Machine Learning-Based Prediction of Shear Wave Velocity: Performance Evaluation of Bi-GRU, ANN, and The Greenberg-Castagna Empirical Method
Year: 2025
Type: article
Abstract: Shear wave velocity (Vs) is recognized as an important elastic parameter for lithology and fluid identification in oil and gas exploration. However, Vs data is not always recorded in well logs. Various empirical approaches are often used to estimate Vs, but these methods show limitations in terms of accuracy and time efficiency. With technological advances, machine learning has become an effective and efficient alternative for predicting Vs from well log data. This study is utilizing the Bi-GRU model, a sophisticated artificial neural network specifically designed to process sequential data. This capability makes Bi-GRU particularly suitable for predicting log Vs data. Four Bi-GRU modeling scenarios are being developed with different hyperparameter configurations and are being compared with ANN models using two input variations: with and without Vp data. The results show that scenario 2 (Bi-GRU with five hidden layers, batch size 64, learning rate 0.005) is achieve the best performance, with R² values of 0.9787 (without Vp) and 0.9868 (with Vp). The MAE values obtained are being recorded as 9.36 (without Vp) and 11.22 (with Vp). Compared to shows ANN, MLR, and empirical Castagna methods, the Bi-GRU model show a more significant improvement in prediction accuracy. These findings are indicating that Bi-GRU have strong potential for accurately and efficiently predicting Vs from well log data.
Institution: Bandung Institute of Technology
Language: English
Subfield: Geophysics
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4415854473
