Application and performances of unconstrained optimization methods in seafloor AVO inversion

Seafloor elastic parameters are important for seafloor engineering and geophysical detection beneath the seafloor. We have proposed to use AVO (amplitude variation with offset) theory to estimate seafloor elastic parameters. However, the previous inversion methods are time-consuming. To improve the computing efficiency, we try to solve the inverse problem as an unconstrained optimization problem in this paper. Three kinds of classical unconstrained optimization methods are applied to seafloor AVO inversion, including the steepest descent method, the Newton’s method, and the conjugate gradient method. Then, we design different initial models to test the convergence behaviors of the three methods. Numerical tests show that the perturbation level of the initial models and the noise level of the observed data have a significant effect on the convergence performances of the three methods. Even for the same perturbation and noise levels, the convergence performances differ with different combinations of the perturbed initial elastic parameters. All three methods have higher computing efficiency than the previous methods. This research also offers a strategy to choose a proper optimization method for a specific case in real seafloor AVO inversion.