Survival Analysis Predicts Flow Regime Changes for Permian Basin Tight Oil Wells

Flow regime changes are important to capture and predict when estimating the future production from tight oil wells. Due to the nature of hydrofracture completions and reservoir heterogeneity, flow regimes are difficult to predict before they are observed. Also, while some wells experience interfractureinterference induced boundary flow quickly, others remain in linear transient flow for years. This creates right-censored data that can only be incorporated in data analysis workflows through applying survival analysis. This flavor of data analysis can identify trends in completion approaches and lead to insights about predicting and engineering flow regime changes. To assess the impact of completion technology on flow regime changes, we first quantified metrics for production data for 16,000 Midland tight oil well production data, including Estimated Ultimate Recovery (EUR), time to boundary-dominated flow (if applicable), and flow regime, using a physics-based scaling solution. Reservoir properties and well spacing estimates for these wells were extracted from a high-resolution 3D geomodel. We explored the changes over time of variables representing completion technologies. Then, we built a gradient-boosting regressor to predict accelerated failure time and used SHAP analysis to determine predictors of flow regime change. Of the 16,000 Midland basin wells investigated, 55% have not yet ended linear transient flow. Kaplan-Meier curves show that flow regimes have been switching earlier for more recently drilled wells. An XGBoost gradient boosting model provides reasonable but not exceptional predictability for flow regimes, achieving a concordance index of 0.7. The model is most influenced by fluid properties, then hydraulic fracturing fluid and proppant intensity, and lastly interwell spacing and lastly by rock properties such as carbonate content. Studies of the impact of completion versus spacing effects on well productivity have generally neither expressly focused on flow regime changes nor used survival analysis. Developing a robust understanding of the nature of this complex data set is an important prerequisite to any attempt to assign causal explanations to well productivity.