Inversion of Surface Waves Using a Dispersion Kernel Neural Network for Continuous Soil Stiffness Profiles

The surface wave method is a noninvasive technique widely used for determining the soil shear wave velocity profile, crucial for various geotechnical and earthquake engineering applications. Traditional inversion methods for analyzing surface waves often assume a layered soil stiffness profile, which may not accurately represent conditions where soil stiffness varies continuously with depth. This study introduces a novel neural network architecture, the dispersion kernel neural network (DKNN), designed to address this limitation by incorporating prior knowledge that longer wavelengths reflect deeper soil stiffness into the network. A unique data generation procedure for creating dispersion curves with continuous stiffness profiles is also proposed, enabling the production of extensive labeled datasets for training the DKNN model. The effectiveness of the DKNN is demonstrated through its application to both synthetic and field case studies. Key findings include the DKNN's ability to accurately predict soil shear wave velocity profiles with continuous stiffness variations, surpassing traditional methods in terms of efficiency. These results highlight the DKNN's potential to enhance the accuracy and reliability of continuous subsurface stiffness profile predictions.