Direct Hydrocarbon Indicators Mapping via Joint Cluster Analysis: A Two-Step Approach over 3D Seismic Data

Abstract

This paper presents a novel methodology developed in Python to map direct hydrocarbon indicators anomalies in 3D seismic data using the unsupervised machine learning algorithms K-Means and Gaussian Mixture Models. The joint cluster analysis consists of implementing the spatial density-based filtering after clustering analysis and investigates the groups interpreted as DHI aiming to distinguish sparsely dense samples and noisy information from samples that are, in fact, areas of interest for hydrocarbon exploration. The experiments were performed on the 3D seismic data F3 Block from Central Graben Basin, Dutch North Sea. To conduct the experiments, the following seismic attributes were extracted: Spectral Decomposition of 25 and 45 Hz, Relative Acoustic Impedance, Coherence, Logarithm of Sweetness, and Reflection Strength. The working flowchart took advantage of good artificial intelligence practices to train the models, such as seismic attributes preconditioning, dimensionality reduction via Principal Component Analysis (PCA), and model validation through statistical tests. Despite the initial challenges faced in isolating DHI anomalies through the K-Means algorithm, the two-step approach ultimately succeeded in accurately mapping them