Evaluation des modeles de prediction des parametres petro physique des reservoirs petroliers a partir des donnees de diagraphies : Cas du Champ de Hassi Messaoud

Abstract

This study evaluates the performance of four machine learning models (Ridge, Bagging, Extra Trees, and XGBoost) for predicting porosity, permeability, and water saturation from well log data in the Cambrian reservoir of the Hassi Messaoud field. The objective is to identify which model performs best, depending on the nature and availability of input data. XGBoost showed the highest accuracy for porosity prediction (R² = 0.997), while Extra Trees performed best for permeability and saturation (R² > 0.97). The workflow includes feature selection based on Spearman correlation and feature importance, along with cross-validation. The results highlight the potential of non- linear algorithms while also acknowledging limitations due to data heterogeneity across wells