Machine-learning guided fracture density seismic inversion: A new approach in fractured basement characterisation

• Main Authors: Shamsuddin, A.A.S., Purnomo, E.W., Ghosh, D.P.
• Format: Conference or Workshop Item
• Published: Offshore Technology Conference 2020
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85097630998&partnerID=40&md5=5a47d07a1e7882ca42efbdfcd42319a6; http://eprints.utp.edu.my/24645/
• Institution: Universiti Teknologi Petronas

The main objective of this study is to map potential fracture density based on a new integrated study of a fractured basement area. A machine learning algorithm of well log fracture density - borehole image log (BHI) guided seismic inversion was performed. Structural seismic attributes which are including maximum and minimum curvature, coherence, ant tracking, neural network, and principal component analysis were used to validate this potential fracture density map. All fractures were interpreted after image from BHI was processed. These fractures were quantified as fracture density log (FDL) based on weight density fractures per meter. The inversion was performed based on the enhanced of Extended Elastic Impedance (EEI) with the FDL. It was done by defining the training data set relating the FDL and its associate EEI. The resolution of inversion is increased by conducting stochastic inversion to generate high frequency relative EEI. The absolute value EEI was obtained by adding the relative EEI with a low frequency model derived from the FDL. A study was conducted in the Field A of the Malay Basin, which is located in transtensional tectonic regimes with the domination of various rock types ranges from metasedimentary rocks to volcanic and igneous rock. The integrated study has been implemented to invert the potential fracture density of the fractured basement area. Based on the BHI analysis, the highest fracture density is defined as 1.3 fractures per meter. Inversion at the well location shows a high correlation and small error of the reconstructed fracture density (FDL). The analysis of cross section and the horizontal slice of inverted fracture density displays good fit with a fracture sensitive analysis, which is structural seismic attributes such as minimum and maximum curvature, coherence, ant tracking, principal component analysis, and neural network. This study provides a better fracture understanding as the resulted inverted fracture density is quantitative. It allows potential fracture density mapping by using geobody analysis through classification or cut off value. The novelty of the study is presented based on the potential to generate a 3D volume of well log calibrated fracture density by empowering the seismic elastic inversion with a sophisticated machine learning algorithm. Copyright 2020, Offshore Technology Conference