties which then enables data‐driven control or model reduction. We summarize the aspects of nonlinear model reduction techniques adapted to gas pipeline networks and orchestrated to reduce the order of this challenging class of systems originating from hyperbolic systems of partial differential‐algebraic equations, and demonstrate the applicability of our approach numerically.