Acoustic impedance inversion via voting stacked regression (VStaR) algorithms

Area of Science:

• Geophysics
• Petroleum Geoscience

Background:

• Acoustic impedance (AI) is vital for hydrocarbon exploration, as low AI in sandstones and carbonates often signifies high porosity.
• Accurate AI estimation is critical for identifying potential hydrocarbon reservoirs.

Purpose of the Study:

• To refine acoustic impedance estimation using advanced machine learning algorithms.
• To compare the performance of the novel VStaR model against existing methods like VSR and BLIMP.

Main Methods:

• Employed stacking and voting regression algorithms with depth, two-way travel time (TWTT), and nine seismic attributes as inputs.
• Implemented models using scikit-learn, focusing on hyperparameter tuning for the VStaR model.

Main Results:

• The VStaR model demonstrated superior predictive performance (R²=0.9973) and fitting accuracy (True > VStaR > VSR > BLIMP).
• VStaR reduced Root Mean Square Error (RMSE) by 14.74% compared to the best base model, despite longer computation time.
• VStaR outperformed VSR and BLIMP in Mean Squared Error (MSE), RMSE, and R² metrics.

Conclusions:

• The VStaR method offers a novel and superior approach to acoustic impedance prediction.
• This improved AI estimation has the potential to significantly enhance hydrocarbon exploration effectiveness, particularly in carbonate datasets like Illam.