Facies control on machine learning of acoustic impedance
Hongliu Zeng, Yawen He, Leo Zeng · 2021
We develop a random-forest machine-learning (ML) workflow to predict acoustic impedance from stacked and migrated 3D seismic data. The training, validating, and testing are performed with 40 features extracted from a geologically realistic 46×66-trace model built in Miocene Powderhorn field in south Texas. We focus on the responses of the random forest model to sedimentary facies and the strategies to achieve better prediction with various data conditions. We apply explained variation (R2) and root-mean-square prediction errors (RMSE) to map the relationship between quality of prediction and sedimentary facies. The training with a small well data set (1–10 wells) leads to low and unstable test scores (R2 = 0.2–0.7). The R2 score increases and stabilizes with more (as many as 1000) training wells (R2 = 0.7–0.9). Sparse well-supported models can outperform linear regression and model-based inversion and can be useful if cautions are exercised. Such results can be applied to exploration and production practices.