Dynamic Prediction of Natural Gas Calorific Value Based on Deep Learning

The natural gas quality fluctuates in complex pipeline networks, because of the influence transmission process, changes source, and fluctuations customer demand mixing process. Based on dynamic characteristics system with large time lag non−linearity, this article establishes a deep−learning−based prediction model for calorific value which is used to accurately efficiently analyze networks caused by non−stationary processes. Numerical experiment results show that deep−learning can effectively extract effects hydraulic distribution. method able rapidly predict network, based real−time operational data such as pressure, flow rate, parameters. It has accuracy over 99% calculation only 1% physical simulation (built solved TGNET commercial software). Moreover, noise missing key parameters samples, still maintain an rate 97%, provide new assignment values on−site metering management.