Permeability prediction using neural network model with post-processing

Abstract

Practicable and realistic reservoir characterization is essential for optimal reservoir management. In this study, a randomized back-propagation neural network model is developed for fonnation permeability prediction. The model has only one hiddenlayer, and tbe inputs to the model are core porosity, facies identifier, sample thickness, and well sample location. A number of sensitivity studies for permeability prediction are performed. Prediction errors from the model are analyzed and a post. processing scheme for error mitigation is investigated. Neural network responses were compared with those using conventional methods for permeability determination. There are some specific advantages of using the developed model. Characterization of prediction space is observed to be better. However, the limitations of the study were also highlighted. A variety of applications of artificial neural networks in reservoir engineering problems are reviewed in this study.