Machine learning based modeling of the dynamic Poisson’s ratio for petroleum oil field

Abstract Poisson’s ratio is the fundamental geomechanical property that governs the stress and strain profiles of the geological formations. It is the prime characteristic in geomechanical modelling of the rock deformation and failure criteria. However, the accurate prediction of the Poisson’s ratio from geophysical log data using conventional techniques is a challenging task, owing to their time consuming and characteristic reliability issues. In this context, the robust machine learning (ML) based decision tree (DT) and linear regression (LR) models have been developed in the present work to evaluate the dynamic Poisson’s ratio of Volve field formation. The modelling is carried on geophysical log data gathered from 15/9-F-1B well of the Volve field. 1,500 data points captured in the depth range of 3,250–3,400 m of the 15/9-F-1B well are trained, tested, and validated to establish the reliability of the developed model. The analysis projected a test score of 0.99 for LR based model training and testing. Further, the score metrics projected the lowest maximum error for LR among the various machine learning (ML) methods employed, with a magnitude of 0.0173. The validation analysis projected a good history match of the modeled Poisson’s ratio values with a validation score of 0.982. While LR demonstrated a high test of R 2 0.994, residual analysis revealed systematic errors, indicating potential nonlinearity in the data. Conversely, DT, with a lower test R 2 of 0.904, exhibited well-distributed residuals, effectively capturing nonlinear relationships. However, DT’s tendency to overfit ( R 2 = 1.000 on training data) limits its generalization. The findings suggest that while LR provides superior generalization, the presence of nonlinearity necessitates exploring polynomial regression or ensemble tree-based models to enhance predictive accuracy. The present work demonstrates that the use of machine learning to estimate Poisson’s ratio from geophysical logs is a promising strategy and hence, is useful in capturing insightful information from geo-mechanical engineering applications.