Application of Stacking Model with Feature Weighting Method in Shale Gas Production Prediction

Shale gas is gaining attention as a non-traditional clean energy source. Predicting shale gas well production capacity is vital for production planning. Conventional methods struggle due to complex relationships between parameters. This study proposes a feature-weighted fusion and Stacking ensemble model. Using data from 242 gas wells, feature weights are calculated and applied to training/testing sets. Random Forest Regression, Support Vector Machine, and Light Gradient Boosting Machine models are used, with a Multilayer Perceptron model serving as the meta-learner. Results show the Stacking model significantly outperforms base learners in RMSE, MAE, and R-Square metrics, improving prediction accuracy and generalization. This research provides valuable insights for shale gas well production capacity prediction and production planning.