Speaker
Description
Abstract
One of the key logs used for lithological differentiation and the analysis of facies in petroleum reservoir characterization is the Gamma Ray (GR) log, however these logs are mostly limited or incomplete especially in frontier basins. This presents a challenge in predicting lithologies for reservoir characterization and seismic to well integration. This research develops a machine learning workflow to predict a robust 3-Dimensional GR volume directly from seismic attributes using Artificial Neural Networks (ANNs). The research was applied in the deepwater Tano Basin, Offshore Ghana which is a turbidite heterogenous stratigraphy with well scarcity and minimal well control.
Nine (9) vertical wells and three (3) derived attributes from Seismic data (Sweetness, Gradient Magnitude and Envelope) were used for the study. The datasets were rigorously quality controlled and normalized. A Multilayer perceptron was tested and optimized on a blind test well. High predictive performance with R2 values above 0.7 and Root Mean Square Error below 10 API. The predicted ANN GR volume accurately captured the differences in lithologies and transitions, stratigraphic boundaries and channel geometries. These observations from the predicted ANN volumes were consistent with well log responses and depositional models. Clean sand intervals were delineated from the blind well test used for independent validation. The generated GR volume enhances seismic facies identification and reservoir analysis in areas with no wells.
This methodology provides a data driven scalable basis for improved seismic interpretation and reservoir characterization enhanced by Machine learning, specifically Artificial Neural Network in deepwater frontier Basins.