Inversion of Rayleigh Wave Dispersion Curves Via BP Neural Network and PSO

Abstract

Rayleigh wave is widely applied in engineering exploration and geotectonic research. While how to reconstruct the corresponding geological information via Rayleigh wave is the critical process and difficulty. This paper presents an inversion method of Rayleigh wave dispersion curves based on BP neural network and PSO. In this work, a sample set that referring to the actual stratum distribution is firstly generated. Then, BP neural network is adopted to train the nonlinear mapping relationship between the dispersion curves and the shear wave velocity of each stratum. The trained BP neural network can quickly output a predicted value with rationality but poor precision, which can be utilized as the initial model of PSO inversion. PSO will then be adopted to further optimize the inversion result on the basis of BP neural network prediction. The combination of BP neural network and PSO aims at overcoming the defects of BP neural network that unable to carry out continual optimization and the slow optimization of PSO in the absence of reasonable initial solution. Finally, the effectiveness of the proposed algorithm is verified by a series of synthetic models and an active-source Rayleigh wave experiment carried out in a new railway project from Baotou, Inner Mongolia to Yinchuan, Ningxia.

Ambient Seismic Noise Tomography Within the Floridan Aquifer System, Santa Fe River‐Sink Rise, Florida, US

“…Journal of Geophysical Research: Solid Earth 10.1029/2023JB027644 Additionally, we extend our analysis by performing an initial inversion of the network-wide array average group velocity curve after Woo et al (2023a), employing the evodcinv Python library (Luu, 2023). This process iteratively inverts the sensitivity kernel velocity model with the average group velocity curve, yielding a 1D shear wave velocity model as a function of depth (Figure S22 in Supporting Information S1).…”