Neural Network for Porosity Prediction in Carbonate Formation: A Case Study of the South of Iraq

Abstract

This study presents the application of a Sequence-to-Sequence (Seq2Seq) recurrent neural network for estimating porosity log data by leveraging information from other logs. The effectiveness of this technique in simulating porosity for heterogeneous reservoirs is demonstrated by employing data from the Yamama Formation in the Faihaa Oil field in southern Iraq. Four wells in the field were used for the model training and evaluation, where input data comprised density, neutron, gamma-ray, and porosity logs. The model's performance was assessed using absolute percentage, and root means squared errors, and the results were compared against actual data, revealing a significant correlation. These results establish the Seq2Seq recurrent neural network model as a superior option for predicting reservoir porosity from other good log data. They could have practical implications in estimating porosity for petroleum calculations. .