Hierarchical Clustering on Principal Components (HCPC) for Oil and Gas Reservoir Fluid Typing Based on Multifractal Attributes of Geophysical Well Logs

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Accurate identification of reservoir fluid type—oil or gas—is a critical aspect of petroleum exploration and production, as it directly influences the economic viability of field development, reservoir management strategies, and investment decision-making. Conventional well log interpretation techniques often encounter limitations in reliably distinguishing between oil and gas, especially in geologically complex environments where petrophysical signatures overlap. To overcome these challenges, this study investigates the use of fractal and multifractal attributes derived from geophysical well log data as advanced indicators for fluid classification. The primary objective is to assess how the nature of reservoir fluids influences multifractal behavior and to determine the discriminative power and relative importance of each multifractal attribute in fluid type differentiation. A nonlinear signal analysis framework based on Wavelet and Legendre transforms was used to extract a comprehensive suite of multifractal descriptors, including Box dimension and Regularization dimension, spectrum Width, Left and Right heights, αₚₑₐₖ, αₘᵢₙ, αₘₐₓ, f(αₚₑₐₖ), f(αₘᵢₙ), f(αₘₐₓ), and C-values. These attributes were analyzed through a hybrid machine learning workflow—Hierarchical Clustering on Principal Components (HCPC)—which integrates Principal Component Analysis (PCA) for dimensionality reduction, Agglomerative Hierarchical Clustering (AHC) for discovering structural patterns, and K-means for optimal classification. This methodology was applied to a dataset comprising 1085 geophysical well logs collected from gas- and oil-bearing formations within the sedimentary basins of southern Algeria. The classification process revealed three naturally distinct fluid-related classes. Gas zones (Class_1) are characterized by high values of f(αₘᵢₙ) and REG dimension, alongside low values of BOX dimension, Width, and Right height—indicating strong singularities and narrow multifractal spectra. In contrast, oil zones exhibit broader and more variable multifractal signatures. Class_2 oil is defined by elevated Right height, Width, and BOX dimension, reflecting a more complex and heterogeneous structure. Class_3 oil is distinguished by high C-values, as well as elevated Left height and αₚₑₐₖ, suggesting the presence of distinct multifractal patterns linked to specific fluid-rock interactions. The analysis of attribute importance further confirmed the discriminative roles of these parameters. Among the extracted attributes, f(αₘᵢₙ) and REG Dimension emerged as critical indicators for identifying gas zones, given their consistently high values in the gas class and lower values in the oil classes. Conversely, Right height, Width, and BOX Dimension proved essential in distinguishing between oil types, as they were elevated in both oil-related classes while capturing different multifractal features. Notably, C-values were consistently high in Class_3 oil, establishing them as a key marker for this specific oil subtype. Moreover, resistivity logs were identified as the most sensitive and reliable for fluid typing, due to their strong correlation with pore fluid conductivity and saturation—further reinforcing their interpretive value in reservoir characterization. This study demonstrates the potential of combining multifractal analysis with multivariate clustering techniques to improve the accuracy of reservoir fluid classification. The integrated approach enhances interpretive clarity, supports data-driven reservoir evaluation, and offers a scalable methodology adaptable to various geological settings. Future work will focus on incorporating broader geological and petrophysical variables and applying advanced machine learning algorithms to further refine classification performance and reservoir insight.

Article activity feed