An Application of Machine Learning to Predicting Shear Wave Velocity of Marine Sand

1. Home
2. Publications
3. Online Library
4. An Application of Machine Learning to Predicting Shear Wave Velocity of Marine Sand

ABSTRACT: Empirical correlations from multiple regression analysis (MRA) and deep-learning algorithm (DLA) models were developed to estimate and predict shear wave velocity (Vs) in marine sand. This study presents a database of ten sites with in situ measurements of Vs using P-S logging and three empirical correlations developed using MRA. DLA models also were developed to predict  Vs using two and six input soil parameters. A comparison across nine sites between  Vs predictions using DLA and MRA equations is presented, followed by blind  predictions. Results showed both DLA and MRA were effective in predicting Vs   when using two soil parameters, with the highest accuracy using s'v and qc. However, using six soil parameters did not improve predictions compared to using two soil parameters. Blind predictions at an untrained site showed DLA outperformed MRA, confirming s'v and qc  as critical parameters for  determination. The results indicate DLA achieved higher accuracy in predicting Vs profiles than MRA, highlighting DLAs potential to support project planning and survey optimization for geotechnical and geophysical site investigations. The DLA model using s'v and qc  developed in this study can be applied to predict Vs  at various offshore wind turbine locations within an offshore wind farm. By using in situ measurements from a subset of locations, the model can be trained and tested to predict Vs at other locations that have in situ measurements of  qc but lack  Vs measurements, thereby supporting survey planning and potentially reducing survey costs through optimized field testing.

5th International Symposium on Frontiers in Offshore Geotechnics (ISFOG2025)

5 - Data Analytics and Machine Learning