TY - JOUR
T1 - Evolving strategies for shear wave velocity estimation: smart and ensemble modeling approach
AU - Olayiwola, Teslim
AU - Tariq, Zeeshan
AU - Abdulraheem, Abdulazeez
AU - Mahmoud, Mohamed
N1 - Generated from Scopus record by KAUST IRTS on 2023-09-20
PY - 2021/12/1
Y1 - 2021/12/1
N2 - Shear wave velocity has many applications in subsurface engineering such as reservoir engineering, rock mechanics, and seismic studies. To estimate the shear wave velocity in rocks, different methods such as laboratory experiments and well logging approaches are used. These approaches are usually involved with high cost and are time-consumption. To overcome these issues, several methods such as the use of empirical correlations and machine learning methods were proposed in the past. Most of these methods involve the usage of a large number of features and the formation of specific data points. In this study, a new method is proposed for feature selection by coupling regression algorithms to reduce the likelihood of the error introduced because of the redundant variables. A substantial number of conventional well logs and acoustic data from the open-source data of Norwegian Volve field were utilized. Four machine learning algorithms such as Random forest Regressor, Adaboosting regressor, Gradient Boosting tree, and Support Vector Regressor were implemented. The prediction results on testing dataset showed that the Adaptive boosting Regressor algorithm accurately predict the interrelationship between the selected features and the desired output by yielding MAE of 0.032 and R2 of 0.93. The previous methods resulted in MAE up to 2.0 and low as R2 of 0.5 between actual and predicted values.
AB - Shear wave velocity has many applications in subsurface engineering such as reservoir engineering, rock mechanics, and seismic studies. To estimate the shear wave velocity in rocks, different methods such as laboratory experiments and well logging approaches are used. These approaches are usually involved with high cost and are time-consumption. To overcome these issues, several methods such as the use of empirical correlations and machine learning methods were proposed in the past. Most of these methods involve the usage of a large number of features and the formation of specific data points. In this study, a new method is proposed for feature selection by coupling regression algorithms to reduce the likelihood of the error introduced because of the redundant variables. A substantial number of conventional well logs and acoustic data from the open-source data of Norwegian Volve field were utilized. Four machine learning algorithms such as Random forest Regressor, Adaboosting regressor, Gradient Boosting tree, and Support Vector Regressor were implemented. The prediction results on testing dataset showed that the Adaptive boosting Regressor algorithm accurately predict the interrelationship between the selected features and the desired output by yielding MAE of 0.032 and R2 of 0.93. The previous methods resulted in MAE up to 2.0 and low as R2 of 0.5 between actual and predicted values.
UR - https://link.springer.com/10.1007/s00521-021-06306-x
UR - http://www.scopus.com/inward/record.url?scp=85111277413&partnerID=8YFLogxK
U2 - 10.1007/s00521-021-06306-x
DO - 10.1007/s00521-021-06306-x
M3 - Article
SN - 1433-3058
VL - 33
SP - 17147
EP - 17159
JO - Neural Computing and Applications
JF - Neural Computing and Applications
IS - 24
ER -