Data-Driven Forecasting of Well Production Considering Re-Fracturing in Shale Reservoirs

Abstract Re-fracturing is a widely adopted technique to enhance production in mature shale fields. The economic feasibility of re-fracturing should be carefully evaluated because re-fracturing is not always effective. However, it is hard to evaluate the economic feasibility of re-fracturing because re-fracturing records are often missing from public datasets. This study introduces a re-fracturing detection method that identifies re-fracturing events from production data using a DCA-based approach. The performance of the re-fracturing detection method is demonstrated by correctly identifying 96.0% of re-fractured wells in the Eagle Ford. This study also proposes a data-driven method that uses a long short-term memory (LSTM) model to forecast production while incorporating re-fracturing data. In the proposed LSTM model, incorporating re-fracturing data as input improves estimated ultimate recovery prediction in the Delaware, Eagle Ford, Midland, and SCOOP, with the coefficient of determination (R2) increasing from 0.595 to 0.852. The proposed LSTM model further outperforms random forest model, achieving an R2 of 0.852 compared to 0.804. The economic feasibility of re-fracturing can be efficiently and accurately evaluated by forecasting oil and gas production using the LSTM model combined with the re-fracturing detection method.