基于APSOＧMCMC的叠前三参数同步随机 反演方法研究

Deterministic inversion of prestack data has problems concerning inaccuracy, instability, poor noise resistance, and exces-sive dependence on the initial model. Therefore, a stochastic inversion method was previously proposed to determine the prestack three-parameter simultaneous inversion; however, the efficiency of conventional stochastic inversion is low. In this study, the AP-SO-MCMC method was developed to improve the computational efficiency of the prestack simultaneous inversion. According to the statistical perturbation relationship of the P-wavc velocity, S-wavc velocity, and density, the Gaussian distribution of the model per-turbation is constructed. Based on the acceptance-rejection method, an initial swarm is generated as an input for the inversion process. During the iteration, a transition matrix is employed to change the evolution of the particles. Additionally, the evolution fac-tor was calculated based on the distance between the particles to determine the convergence state of the particle swarm. In each it-eration, an elite learning strategy was employed to determine the jump-out of the local minimum. Because the conventional forward modeling algorithm is not accurate in wide-angle regions, this study employed the Zocppritz equations to ensure the accuracy of the prcstack three-parameter inversion. The field data test of the NS area shows that the method correctly estimates the P-wavc veloci-ty, S-wave velocity, and density, and has a stability of convergence, high resistance to noise, and better efficiency than conventional stochastic approaches.