TY - JOUR
T1 - Beyond conventional approach
T2 - hybrid supervised learning and feature selection algorithms for prediction sonic logs – a study in a tight gas sand, North of Oman
AU - Al-Handhali, Suad
AU - Al-Aamri, Mohammed
AU - Sundararajan, Narasimman
N1 - Publisher Copyright:
© 2023 Inderscience Enterprises Ltd.. All rights reserved.
PY - 2023/11/29
Y1 - 2023/11/29
N2 - The sonic log is an essential petrophysical log, and it is used in many petroleum applications. Since sonic logs are expensive to run in all boreholes, oil companies conduct them in a few wells. Thus, several workflows incorporate sonic log synthetisation using conventional empirical correlations. However, these traditional approaches are less reliable than modern-day machine learning techniques. This study combines machine learning and feature selection algorithms to predict synthetic sonic logs from basic petrophysical logs. The implemented machine learning algorithms are multi-linear regression (MLR), artificial neural network (ANN), support vector machine (SVM) and random forest (RF). This study was implemented with data from seven wells in North Oman’s tight gas sandstone. The models developed were built and evaluated. The results show that the hybrid random forest algorithm with a backward elimination feature selection approach was more robust and reliable for predicting sonic logs. [Received: November 22, 2022; Accepted: April 12, 2023].
AB - The sonic log is an essential petrophysical log, and it is used in many petroleum applications. Since sonic logs are expensive to run in all boreholes, oil companies conduct them in a few wells. Thus, several workflows incorporate sonic log synthetisation using conventional empirical correlations. However, these traditional approaches are less reliable than modern-day machine learning techniques. This study combines machine learning and feature selection algorithms to predict synthetic sonic logs from basic petrophysical logs. The implemented machine learning algorithms are multi-linear regression (MLR), artificial neural network (ANN), support vector machine (SVM) and random forest (RF). This study was implemented with data from seven wells in North Oman’s tight gas sandstone. The models developed were built and evaluated. The results show that the hybrid random forest algorithm with a backward elimination feature selection approach was more robust and reliable for predicting sonic logs. [Received: November 22, 2022; Accepted: April 12, 2023].
KW - artificial neural network
KW - backward elimination
KW - feature selection
KW - machine learning
KW - multi-linear regression
KW - petrophysics
KW - random forest
KW - sonic
KW - support vector machine
KW - tight sand
UR - http://www.scopus.com/inward/record.url?scp=85181677398&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85181677398&partnerID=8YFLogxK
U2 - 10.1504/IJOGCT.2023.135055
DO - 10.1504/IJOGCT.2023.135055
M3 - Article
AN - SCOPUS:85181677398
SN - 1753-3309
VL - 34
SP - 359
EP - 385
JO - International Journal of Oil, Gas and Coal Technology
JF - International Journal of Oil, Gas and Coal Technology
IS - 4
ER -