USING MACHINE LEARNING FOR WELL PERFORMANCE ANALYSIS & FORECASTING
Managing operating conditions and well maintenance programs are efforts of operators that require wells to be shut-in to optimize equipment for well productivity. As such, there is great interest in both forecasting the behavior of a well after it returns to production, and in understanding the operational and geologic factors that determine that behavior.
In this study, datasets were constructed by taking publicly available production data for the Midland and Williston basins and extracting spans of well shut-in events. These data were used as features in predicting the percent change in production for the first six months following return to production. Shapley values were used to draw inferences about the drivers of well behavior after returning to a producing status.