TY  - JOUR
KW  - Backpropagation; Feedforward neural networks; Forecasting; Organic carbon; Petroleum reservoirs; Petrophysics; Statistical tests; Tight gas; Well testing
KW  -  British Columbia 
KW  -  Canada; Feedforward neural networks (FFNN); Geochemical datasets; Mathematical relationship; Petrophysical properties; Regularization technique; Tight gas reservoirs; Total Organic Carbon
KW  -  Oil well logging
ID  - scholars14900
N2  - Total Organic Carbon (TOC) and maturity level (Tmax) for any source rock considered to be the key parameters for evaluating its potentiality. The TOC and Tmax are estimated mainly by analyzing core samples or cuttings using the common nonfilter acidification combustion and pyrolysis, both methods are time-consuming and costly. Therefore, in recent years, the search for fast, cheap, and appropriate methods has been the key focus of the literature. This study focuses on the application of artificial neural networks (ANNs) and principal component analysis (PCA) to develop an accurate model for TOC and Tmax prediction, based on petrophysical logs of depth, sonic, natural gamma, deep resistivity, and density. However, relying on petrophysical properties alone is complex, involves mathematical relationships that are difficult to use as well as dealing with enormous data that may not be applicable on a large scale, hence the need to use ANN has emerged. The data were collected from the British Columbia Oil Gas Commission for 16 wells representing Montney Formation tight gas reservoir located in the northeastern part of British Columbia Canada. The use of ANN coupled with PCA in this study has proven to be reliable and effective for the prediction of TOC and Tmax. The results of feedforward neural network (FFNN) with back-propagation algorithm and the BR regularization technique for 9 input parameters produced satisfactory performances in TOC prediction (R2 = 94 and 89) in training and validation phases, respectively, and for the Tmax prediction (R2 = 88 and 86) with 5 input parameters in the training and validation phases, respectively. The testing and validation of the ANN models were carried out on 2 wells that were randomly selected with the same log dataset, but not involved in the main training and testing processes. The results of the testing were very high, R2 = 88.71 and 85.39 for TOC and Tmax respectively in well one, and R2 = 85.19 and 84.28 for TOC and Tmax in well two. The findings of this study show that the use of ANN is feasible and can be applied to applications involving the evaluation of organic richness and maturity of a target source rock using conventional logs especially when there are incomplete or missing geochemical datasets. © 2021 THE AUTHORS
IS  - 3
Y1  - 2021///
VL  - 60
UR  - https://www.scopus.com/inward/record.uri?eid=2-s2.0-85100671361&doi=10.1016%2fj.aej.2021.01.036&partnerID=40&md5=eb026a772744855e39a196f13585b4da
A1  - Barham, A.
A1  - Ismail, M.S.
A1  - Hermana, M.
A1  - Padmanabhan, E.
A1  - Baashar, Y.
A1  - Sabir, O.
JF  - Alexandria Engineering Journal
AV  - none
TI  - Predicting the maturity and organic richness using artificial neural networks (ANNs): A case study of Montney Formation, NE British Columbia, Canada
SP  - 3253
N1  - cited By 5
SN  - 11100168
PB  - Elsevier B.V.
EP  - 3264
ER  -