Two-Step Process to Quantify Effects of Geological and Completion Parameters on Unconventional Well Performance

Machine learning algorithms have been frequently used to evaluate the performance of unconventional wells. Typically, all available features (geological and completion) are used to correlate the data to production performance of wells. Unlike the traditional approach, we correlate production data first to geological and geophysical parameters (depth, porosity, permeability, brittleness, etc.). We then take the residual information after the first correlation and correlate the residual to completion parameters such as length, proppant amount and fracturing fluid. By separating geology from completion, we can isolate the effect of geology removed from completion. This is very useful for exploration prospects where very limited or no data are available from existing wells. Further, by correlating completion parameters separately to residual values, we can quantify the impact of each completion parameter which can help optimize completion parameters to get the most benefit of various completion strategies. We use SHAP (Shapley Additive Prediction) to isolate the effects of each completion parameter. Using large, confidential, data available, we evaluated multiple basins and applied the proposed methodology. These include: Eagleford, and Marcellus Basins. What we observed, in general, is that approximately 70 % of the variance in production data can be explained by geological parameters and about 20 % of the variance can be explained by completion parameters. True vertical depth consistently ranked as the most important geological feature in predicting performance, but other features varied in importance. If appropriate analog can be understood, the methodology from mature formations can be used in exploration blocks to identify sweet spots. The evaluation of completion parameters revealed that the relationship between performance and completion is not always linear and in some instances, incremental benefits of larger completion inputs are negligible and counter-productive. The length of the well is the most important completion variable in determining the performance. The impact of individual parameters varies depending on production parameter selected. Geology becomes increasingly important as more production data are added (12 month vs 24 month vs EUR) and completion parameters are less able to explain the performance as the length of the production data increases.