Abstract

Krishna-Godavari (KG) Basin is a poly-historic basin on the east coast of India that extends from onland to deep water. It is characterized by series of horst and graben systems and consists of more than 5km thick sedimentary sequence with an age range from Late Carboniferous to Recent with proven hydrocarbon repositories. KG Basin is a high-pressure high-temperature (HPHT) basin. The research area focuses on lithofacies identification, where temperatures reach up to 170 C and formation pressure exceeding 10,000 psi. Accurate lithofacies identification is crucial for reservoir characterization, which in turn influences the hydrocarbon exploration approaches. The presence of HPHT in the basin has severely affected the lithofacies, altering them diagenetically with the growth of authigenic clays, dissolution of minerals, and formation of cements. The presence of radioactive sands and over-pressure shale makes the lithofacies modelling complex and challenging. Traditional lithofacies identification techniques are challenged by complex, time-consuming, and substantially laborious work. To address these problems, we have employed supervised Machine Learning (ML) classification algorithms, utilizing multi-parametric well log data to accurately identify the lithofacies. Multiple supervised learning models were trained and evaluated using Precision, Recall, F1-score, and Accuracy metrics on labeled datasets. The ML models are optimized by tuning hyperparameters, and the final models are used for prediction on blind well data, achieving an accuracy of more than 83% and reaching up to 90%.