Prediction\nof Oil and Gas Well Integrity Using Well\nConstruction Physical Parameters and Geospatial Metrics

Oil and gas (O&G) well integrity\nissues, including\nsustained\ncasing pressure and casing vent flow, could cause significant environmental\nand climatic risks by releasing methane-containing fluids into the\nenvironment. The large number of O&G wells makes routine inspection\nor spatiotemporally adequate screening of well integrity issues expensive\nand sometimes intractable. Machine learning (ML) algorithms with the\nability to predict O&G well integrity issues could help prioritize\nsuch inspections. We manually extracted well characteristics from\n1223 O&G well completion reports in Bradford County, PA, appended\ngeospatial metrics, and used them to predict well integrity issues\nreported by field tests. Different ML models (e.g., Random Forrest,\nXGBoost, and Logistic Regression) were compared, and an accuracy of\n68% was achieved by comparing the predictions with unseen field test\nrecords. Important predictive features were identified by the Random\nForrest model, such as the length of casings, amount of cement, and\nwell operator. Moreover, the wells with integrity issues were geospatially\nclustered in our study region, which could be explained by the clusters\nof important physical features. Overall, the predictions would help\nprioritize well integrity inspection and have the potential to guide\nwell design and well location selection for mitigating integrity issues\nthat lead to methane emissions.