Economic benefit forecasting and parametric sensitivity analysis for shallow shale gas wells in the Zhaotong area using machine learning

Abstract

In shale gas development, Net Present Value (NPV) and Internal Rate of Return (IRR) are influenced by the coupling of multi-source geological and engineering parameters, and quantitative research on the marginal effects and risk thresholds of key parameters remains lacking. A deep feedforward neural network prediction model was constructed to achieve mapping from 19-dimensional features to NPV and IRR. The trained model was then utilized as a digital surrogate model to conduct univariate sensitivity analysis, quantifying the marginal impacts of parameters such as clay content, carbonate content, effective porosity, Poisson’s ratio, horizontal stress difference, and first-year average daily production on economic benefits. Cross-validated results indicate that the model achieves a mean R 2 of 0.6071 (±0.0957) for NPV prediction and 0.4168 (±0.1294) for IRR prediction. Furthermore, it identifies economic risk thresholds including 30% for clay content, 0.2 for Poisson’s ratio, and 17 MPa for horizontal stress difference, which are highly consistent with oilfield engineering experience. SHAP-based interaction analysis reveals that these thresholds are context-dependent, with interaction effects accounting for approximately 30%–32% of the main effects for clay content and horizontal stress difference.