Machine learning shear wave velocity prediction based on multi-channel analysis of surface wave
Shear Wave Velocity (Vs) profile plays a crucial part in determining seismic site classification. However, field measurement incurs extra cost and time. Economic factors urge the need for a cheaper and faster alternative. Previous studies proposed the use of empirical equations, however there are gr...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/102258/1/MarkRubenUpomMSKA2021.pdf.pdf http://eprints.utm.my/id/eprint/102258/ http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:145643 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Shear Wave Velocity (Vs) profile plays a crucial part in determining seismic site classification. However, field measurement incurs extra cost and time. Economic factors urge the need for a cheaper and faster alternative. Previous studies proposed the use of empirical equations, however there are growing evidence that Machine Learning (ML) methods may produce better results. Thus, this study was designed to develop a feasible method of predicting Vs value using ML Models. Due to the impact of weathering profile on seismic site classification, the results of this study are limited to sites with similar geological formation of the study area, which is composed of granitic rocks. The study utilized four types of ML algorithms to develop the predictive model. The ML algorithms used were Multi Linear Regression (MLR), Random Forest (RFR), Artificial Neural Network (ANN) and Support Vector Machine (SVR). The independent variables are Standard Penetration Resistance (Nspt) and depth of soil (Ds), while the dependent variable is Vs. Consequently, this study conducted a Multichannel Analysis of Surface Wave (MASW) survey to get the required dataset. Furthermore, this study verified the Vs profiles using Nspt data. In addition, the hyperparameters for the ML models were determined using Random Search and k-fold Cross Validation. On top of that, this study also used Coefficient of Determination (R ), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) as the performance metrics for model selection. The best ML model was determined to be RFR based on the performance metrics (R = 0.9, MAE = 16.93 and RMSE = 19.79). It was then determined that the average percentage difference between the actual and predicted Vs30 was 10.7%. This study also presents the development of a software, pyMASW, for the processing of the raw seismic data. In conclusion, the RFR model can predict Vs30 values for seismic site classification with an accuracy of 89.3%. |
|---|