Prediction of coalbed methane fracturing effect based on GA-Elman neural network model

Coalbed methane (CBM) development relies on hydraulic fracturing, yet reliable prediction of fracture geometry remains challenging due to complex reservoir heterogeneity and strong nonlinear coupling among treatment parameters. This study develops a genetic algorithm (GA)-optimized Elman neural network to predict fracture length in horizontal wells of the South YC CBM field. The GA is employed to enhance global optimization capability and improve convergence efficiency, enabling robust mapping between operational parameters and fracture propagation behavior. Key controlling factors are identified via Pearson correlation analysis, including sand ratio, proppant volume, fluid volume, pre-fluid volume, extension pressure, and injection rate. The proposed model significantly outperforms the conventional Elman network, with the R-squared increasing from 0.74 to 0.97 and the mean absolute percentage error (MAPE) reduced to 2.19%. Blind validation on an independent well demonstrates strong generalization, yielding the root mean square error (RMSE) of 19.67 and the mean absolute error (MAE) of 17.74 m. Compared with backpropagation-based models, the GA-Elman approach exhibits superior accuracy and stability. The developed model provides a practical, data-driven tool for rapid fracture geometry prediction, supporting treatment design optimization and reducing uncertainty in CBM field development.