Behnia Azizzadeh mehmandoust Olya - School of Mining Engineering, College of Engineering, University of Tehran, Tehran, Iran.
Reza Mohebian - School of Mining Engineering, College of Engineering, University of Tehran, Tehran, Iran.

Potential mapping of Permeability is a crucial factor in determining the productivity of an oil and gas reservoirs. Accurately estimating permeability is essential for optimizing production and reducing operational costs. In this study, we utilized the CUDNNLSTM algorithm to estimate reservoir permeability. The drilling core data were divided into a training pool and a validation pool, with ۸۰% of the data used for training and ۲۰% for validation. Based on the high variation permeability along the formation, we developed the CUDNNLSTM algorithm for estimating permeability. First, due to the highly dispersed signals from the sonic, density, and neutron logs, which are related to permeability, we adjusted the algorithm to train for ۱۰۰۰ epochs. However, once the validation loss value reached ۰.۰۱۵۸, the algorithm automatically stopped the training process at epoch number ۵۰۰. Within ۵۰۰ epochs of the algorithm, we achieved an impressive accuracy of ۹۸.۴۲%. Using the algorithm, we estimated the permeabilities of the entire set of wells, and the results were highly satisfactory. The CUDNNLSTM algorithm due to the large number of neurons and the ability to solve high-order equations on the GPU is a powerful tool for accurately estimating permeability in oil and gas reservoirs. Its ability to handle highly dispersed signals from various logs makes it a valuable asset in optimizing production and reducing operational costs, because it is much cheaper than the cost of core extraction and has very high accuracy.

\otential mapping, Permeability estimation, Deep learning, CUDNNLSTM, Oil and gas reservoir’s