Research On Geoacoustic Inversion Using Marine Ambient Noise

Ocean ambient noise carries abundant environment information about water and subsurface structure.The usage of ambient noise for passive interferometry and geoacoustic inversion has attracted ever-increasing attention in the past decades.The technique requires that the noise field has to be equipartitioned,which means the field has to be generated by a homogeneous distribution of random and uncorrelated noise sources.However,the existence of strong interferences destroys the equipartition property of noise fields in the real world,making the ambient noise difficult to use.Recently,a number of temporal and spectral pre-processing techniques have been developed to improve the quality of the noise field.But it is still very hard to deal with stubborn interferences,especially the interferences which last for a long time.On the other hand,traditional geoacoustic inversion methods suffer the drawbacks of high computational cost and slow inversion speed for a long time,which can not meet the need of real-time monitoring and fast inversion.This thesis focuses on several aspects,including the analysis of ambient interferometry,the suppression of interferences,ambient tomography and inversion approaches,to help solve these problems.The main work and contributions of this thesis include:1.The method,theory and technical route of using ocean ambient noise to extract the surface-wave Green’s function are studied,the ocean ambient noise data collected by submarine arrays are processed and analyzed,and the influence of the strong directional noise on the equipartitioned property of the noise field and on the reliability of Green’s function are discussed.By studying the statistical characteristics of the eigenvalues of the noise field sample covariance matrix under the condition of ideal equipartition,and using the random matrix theory,an adapted eigenvalue-based filtering method is proposed,which can effectively suppress the influence of stubborn and strong directional noise on the Green’s function.Experimental results show that the adapted eigenvalue-based filtering method can restore the equipartition of the noise field to a certain extent,thereby improving the quality and reliability of the surface-wave Green’s function.2.The theory and technical route of shear wave tomography using surface-wave information have been studied.The surface-wave dispersion curve is extracted using the slowness-frequency transformation method,and the adaptive simplex simulated annealing algorithm is used to inverse the corresponding subsurface structure and shear wave information from the extracted dispersion curves.The inversion results were used to construct a three-dimensional tomography map of the stratigraphic structure.The comparison with the existing tomography results shows that although the noise record with a short duration was used and the existence of the strong interferences,a reliable tomography results could still be obtained.3.The mixture density neural network under the Bayesian inference framework and its application in the field of geoacoustic parameter inversion are studied.The Bayesian geoacoustic parameter inversion framework is extended by using the mixture density network theory.By deriving the analytical expression of the geoacoustic parameter statistics,the relationship between the geoacoustic parameter statistics and the multi-dimensional posterior probability density is clarified,and the challenges in using the mixture density network inversion is solved.A mixture density neural network is designed and implemented for geoacoustic parameter inversion.Simulation analysis and real data inversion experiments show that the network can give reliable predictions and has good generalization performance on new data.The problems of high computational cost and slow inversion speed of traditional inversion methods have been well solved by the application of the mixture density network inversion method,providing a possibility for real-time monitoring and fast inversion.