Date of Graduation
2002
Document Type
Thesis
Degree Type
MS
Committee Chair
Khashayar Aminian
Abstract
Although many attempts have been made in the recent years for permeability prediction using Artificial Neural Network (ANN), none of the approaches has employed prespecified test set instead of a randomly generated test set. The methodology for selecting proper pre-specified test set was presented in chapter four of this report. The pre-specified test sets were chosen from a plot of log of permeability versus density. This approach was explicitly discussed later in the report. In this study, a pre-specified test set approach for training the network for field applicability has been developed using inputs from electric logs and flow unit obtained from geological interpretation of the pay zone. The developed ANN model was successfully applied to the Stringtown Oilfield in West Virginia. The results of this research demonstrated that the embedded powerful abilities of the ANN could be utilized to predict permeability among other important petrophysical parameters provided it was properly trained with the right pre-specified test set for field applicability.
Recommended Citation
Oyerokun, Ademola Akinwumi, "A new approach for training and testing artificial neural networks for permeability prediction." (2002). Graduate Theses, Dissertations, and Problem Reports (ETD). 10637.
https://researchrepository.wvu.edu/etd/10637