A fusion network for surface wave dispersion curves inversion

In order to obtain the subsurface shear wave velocity structure quickly and accurately, in this paper, a fusion network inversion analysis method is proposed to solve the disadvantages of the traditional population search optimization inversion method, which has high computational cost and limited applicability. The method uses deep learning training time instead of computing time, so as to achieve near real-time inversion of the dispersion curve. This paper establishes the data set of the velocity structure of the dispersion curves according to the Vs-h empirical relationship in different regions, and constructs and trains the fusion network model (PSO-LSTM-1DCNN-DNN) on this basis, which has a good effect on both validation and test sets. In addition, this study also add 10 % Gaussian noise to the dispersion curve to analyze its uncertainty, and improve the anti-noise ability of the algorithm by Adversarial Autoencoder (AAE). The results show that the accuracy and efficiency of surface wave dispersion curve inversion are significantly improved by the fusion network.