大地电磁的多尺度反演

A MULTIRESOLUTION INVERSION OF ONEDIMENSIONAL MAGNETOTELLURIC DATA

对于迭代方式的参数化反演方法，如何使反演结果稳定地收敛到整体极小仍是目前大地电磁（MT）反演中急需解决的问题．本文利用小波变换理论中的多尺度分析方法将大地电磁反问题分解为依赖于尺度变量的反问题序列，然后按尺度从大到小的次序依次求解，求解过程中前一个尺度反问题的解作为下一个尺度反问题的初始模型，直到来出对应于尺度为0的原反问题的解为止．该方法称为多尺度反演方法．数值试验和实际资料的反演结果表明，该方法可有效改善传统广义逆反演方法易陷入局部极小的弊端．

For magnetotelluric (MT) inversion, when initial guesses are far from the global minimum, the results are often trapped by some local minimums by using linear inversion methods. This moviated us to develop a new method, called multiresolution inversion (MI) method, to overeome the troubles mentioned above. The basic idea of MI is tha the original MT inverse problem can be decomposed into a set of inverse problems at different scales by multiresoluhon analysis (MRA), then the inverse problem of the coarest scale is solved firshy and its final result is implemented as an initial model of the inverse problem of the second coarest scale. This process is repeated unhl the inverse problem of the finest scale is solved. Two numericalexamples of ID MT theorehcal model and a real MT profile interpretation show the algorithm is stable, less dependent on the initial guesses, and can improve the inverseresolutions.