A Method to Identify Lithofacies Based on Wavelet Transform, Principal Component Analysis and K-Means Clustering

Abstract

A method to extract lithologic interfaces and identify lithofacies based on the continuous wavelet transform (CWT), principal component analysis (PCA) and K-means clustering is proposed. Well-logs which can reflect lithofacies are selected by correlation analysis of multiple well-logs and their principal components are determined by PCA of them. The CWT of the 1st principal component (PC) based on the Gaussian wavelet at a fixed scale is used to detect temporary interfaces which include lithologic interfaces as well as those reflecting intra-bed variations. Interval signal is formed by averaging the 1st PC values between adjacent interfaces. Accurate and practical lithologic interfaces are reset by considering variances of the interval signal to select interfaces using the difference moduli of the interval signal. The K-means clustering in the main PC space is effectively employed to classify and identify sedimentary lithofacies from well log data. The application to well log data indicates that the method is useful and practical in detecting lithological interfaces and identifying lithofacies.