A novel approach for identifying sweet spots in tight reservoir fracturing engineering based on physical-data dual drive

Abstract

Reservoir engineering sweet spot identification is a crucial prerequisite for fracture interval selection and hydraulic fracturing design. Rock mechanical parameters serve as key indicators for evaluating engineering sweet spots. To accurately predict the rock mechanical parameters of tight reservoirs, a physics-informed NSGA-PINN (Non-dominated Sorting Genetic Algorithm combined with Physics-Informed Neural Networks) model was developed, achieving prediction accuracies exceeding 90 % for four rock mechanical parameters, outperforming purely data-driven models such as RF (Random Forest), CatBoost, LightGBM, and BPNN (Back Propagation Neural Network). On this basis, an intelligent evaluation method for engineering sweet spots was established by integrating mechanical parameters and brittleness index, and a fracturing sweet spot calculation model was constructed using a combined weighting approach. The results indicate that the physics-informed neural network model exhibits superior generalization and robustness, and the calculated sweet spot index shows a 91.2 % correlation with post-fracturing gas well productivity, demonstrating the reliability of the proposed method. This approach can be effectively applied to the efficient development of gas reservoirs in the target block.