大地电磁测深阶段式自适应正则化反演

如何合理确定正则化因子一直是地球物理反演的研究热点和难点。从提高反演稳定性的角度对正则化因子的选择进行考量，是确定正则化因子取值的新思路。此外，以往的正则化反演研究对非线性优化算法的随机性考虑不足。基于此，以Zhdanov提出的自适应算法为框架，提出了一种新的自适应正则化算法，即阶段式自适应算法，按照“阶段”自适应地调整正则化因子。将此方法分别应用于大地电磁测深（MT）的共轭梯度和差分进化算法（DE）反演。模型试验表明该算法可以提高反演的稳定性，在一定程度上降低衰减因子的影响，而且该算法具有同时适用于线性和非线性优化算法的特点。

Staged adaptive regularized inversion of magnetotelluric data

How to reasonably choose a regularization factor has always been a hotspot and difficulty in geophysical inversion.Choosing the regularization factor from the perspective of improving the stability of inversion is a new idea. In addition,the randomness of nonlinear optimization algorithm is not considered enough in previous studies on regularized inversion. Based on the adaptive algorithm proposed by Zhdanov,we propose a self-adaptive regularized algorithm,also called "Staged Adaptive Algorithm",which adjusts the regularization factor stage by stage. We use the algorithm for magnetotelluric (MT) conjugate gradient inversion and differential evolution inversion. Model tests show that the new adaptive algorithm can improve the stability of inversion,reduce the influence of the attenuation factor to some extent,and it is applicable for linear and nonlinear optimization algorithms.