Comparative Study of Predictive Models for Permeability from Vertical wells using Sequential Gaussian Simulation and Artificial Neural Networks

Abstract

This study attempts to estimate permeability from well logs data and also predict values from existing rock sections to points that are missing using Artificial Neural Network (ANN) and Sequential Gaussian Simulation (SGS). Potentially, exploration data is prone to trends that are initiated by the sedimentation process, but a detrending method using Semi-variogram (vertical) algorithm was applied to remove this from the interpreted wells which are all vertical. Permeability modeled for ANN gave an estimated root mean square error (RMSE) of 0.0449, while SGS gave RMSE of 0.1789, both giving a ‘K’ range of 100 – 1000 mD. Although the spatial geology of the area was relegated and not considered, making a spatial prediction influenced from the temporal reference point un-assessable. However, the independent prediction on the overall result shows a better prediction from the ANN, perhaps due to the optimization algorithm used.