基于模型分块交叉移动的学习型模拟退火算法地震反演

针对地震非线性反演问题,提出一种基于模型分块交叉移动的学习型模拟退火的全局优化地震反演方法.其步骤为:首先,在模拟退火算法及粒子群算法基础上,在算法模型扰动项里面加入1个向目标优化的方向移动的学习项；其次,针对地震反演模型数量多及地震记录为褶积形式的特点,采用模型分块交叉移动的方法来实施模拟退火反演,给出模型分块交叉移动的学习型模拟退火算法流程.研究结果表明:该方法具有收敛速度快、精度高、实现简单、高效的特点,可以用于其他多维多极值的目标函数反演.

Seismic inversion based on simulated annealing of learning and cross-moving with divided block model

To solve the seismic nonlinear inverse problem,a seismic inversion method was presented based on simulated annealing of learning and cross-moving with divided block model.The algorithm was proposed in two aspects.Firstly,based on the algorithm of simulated annealing and particle swarm optimization,simulated annealing formula of perturbed model was added with a learning item which moved to the direction of objective optimization.In addition,because the seismic inversion model’s number are large and form of seismic records is convolution,simulated annealing algorithm was implemented by the way of divided block model cross-moving.The results show that the proposed method has fast convergence and high accuracy.The new simulated annealing is simple and efficient,and it can be used to other multidimensional object function inversion.