Multi-Layer LSTM Implementation in Operational Condition Forecasting of a Natural Gas Transmission Pipeline Network

Proper monitoring of natural gas transmission in a pipeline network is necessary to ensure balance in gas supply and demand. Physical correlations such as hydraulic and thermodynamic equations, although has been applied widely in many use cases to simulate gas flow throughout pipelines, have their own limitation in forecasting future trends. For that matter, data-driven technique is needed. In this paper, deep neural network with multi-layer Long Short-Term Memory (LSTM) cells is implemented to forecast future values of gas operational conditions in receiving facility (Sink) of a gas transmission network. Past flowrate data in all network end-nodes are used as input with additional historical pressure data in the sink itself. The model successfully learns from data and able to forecast pressure and flowrate condition in the sink at some time in the future with acceptable error.