Information Theory Based Probabilistic Machine Learning And Wireline Inversion: Surat Basin Case Study

• Published in: ASEG Extended Abstracts Vol. 2019; no. 1; pp. 1 - 4
• Main Authors: Gaede, Oliver, Wellmann, Florian, Flottmann, Thomas, Regenauer-Lieb, Klaus
• Format: Journal Article
• Language: English
• Published: Taylor & Francis 01.12.2019
• Subjects: Coal Seam Gas; Information Entropy; Machine Learning; Probabilistic Method; Wireline Data
• ISSN: 2202-0586
• DOI: 10.1080/22020586.2019.12073234

We present a probabilistic machine learning approach to determine lithologies for the wireline data from the Springbok Sandstone formation in the Surat Basin. Deterministic inversions of the data are compared to the new machine learning approach in order to develop a generic method for wireline log inversions. The approach is designed to combine the benefits of classical deterministic inversion with new machine learning algorithms hence providing an approach to a statistical learning method. In this approach model parameters have a physical meaning and therefore can be readily related to petrophysical standards. We start by establishing a prior distribution of minerals within the Springbok Sandstone Formation using geochemical laboratory data. The prior distributions are used in a Bayesian framework to improve wireline-based predictions of the mineral assemblages within the formation. The uncertainties of the predictions are quantified using the associated information entropy measure. This approach allows us to choose between various model mineral assemblages and apply an information theory augmented statistical learning approach.