Abstract:
Classification of different lithofacies is crucial in seismic interpretation because different rocks have different permeability and fluid saturation for a given porosity. The ideal sources for lithofacies classification are core samples of rocks extracted from wells. Nevertheless, core samples cannot always be obtained due to associated costs. The conventional classification method is based on manually assigning lithofacies by human interpreters and is a very tedious and time consuming process.
I aimed at automating this classification process through the use of machine learning and deep learning methods. I selected wells from the region which have the mud log present with them. Machine learning algorithm, Support vector machines (SVM) was employed to build an automatic lithofacies classifier. The algorithm was trained on the dataset that has lithofacies marked from the mud log. The accuracy of the model was validated on a well with unlabeled lithofacies and accuracy of 0.63 was achieved.
In the next phase, a Deep learning (DL) model based on Convolutional Neural Network was developed and training was carried out using the labeled dataset (same dataset as in case of SVM algorithm). It achieved an classification accuracy of 0.71 on blind dataset. The model performed fairly good at classifying facies which have a larger number of training examples. Due to the skewed nature of the dataset, the validation accuracy of the model showed a stark drop when compared with training accuracy. This major drop in accuracy occurs while classifying those facies which have limited number of training example. The accuracy could be further improved by incorporating adjacent lithofacies in classification task, which was the limitation of the target dataset
