Abstract:

An iterative inversion of microtremor surface wave dispersion curves by improved neural network is proposed. The learning speed is improved by batch processing and optimizing the network structure in network training. In order to reduce the iteration number of the inversion, the input model is adjusted according to the changes of the extreme points of  v_R/λ_R-f curve during the inversion. The inversion program is designed and applied to the inversion of dispersion curve from a six-layer model to validate the method. The results show that the method can reduce the iteration number significantly and improve the convergence rate, while is of good anti-noise capability.

Key words: improved BP neural networks, microtremor, dispersion curve, surface waves, iterative inversion