Data-driven computation of natural gas pipeline network hydraulics

Hydraulic calculation is a major path to the management of natural gas pipeline network. However, the assumed physical parameters such as the length of the pipes are usually investigated from blueprint or factory technical tests, which may differ from actual values and cause inaccurate hydraulic calculation result. Moreover, an initialization sufficiently near the root guarantees the convergence of the Newton's Iteration (NI) for solving the hydraulic equations. In this paper, a data-driven hydraulic calculation method is proposed, where we make two major contributions. (i) A parameter estimation method is provided for adjustment to the hyper parameters of the hydraulic calculation. (ii) An initialization method using neural network is proposed to make an estimate for the root at a first glance to ensure the convergence. The numerical results show that the proposed data-driven parameter estimation improves the hydraulic calculation accuracy and initialization method reduces the calculation time cost.