Prediction of fast, slow, and average shear-wave velocities from well logs using artificial neural networks and statistical regression
Abstract
Accurate estimation of fast, slow, and average shear-wave velocities (Vs Fast , Vs Slow , and Vs Average ) is important for evaluating reservoir anisotropy, fracture networks, and stress regimes. In this study, multivariate statistical regression and artificial neural network (ANN) models were developed and compared to predict these velocity components in a well from the Sarvak Formation, Zagros Basin, Iran. The initial dataset comprised 3,234 depth points, of which 2,264 valid points were retained after data preprocessing and normalization. Compressional-wave velocity (Vp), neutron porosity (NPHI), and bulk density (RHOB) were used as input variables. For the statistical models, 70% of the data were used for calibration and 30% for testing, whereas the ANN employed a feedforward backpropagation architecture trained using the Levenberg–Marquardt algorithm, with 70, 15, and 15% of the data allocated to training, validation, and testing, respectively. The statistical models achieved test-set R² values of 0.769129, 0.616225, and 0.725904 for Vs fast , Vs slow and Vs Average , respectively, while the corresponding ANN models achieved R² values of 0.786769, 0.641601, and 0.727609. The ANN therefore provided modest improvements over statistical regression, with the largest improvement observed for Vs Fast and Vs slow . Minor discrepancies at lower velocity ranges may reflect local geological complexities, including intense fracturing. Overall, the proposed approach, particularly the ANN models, provides a practical indirect method for estimating shear-wave velocity components from conventional well logs and may support reservoir characterization where direct dipole-shear measurements are unavailable.
Related
- Mapping soil salinity and prediction uncertainty across different depths using quantile regression forests
- Time Series Fuzzy Bayesian Logistic Regression for Wildfire Event Prediction
- Optimization of UV-C irradiation time for biologically treated saline wastewater using artificial neural networks and a hybrid multi-criteria decision-making framework
- WP-MIP: An Artificial Intelligence, Hybrid, and Physically Based Model Intercomparison Project for Weather Prediction
- Trends in soil organic carbon prediction research using geospatial technologies and artificial intelligence: an exploratory bibliometric analysis (2005-2025)
BibTeX
@article{teymouri2026prediction,
title = {{Prediction of fast, slow, and average shear-wave velocities from well logs using artificial neural networks and statistical regression}},
author = {Hamed Teymouri and Ramin Nikrouz and Aliasghar Ciabeghodsi and Ali Kadkhodaie},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-72419-9},
url = {https://www.nature.com/articles/s41598-026-72419-9},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Prediction of fast, slow, and average shear-wave velocities from well logs using artificial neural networks and statistical regression
AU - Hamed Teymouri
AU - Ramin Nikrouz
AU - Aliasghar Ciabeghodsi
AU - Ali Kadkhodaie
PY - 2026
DA - 2026/10/02
JO - Scientific Reports
DO - 10.1038/s41598-026-72419-9
UR - https://www.nature.com/articles/s41598-026-72419-9
AB - Accurate estimation of fast, slow, and average shear-wave velocities (Vs Fast , Vs Slow , and Vs Average ) is important for evaluating reservoir anisotropy, fracture networks, and stress regimes. In this study, multivariate statistical regression and artificial neural network (ANN) models were developed and compared to predict these velocity components in a well from the Sarvak Formation, Zagros Basin, Iran. The initial dataset comprised 3,234 depth points, of which 2,264 valid points were retained after data preprocessing and normalization. Compressional-wave velocity (Vp), neutron porosity (NPHI), and bulk density (RHOB) were used as input variables. For the statistical models, 70% of the data were used for calibration and 30% for testing, whereas the ANN employed a feedforward backpropagation architecture trained using the Levenberg–Marquardt algorithm, with 70, 15, and 15% of the data allocated to training, validation, and testing, respectively. The statistical models achieved test-set R² values of 0.769129, 0.616225, and 0.725904 for Vs fast , Vs slow and Vs Average , respectively, while the corresponding ANN models achieved R² values of 0.786769, 0.641601, and 0.727609. The ANN therefore provided modest improvements over statistical regression, with the largest improvement observed for Vs Fast and Vs slow . Minor discrepancies at lower velocity ranges may reflect local geological complexities, including intense fracturing. Overall, the proposed approach, particularly the ANN models, provides a practical indirect method for estimating shear-wave velocity components from conventional well logs and may support reservoir characterization where direct dipole-shear measurements are unavailable.
SN - 2045-2322
ER -