Abstract

Accurate estimation of effective porosity (Φe) is crucial for reliable reservoir characterisation and formation evaluation, which directly influences hydrocarbon recovery strategies based on the estimated volume of prospect. This study investigates reducing the uncertainty of Φe using deep learning and core data over the conventional practice in the E&P industry. The Upper Devonian Tahara Formation, a major reservoir in the Ghadames Basin, Libya, plays an important role in hydrocarbon exploration due to its significant potential. To enhance the Φe estimation in this formation, a combination of core data and basic well log inputs was utilised, along with advanced deep learning (DL) algorithms namely Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Convolutional-LSTM (Conv-LSTM). These DL techniques were trained using data from Well-A to improve the prediction of Φe, capturing the complex relationships between well logs and Φe more effectively than the traditional industrial methods. The best-performing models were then applied to a blind test well (Well-B), for Φe prediction. Results indicate that the Conv-LSTM model significantly outperforms, achieving a root mean square error (RMSE) of just 2.23%. This data-driven approach successfully captures the complex, nonlinear relationships between well logs and Φe, minimizing interpreter bias and reducing the dependency on extensive core sampling. The study highlights the potential of integrating DL techniques into routine petrophysical workflows, offering scalable and accurate alternatives for reservoir property estimation, especially in data-constrained environments.