Data-Driven Analysis and Template Construction for Casing Deformation in Deep Shale Gas Basin

Summary In this study, we develop a novel and systematic data-driven approach for analyzing the key drivers of casing deformation and construct a risk assessment template for fracturing design optimization. Casing deformation occurs frequently in deep shale gas fields in the Sichuan Basin of China. It impairs well integrity and causes economic loss due to production deprivation. Traditional methods for predicting casing deformation that build upon theoretical derivation and numerical simulation involve idealized assumptions and simplified or uncertain subsurface modeling. A data-driven approach uses actual field data to build a machine learning model that maps the relationship between geological parameters, fracturing designs, and casing deformation risk in an end-to-end fashion. This way, the true mechanisms can be captured more comprehensively and rigorously, while the governing equations do not need to be explicitly derived and solved. We use extreme gradient boosting (XGBoost) to build a collection of decision tree models that capture complex nonlinear relationships with a relatively small data set and interpret the model using SHapley Additive exPlanations (SHAP) analysis to quantitatively determine the contribution of every geological and engineering parameter to casing deformation. SHAP analysis provides the weights of important parameters and the technical thresholds that cause notable changes in casing deformation probability. We use these weights and thresholds to build a 2D template that quickly predicts the probability of casing deformation from normalized and unified geological and engineering factors and establish a categorization criterion for casing deformation risk assessment. Field experts without data science background can conveniently apply the template to evaluate the integrity of new wells and adjust the development plan accordingly to reduce casing deformation risk.