Facies Classification Based on Well Logs by Using an Convolutional Neural Network

Abstract

Facies classification, assigning a rock type or class to specific rock samples on the basis of measured properties, is fundamental in geologic investigations. Core observation is the most widely used method to identify the lithofacies accurately, however accessing to cores is time and cost limit. Thus, fast and accurate facies classification is a significant step in reservoir characterization and reservoir simulation. Conventional statistical and machine learning approaches such as principal component analysis (PCA), artificial neural network (BPNN), and support vector machine (SVM), have been used to identify lithofacies. However, the results of such a method by using imbalanced dataset may not result in higher accuracy in lithofacies prediction. In this research, we introduce a deep learning algorithm-convolutional neural network (CNN) to classify the complex lithofacies from Kansas and compare its performance to a few common machine learning algorithms. In order to train the CNN model, all of the feature vectors are transformed into geological feature images. CNN shows significantly high lithofacies classification accuracy compared to other algorithms, which indicate that CNN has a good performance on facies classification and strong generalization ability.

AAPG Datapages/Search and Discovery Article #90350 © 2019 AAPG Annual Convention and Exhibition, San Antonio, Texas, May 19-22, 2019