INNOVATIVE APPROACH TO DIAGNOSTICS OF WELL INTEGRITY USING MA-CHINE LEARNING

Abstract The paper considers a method for determination of production casing string integrity using machine learning algorithms. Under conditions of late stage of field development and high water cut of produced fluid, timely diagnostics of leaky production casing becomes an urgent problem. Traditional methods based on well logging often fail to promptly detect casing leaks, which leads to economic losses and environmental risks. The authors propose a new method based on water chemistry analysis, well age, number of well repairs, well production performance and well design. These features are processed and analyzed using machine learning model to obtain a real-time decision on leak presence. The paper presents the results of application of different machine learning algorithms such as decision tree, random forest, CatBoost and LightGBM. The best results have been obtained using LightGBM, with F1-score metric of 0.764. The paper also analyzes the impact of different features on model predictions using SHAP analysis. Finally, the advantages of the proposed workflow are discussed, such as reduction of decision-making time and improved efficiency of production casing leak detection, which in turn improve the economic performance and reduce environmental risks.