实数编码遗传算法在大地电磁测深二维反演中的应用

作为全局非线性优化的新方法之一的遗传算法,近年来已从生物工程流行到大地电磁测深资料解释中.然而,大地电磁反演问题具有不适定性,解的非唯一性.通过结合求解不适定问题的Tikhonov正则化方法,本文采用实数编码遗传算法求解大地电磁二维反演问题.此算法在构建目标函数时引入正则化的思想,利用遗传算法求解最优化问题.常规的基于局部线性化的最优化反演方法易使解陷入局部极小值,而且严重的依赖初始模型的选择.与传统线性化的迭代反演方法相比,实数编码遗传算法能够克服传统方法的不足且能获得更好的反演结果.通过对大地电磁测深理论模型进行计算,结果表明:该算法具有收敛速度快、解的精度高和避免出现早熟等优点,可用于大地电磁资料解释.

Application of real coded genetic algorithm in two-dimensional magnetotelluric inversion

Genetic algorithm, one of the new methods for global non-linear optimization, has been applied to magnetotelluric sounding (MT) data analysis. But the magnetotelluric inverse problem is ill-posed, and therefore unstable and non-unique. In this paper, the inverse problem of two-dimensional magnetotelluric has been solved by using a real coded genetic algorithm that employing Tikhonov regularization method. The approach is based on regularization theory and genetic algorithm is utilized for searching for the minimum of the parametric functional. The normal optimum inversion methods based on local linearization are usually lost in local minimum values, and they seriously depend on the selection of the initial model. Compared with the traditional iterative inversion methods through linearization, the real coded GA is able to overcome disadvantages of the traditional inversion and obtain better results. The inversion results of magnetotelluric sounding synthetic models are ideal, which indicates that the algorithm possesses advantages of expediting convergence, avoiding earliness and improving precision. The real coded GA is highly adaptable and well suited to non-linear hypothesis testing as well as to inverse modeling. So it can be used in magnetotelluric sounding data analysis.