Estimating wave velocity profiles and earthquake ground motion intensities using machine learning algorithms

Abstract

Optimization techniques are essential for solving complex problems across various domains. Rule-based and data-based algorithms have been proven effective in solving the issues. Machine learning is an optimization technique that utilizes computational algorithms to enable computers to statistically capture data features. This technique allows computers to learn without being explicitly programmed and involves adjusting algorithms based on predefined information to achieve acceptable performance on unseen data. In earthquake engineering, machine learning-based approaches are gaining popularity for automated decision-making processes with low computational costs. One of the concerns in the field of earthquake engineering is earthquake hazard assessments. These assessments are important to evaluate the potential risks associated with earthquakes and ensure the safety and resilience of social infrastructure. The assessment involves analyzing the seismic responses at a particular site during an earthquake, which are affected by the geotechnical conditions, such as the properties of the soil and rock. Therefore, geological surveys are conducted to obtain the velocity profiles that can represent the ground conditions. Obtained velocity profiles are frequently utilized in various research areas, such as mitigating earthquake damage. It is preferable to measure wave velocities directly if it is possible. However, such measurements are often constrained by time, cost, and site conditions. If measurements are unavailable, wave velocities need to be estimated based on other known proxies. These predictions can help assess the seismic intensity of ground motions, by serving as an independent variable in ground motion prediction models. Considering the prediction of velocity profiles and seismic intensities of ground motion can help address future uncertainties in the field of earthquake engineering. This dissertation aims to make predictive models of wave velocities and seismic intensities of ground motion prediction using machine learning approaches. Various factors are considered, such as depth, N-value, density, slope angle, elevation, geological feature, soil/rock type, and site coordinates to predict the compression and shear wave velocities (VP and VS, respectively) in Japan. Furthermore, the following independent variables: ML, Repi, VS30, focal depth, and slope angle are utilized to predict 5% damped pseudo-spectral acceleration (PSA) for 27 periods ranging from 0.01 to 10 s in South Korea. The variables that exhibit a strong influence on predicting abilities are identified.