Determination of Horizontal Permeability Through a Probability Neural Network Approach

Abstract

The mam objective of this study is to present an alternative methodology based on the Probability Neural Network (PNN) fonnalism to estimate and further predict the penneability values in any location of a 3D grid of a reservoir model. An experimental data set containing measurements of log porosity, true vertical depth, transitive time and lithologies at well locations, was used for training the PNN. Afterwards, the trained PNN is applied to the wells where does not exist core infonnation, but logs. The penneability values estimated by PNN in the wells are used to simulate the penneability at all spatial locations of the reservoir. This methodology was applied to experimental data from a reservoir located at West Africa with encouraging results.

Prediction of Reservoir Compressibility Using Subsurface Cores, Well Logs, and Seismic Data by Neural Network

“…The most straightforward and e cient type of neural network is the multilayer perceptron model, which consists of an input layer, one or more hidden layers, and an output layer (Puskarczyk 2019) asymptotically approach the underlying parent density, provided that it is smooth and continuous (Soares et al 1996). PNN was also applied for a more precise prediction.…”