Petrophysical Regression regarding Porosity, Permeability, and Water Saturation Driven by Logging-Based Ensemble and Transfer Learnings: A Case Study of Sandy-Mud Reservoirs
Table 6
Measured information of porosity, permeability, and water saturation gained by four validated predictors in three experiments.
Model 1
RMSE results for porosity (%) evaluated by Equation (7) 2
1st experiment
2nd experiment
3rd experiment nontrans/trans 3
KNN-cored predictor
1.2048
0.9882
0.8234/0.5543
SVR-cored predictor
1.1751
0.9215
0.7436/0.5168
RF-cored predictor
1.1024
0.8502
0.6466/0.4599
LightGBM-cored predictor
0.8947
0.7091
0.6096/0.3934
RMSE results for permeability (mD) evaluated by Equation (8)
1st experiment
2nd experiment
3rd experiment nontrans/trans
KNN-cored predictor
0.4027
0.3466
0.1375/0.1015
SVR-cored predictor
0.3956
0.3440
0.1700/0.1189
RF-cored predictor
0.3660
0.3201
0.1499/0.1070
LightGBM-cored predictor
0.3024
0.2510
0.1115/0.0761
RMSE results for water saturation (%) evaluated by Equation (7)
1st experiment
2nd experiment
3rd experiment nontrans/trans
KNN-cored predictor
4.5537
3.8576
6.5779/4.8992
SVR-cored predictor
4.3016
3.7810
6.1735/4.6108
RF-cored predictor
3.9348
3.4938
5.7748/4.3459
LightGBM-cored predictor
3.6630
2.8941
4.8863/3.2708
1KNN = k-nearest neighbors; SVR = supper vector regression; FR = random forest; LightGBM = light gradient boosting machine; 2RMSE = root-mean-square error; 3“nontrans” means normal prediction, and “trans” stands for the prediction implemented under the support of transfer learning.