中文摘要：

针对波数域中Parker-Oldenburg位场迭代反演方法中存在向下延拓的问题，提出了利用改进的遗传算法进行三维Moho界面反演的位场遗传反演方法。通过模型试算证明了该方法的有效性和稳定性，并利用该方法反演了青藏高原东缘三维Moho界面的分布。

英文摘要：

Parker-Oldenburg method is an iterative inversion method with potential field data in wavenumber domain, which is commonly used for the inversion of interface depth. But this method has a problem of downward continuation in course of the inversion. The downward continuation will bring about high frequency noise. An improved genetic algorithm can be used to invert the three-dimensional Moho structure with potential field data. This new inversion method includes both the advantages of genetic algorithm and Parker formula. The Parker formula is used for forward of gravity anomaly, and the genetic algorithm is used for inversion of Moho depths. The Oldenburg formula is not used in order to avoid the downward continuation. The validity and stability of the new method are proved by synthetic data. The three-dimensional Moho depths in eastern Tibetan plateau are inverted by this new method with Bouguer gravity anomaly.