Interwell Saturation Prediction by Artificial Intelligence Analysis of Well Logs

Abstract

Well log analysis, through deploying advanced artificial intelligence (AI) algorithms, is key for wellbore geological studies. By analyzing different well characteristics with modern AI tools it becomes possible to estimate interwell saturation with improved accuracy, outlining primary fluid channels and saturation propagations in the reservoirs interwell region. The development of modern deep learning and artificial intelligence methods allows analysts to predict interwell saturation as a function of observed data in the near wellbore logged geological layers. This work addresses the use of deep neural network architectures as well as tensor regression models for predicting interwell saturation from other well characteristics, such as resistivity and porosity, as well as local near-well saturation. Several algorithms are compared in terms of both accuracy and computational efficiency. Sensitivity analysis for model parameters is carried out, which is based on the wells’ geometry, radius, and multiple sampling techniques. Additionally, the impact of local saturation prior knowledge on the model accuracy is analyzed. A reservoir box model encompassing volumetric interwell porosity, resistivity and saturation data was utilized for the validating and testing of the AI algorithms. A prototype is developed with Python 3.6 programming language.