Prediction of Reservoir Permeability and Porosity using a Deep Learning Technique

Summary Reservoir characterization is essential for optimizing production in the oil and gas industry. This study aims to develop a deep learning-based predictive model to estimate permeability and porosity in gas reservoirs using historical production data. Traditional methods of determining reservoir properties rely on core sampling and well testing, which are costly and limited in scope. In contrast, advanced machine learning techniques, including artificial neural networks (ANNs) and convolutional neural networks (CNNs), offer powerful alternatives for analyzing complex, large-scale datasets. The study utilizes simulated data from a 45 km by 47 km reservoir model with 248 wells and 48 permeability realizations. Production data are preprocessed, normalized, and divided into training and validation sets. Deep learning models, including CNNs and hybrid LSTM-CNN architectures, are trained to predict permeability and porosity maps based on well production responses. The results demonstrate the model’s effectiveness in predicting key reservoir properties, reducing reliance on traditional, costly methods. Future work will enhance model accuracy by incorporating additional data types and exploring real-time updates. This approach not only saves time and costs but also provides detailed property maps for better reservoir management and optimization.