TY  - JOUR
A1  - Aulia, A.
A1  - Jeong, D.
A1  - Saaid, I.M.
A1  - Kania, D.
A1  - Shuker, M.T.
A1  - El-Khatib, N.A.
JF  - Journal of Petroleum Science and Engineering
UR  - https://www.scopus.com/inward/record.uri?eid=2-s2.0-85068832492&doi=10.1016%2fj.petrol.2019.106237&partnerID=40&md5=137e564cc6a2b72250529bfa5607fb2d
VL  - 181
Y1  - 2019///
PB  - Elsevier B.V.
SN  - 09204105
N1  - cited By 27
N2  - Sensitivity analysis is typically required to screen unwanted history matching parameters so that computational cost can be reduced. Random Forests is a well-known statistical learning tool that maps a list of input parameters onto a predicted response. A constructed Random Forests model ranks these input parameters efficiently even for highly nonlinear data. The aim of this study is to investigate the capability of Random Forests in assisted/automatic production history matching in reservoir engineering. A history matching case study has been chosen to rigorously compare the capability of Random Forests and one-parameter-at-time (OPAT) in sensitivity analysis. The capability of Random Forests as a sensitivity analysis tool was investigated and compared with OPAT. By implementing the optimal number of decision trees, which the Random Forests' configuration parameters, and the appropriate design of experiments method Fractional Factorial Design (FFD), the Random Forests-based history matching results have shown to be relatively better than OPAT-based results in this case. Here, Random Forests combined with FFD helped in improving the average error by 23.1, along with the reduction of the standard deviation of errors by 28.9 relative to OPAT. Combining Random Forests with FFD is relatively better than OPAT because of the rigorous permutation tests of each parameter. In addition, this combination requires only 16 samples, where OPAT requires 21 samples. The combination of Random Forests and FFD as a sensitivity analysis tool successfully selected the top parameters which are horizontal permeability-related. Such parameters are well known in their direct relationship with water production as described by the well-known fractional flow theory. To conclude, this study found there is a strong indication that Random Forests can be a more reliable sensitivity analysis tool in production history matching, given the right configuration and design of experiments strategy. © 2019 Elsevier B.V.
ID  - scholars11285
TI  - A Random Forests-based sensitivity analysis framework for assisted history matching
KW  - Computation theory; Decision trees; Design of experiments; Genetic algorithms; Oil well flooding; Petroleum reservoirs; Random errors; Reservoirs (water)
KW  -  Assisted history matching; Configuration parameters; Fractional factorial designs; Fractional flow theory; Global sensitivity analysis; Random forests; Reservoir engineering; Statistical learning
KW  -  Sensitivity analysis
KW  -  decision analysis; experimental design; genetic algorithm; historical perspective; hydrocarbon reservoir; permeability; reservoir flooding; sensitivity analysis
AV  - none
ER  -