A Self-Organizing Map, Machine Learning Approach to Lithofacies Classification

Lan Mai-Cao (Ho Chi Minh City University of Technology (HCMUT), Vietnam); Chau Le (Ho Chi Minh University of Technology & Vietnam, Vietnam)

Nowadays, there are two main problems in data analysis, especially in lithofacies classification which are the big data and the fact that human cannot fully understand relationship between seismic attribute.[1] With our machine learning approach, we can not only solve these two problems but also reduce its time-consuming aspect and give an accuracy result even with non-experiences user. Typically, an exploration well is required to build a facies. However, only well log data is given and cores are not sampled. Given these circumstances and the conventional method like regression is unsolvable, our approach is taken by the use of 3 methods: (1) Using Principal Component Analysis (PCA) to select the most meaningful attributes, (2) grouping depth intervalswhich have similar facies into clusters by training Self-Organizing Map (SOM) and (3) Clustering to separate different facies into individual zones. Our case study focus on 2 well. The first one is Well Stuart, Brendon Hall, Enthought. 2016[2]. The second one is Well 1-X located in Oilfield Y, Vietnam. Our model is mathematically done by programming using Python language and then compared to Interactive Petrophysics(IP) software.

Journal: International Journal of Simulation: Systems, Science and Technology IJSSST V19

Published: Jun 30, 2018

DOI: 10.5013/IJSSST.a.19.03.16