Integrated neural network–geostatistical modeling for data-driven estimation of shear wave velocity from conventional well logs: A case study from an Iranian oil reservoir

چکیده مقاله

Shear wave velocity (Vs) estimation is essential in reservoir geophysics, particularly when direct measurements are unavailable. This study introduces a hybrid framework that integrates artificial neural networks (ANNs) with geostatistics to estimate Vs from standard well log data. The ANN was first trained on well log data to capture nonlinear relationships between input features and target. The ANN predictions were then incorporated as local mean values in the ordinary kriging (OK) algorithm, enabling spatial variability and geological continuity to be reflected in the final estimation. The novelty of this work lies in combining ANN with geostatistics, an approach not previously applied in reservoir geophysics for Vs prediction. This integration captures both nonlinear parameter interactions and spatial variability, leading to a clear improvement in accuracy compared to conventional machine learning or empirical methods. Model performance was evaluated using cross-validation, where subsets of data were systematically removed and predicted independently. Results showed that the ANN model achieved a correlation of 97.9 % with an RMSE of 63.9 m/s, while the hybrid ANN–OK approach reached 99.8 % correlation and a lower RMSE of 20.06 m/s. These findings confirm that integrating ANN with geostatistics enhances precision and spatial reliability in Vs estimation, providing a robust tool for subsurface characterization in data-limited environments.