Predictive Modeling of Clay-Bound Water in Carbonate Reservoirs Using Machine Learning Techniques

Abstract The pore structure in carbonate rocks is intricate and heterogeneous, encompassing both intra-particle and inter-particle porosities. Ignoring the presence of clay-bound water during the assessment of hydrocarbon recovery in these reservoirs can lead to inaccurate recovery factor estimates. Conventional well logging techniques often struggle to accurately measure clay-bound water in such complex lithologies. Although Nuclear Magnetic Resonance (NMR) can measure microporosity independently of the rock matrix and mineralogy. However, NMR is very expensive to measure, and not widely available in conventional wells. In this study, we propose an approach utilizing supervised machine learning (ML) techniques to predict clay-bound water using readily available well logs. We have used a dataset comprising of 6000 samples collected from multiple wells within a carbonate reservoir to develop and validate ML models. Five different machine learning techniques were employed, including, Long Short-Term Memory (LSTM) networks, Recurrent Neural Networks (RNN), Gated Recurrent Units (GRU), extreme gradient boosting regressor (XGBoost), and Gradient Boosting Regressor (GBR). Model performance was evaluated using the coefficient of determination (R²) and mean absolute percentage error (MAPE). The results demonstrate that our ML tools can effectively predicts clay-bound water content using well log data, offering a significant time and cost-saving over traditional methods. Among the tested models, the LSTM network emerged as the top-performing algorithm, achieving an impressive R² value of 0.980 and a MAPE of 3%, indicating its superior ability to capture the complex relationships within the data. The GRU and RNN models also performed well, with R² values exceeding 0.95, while the XGB and GBR models provided moderate predictive accuracy. This study highlights the potential of machine learning techniques in enhancing reservoir characterization by providing a cost-effective and accurate alternative to traditional methods for estimating clay-bound water. By relying solely on well log data, our approach eliminates the need for expensive direct measurement methods, thereby offering a practical solution for reservoir engineers and geoscientists working in complex carbonate systems.