Research on intelligent fracturing parameter optimization method based on deep learning and construction curve feature extraction

Introduction Optimizing fracturing parameters under multi-factor, complex conditions remains challenging in low-permeability reservoirs. Methods We extract stage-aware construction-curve features, compute composite correlations (Pearson, Kendall’s Tau, Random Forest), train an SSA-BP surrogate to predict open flow capacity (OFC), and apply a GA to optimize fluid volume, pump rate, and proppant concentration. Results Twenty key factors were retained. Among four regressors, SSA-BP performed best (highest R ² , lowest MSE). GA-optimized parameters improved OFC in multiple wells; a field application (Well A-X) showed increased daily gas and OFC after adjustments. Discussion The integrated feature-extraction + SSA-BP + GA workflow provides accurate OFC prediction and practical parameter optimization. Limitations include single-field data (∼70 wells) and SSA-BP computational cost; future work will expand datasets and explore lighter models.