The 13th SEGJ International Symposium, Tokyo, Japan, 12-14 November 2018
Abstract
Machine learning is getting more popularity nowadays since it has potential to improve the quality of our life in many fields through data learning and optimization. In a short explanation, machine learning can be seen as non-linear equation optimization that try to predict one output from several input parameters although it may not have any logical relationships. In the world of geoscience work, one challenging problem is lithology estimation from several logs input and/or seismic data. Until today, predicting lithology from seismic data is common and popularly known as seismic inversion or color inversion, where seismic data has been converted from impedance to lithology to help interpreter pick horizons. This study is focusing on an effort to estimate facies or lithology by only looking some well logs data input using a method named K-Nearest Neighbor method as one of many machine learning methods. I have tried to pick a unique K value with best errors for predicting facies classification. The value has been tested at training data with satisfactory results that is better than the results from tradidtional KKN. In the summary, although in this study KNN method successfully estimated the lithology, reasonable and practical results should be considered with several things such as regional geological model, a continuous improved model with newer data once acquired, coring data validation, and etc. Future work that try to expand the capability between seismic data and lithology facies would be interesting to pursue.