Predictive and comparative study of petrophysical parameters based on AI

Abstract

This study aims to explore the use of machine learning as a powerful artificial intelligence tool to develop an algorithm capable of estimating and predicting three essential petro- physical parameters: clay volume (VCL), effective porosity (P HIE), and water saturation (SW ), based on raw log data from several production wells in the Berkine Basin. The main challenge lies in the accurate prediction of water saturation. Several models were compared, including XGBoost, MLP, and CNN. The results obtained, especially with the CNN model, demonstrate the high efficiency of machine learning techniques, achiev-ing a global determination coefficient of R2 = 0.81 for water saturation, which is the most complex parameter to predict.