The December issue of the peer-reviewed SPE Journal includes a spotlight section on data analytics. The section will showcase paper SPE 195698, “Prediction of Shale-Gas Production at Duvernay Formation Using Deep-Learning Algorithm.” The paper presents the limitations of using decline-curve analysis (DCA) to predict production from shale-gas wells and explores the use instead of an algorithm that has been widely used for prediction of time-series data.
The paper proposes using the long short-term-memory (LSTM) algorithm for production prediction in shale-gas wells. The authors point out five advantages that LSTM has over DCA in these instances. “First, it is much faster for predicting future production rates for each shale-gas well. Second, it is easy to use with a trained neural network without any inverse modeling or regression. Third, it can analyze the effect of additional features on future production rates, such as the shut-in period. Fourth, the same input sequences guarantee the same prediction result regardless of any subjectivity from an analyst. Fifth, the proposed method can yield a more reliable prediction.”
The authors say that LSTM can also be used with conventional wells but that it is more effective with unconventional wells because of their high density and the poor performance of DCA in those situations.
The SPE Journal is available on the OnePetro online library.