The right placement of fractures helps to enhance gas production in shale gas reservoirs. One parameter that helps to determine the target layers to place hydraulic fractures is the Brittlenex index (BI). However, no universal and appropriate methods can be used to compute BI, with all established correlations being used under different conditions. This paper uses machine learning (ML) methods to predict the BI of Upper Ordovician Wufeng to Lower Silurian Longmaxi formation in the Weiyuan shale gas field, Sichuan Basin, China. Random forest based on particle swarm optimization (PSO-RF) was utilized for the first time to predict BI due to its ability to capture nonlinear relationships between many variables in the dataset, thus giving more accurate results than other models. Collected secondary data from the WY1 well were used for training, whereas WY2 well data were used for testing. The results revealed that PSO-RF outperformed Extreme gradient boosting (XGBoost), Light gradient boosting machine (LightGBM), and K-nearest neighbor (KNN) in predicting BI with high accuracy and minimum errors during training and testing. PSO-RF coefficient of determination (R2), root mean square error (RMSE), and mean absolute errors (MAE) after training and testing were 0.9934 and 0.9533,4.6327 and 15.5308,2.0974 and 5.3896, respectively. In addition, the best-developed PSO-RF model was used to predict BI in WY3 and WY4 wells for model results validation; it was found that the model predicted the BI with high accuracy. This confirms that the developed model can be used to predict the BI of new development wells without depending on laboratory measurements, which are expensive and time-consuming to compute; thus, the developed model can be adopted as an alternative technique to determine the sweet spot for hydraulic fracturing in shale gas reservoirs to enhance gas production.
Read full abstract