Automated control on natural gas pipelines using deep learning algorithms

Natural gas has been recognized as a promising resource for urban energy system due to the less air pollution and a reliable supply transported along pipelines. A simplified physical model is established in this paper, which simulates the direct correlation of compressor operations and the inlet flux at each station to be the theoretical foundation of pipeline control. The deep neural network is designed then, with the corresponding input features and output results connected via certain hidden layers representing the complex hydraulic processes occurred along the pipeline. After training and network tuning using practical operation data, the trained model is shown to be effective in predicting the inlet flux at certain station as a consequence of certain operations on the compressors, which is essentially needed for the controllers in an urban energy controlling center.