Acoustic impedance inversion via voting stacked regression (VStaR) algorithms.

In this study, we focused on improving acoustic impedance (AI) in seismic exploration. AI is a crucial parameter estimated by multiplying the density of a material by the velocity of an acoustic wave passing through it. A low AI in sandstones and carbonates often indicates high porosity, which enhances hydrocarbon accumulation. Accurate AI estimation is thus critical for reliable hydrocarbon exploration. To refine the AI estimation, we used stacking and voting regression algorithms, with depth, two-way travel time (TWTT), and nine seismic attributes as inputs. All models were implemented using scikit-learn . The VStaR model achieved superior predictive performance ([Formula: see text]= 0.9973) and yielded a more accurate fitting parameter (a = 0.1584) in the acoustic impedance-porosity transformation compared to the VSR ([Formula: see text]= 0.9775, a = 0.1583). The VSR approach made the voting of a top-performing base model with two less predictive base models, as used in the existing literature. Relative to the true and BLIMP-derived impedance, the fitting accuracy followed the order of true > VStaR > VSR > BLIMP. While VStaR required longer computation time (≈ 400 s), it reduced RMSE by 14.74% compared to the top-performing base model. VStaR outperformed all evaluated models based on MSE, RMSE, and [Formula: see text] metrics. The novelty of the VStaR method based on hyperparameters lies in its superior performance in obtaining a more precise prediction of acoustic impedance compared to the VSR and conventional BLIMP method, potentially improving the effectiveness of hydrocarbon exploration in Illam carbonate dataset.