Skip to main content Skip to footer content

Using deep learning for facies classification in geological exploration: feedforward neural network vs. convolutional neural network approaches

M. Noroozi and M.A. Riahi

Abstract: 

Facies classification plays a crucial role in geosciences, especially in the exploration and development of resources. Sedimentary facies offer valuable insights into physical, chemical, and biological conditions during sedimentation. Researchers have traditionally studied facies using rock samples, but machine learning offers a promising alternative for predictive modelling. This study employed two deep learning algorithms to classify facies using well-log data from the Hugoton and Panoma fields in North America, which derived from an academic exercise at the University of Kansas. The data set includes log data from nine wells, which was used to train supervised classifiers for predicting discrete facies groups. The first model, a Feedforward Neural Network (FFNN), achieved an accuracy of 72%, while the second model, a Convolutional Neural Network (CNN), demonstrated improved performance with an accuracy of 96%. These results underscore the effectiveness of deep learning for facies classification, with CNN outperforming FFNN in recognising complex geological patterns. Further improvements could be made through hyperparameter tuning and advanced architectures. Additionally, this study provides new insights into improving classification robustness by incorporating feature engineering and uncertainty estimation techniques.