Seismic Fluid Detection with Deep Learning Elastic Property Prediction and Rock Physics

Detecting fluids is crucial for reservoir characterization. Traditionally, this is done using AVO seismic inversion to estimate elastic properties and Rock Physics Templates (RPTs) to link these properties to fluid content. However, AVO inversion can be limited by dependency on low-frequency models, elastic and isotropic medium assumptions, and assumptions of low contrast properties. These limitations can lead to errors in elastic property estimation and fluid interpretation.

DL offers a promising alternative, addressing traditional challenges by predicting elastic properties with a data-driven approach. DL can handle nonlinear relationships, large property contrasts, and input a variety of data features to guide the predictions. In this study, we investigate two key aspects: (1) Deep Learning’s ability to predict fluid-sensitive elastic properties and (2) the impact of seismic data on these predictions.

We ran tests on the Volve dataset and found that DL predictions accurately captured elastic properties and fluid trends in blind wells. DL predictions distinguished HC-bearing wells from dry wells, aligning closely with measurements. Seismic data emerged as the most critical input for detecting fluid anomalies.

The results demonstrate DL’s ability to uncover fluid-related trends quickly and accurately, highlighting its potential as a robust method for reservoir characterization and HC detection.