| Abstract Submission No. | ABS-4.1-0180 |
| Title of Abstract | Hybrid ML & DL Approaches for Estimating Petrophysical Properties from Inverted Volumes and Seismic Attributes |
| Authors | Tignangshu Das*, Partha Pratim Mandal |
| Organisation | Indian Institute of Technology (Indian School of Mines), Dhanbad |
| Address | IIT(ISM), Dhanbad Dhanbad, Jharkhand, India Pincode: 826004 E-mail: 23mc0111@iitism.ac.in |
| Country | India |
| Presentation | Poster |
| Abstract | Accurate Prediction of petrophysical properties from seismic data is a key step in reservoir characterization, but it remains challenging due to the non-uniqueness of inversion and uncertainties in linking seismic to petrophysical parameters. Traditionally, petrophysical properties such as porosity, water saturation, and volume of shale are estimated from inverted seismic volumes (Acoustic Impedance and Vp/Vs) using rock physics inversion. In this work, an alternative data-driven approach is proposed where machine learning and deep learning algorithms are used in place of rock physics inversion. The methodology begins with pre-stack seismic inversion to obtain acoustic impedance and Vp/Vs volumes. In addition to these inverted volumes, seismic attributes (e.g., sweetness, coherence, etc.) are extracted at well locations by considering the traces around the well path to build a comprehensive input dataset. Wireline-log derived petrophysical properties are used as output variables for model training and validation. A range of supervised machine learning models and deep neural networks are trained and tested for predicting porosity, water saturation, and volume of shale. Model validation is carried out using a blind well to evaluate generalization of prediction performance. Uncertainty quantification is also incorporated to provide probabilistic estimates (confidence) of the predicted petrophysical properties, which helps in assessing prediction reliability. Application of the proposed workflow to a field dataset shows that integrating pre-stack inversion with seismic attributes under a machine learning framework provides reliable petrophysical predictions. This approach demonstrates the potential of replacing inverse rock physics modelling with data-driven methods for more reliable and effective reservoir characterization. |
| Keywords | Pre-stack Inversion; Petrophysical Inversion; Machine Learning; Uncertainty Quantification |