Development of Deep Learning-Based Detection Model for Borehole Breakout Depths in Image Logs
Year: 2024
Type: article
Abstract: This study applied the long short-term memory (LSTM) algorithm to detect the depths of borehole breakouts in image logs to overcome the limitations of the extreme gradient boosting (XGBoost) model.Using 99,084 data obtained from the preprocessed image log of a 1 km deep borehole in Gyeongju-si, we developed both XGBoost and LSTM models.While the XGBoost model considers only single-depth data, the LSTM model analyzes the occurrence of borehole breakouts at the central depth using data from seven consecutive depths.The LSTM model showed an 88% reduction in false negatives and 29% improvement in recall on the test data compared to the XGBoost model.It also outperformed the XGBoost model that using the seven-depth data.These results suggest that the LSTM model can systematically analyze sequential depth data to enhance the efficiency of experts in stress analysis.Key words : borehole breakout, image log, long short-term memory (LSTM), extreme gradient boosting (XGBoost), confusion matrix 요 약 본 연구는 이미지로그의 압축파쇄대 심도를 검출하기 위해 장단기기억(long short-term memory, LSTM)을 적용하여 XGBoost(extreme gradient boosting) 모델의 한계를 보완하고자 한다.경주시 심도 1 km 시추공의 이미지로그를 전처리하여 얻어진 99,084개 자료로 XGBoost와 LSTM 모델을 개발하였다.단일 심도만을 고려하는 XGBoost 모델과 달리, LSTM 모델은 연속된 7개 심도자료로 중앙심도의 압축파쇄대 여부를 분석한다.XGBoost 대비 LSTM 모델은 테스트자료에서 false negative가 88% 감소하였고 재현율이 29% 향상되었다.
Authors: Hayoung Kim, Juyeol Yeom, Chandong Chang, Yeonguk Jo, Kyungbook Lee
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4406076350