Magnetotelluric Genetic Algorithm Inversion And Comparison Of The Inversion Methods

Magnetotelluric inversion method is multiple, different inversion method has its limitations in practical application, inversion results may fall into local optimum, and cause difficulties to explain personnel. Currently linear algorithm is widely used. Linear algorithm is an iterative algorithm based on the calculation of the gradient of the objective function to update the model, greater dependence on the initial model, high requirements to inversion data. Nonlinear genetic algorithm, it is a heuristic global random search method, do not need to calculate the gradient information of the objective function, and does not depend on the initial model, is based on multiple model search method, the requirements of inversion data is low, can make up for the shortcomings of linear inversion.The practical application of genetic algorithm in magnetotelluric is the relatively small. Magnetotelluric genetic inversion procedure is designed in this paper for magnetotelluric inversion characteristics. In the program, using the idea of linear algorithm, fixed inversion grid, resistivity is calculated based on the logarithmic domain, inversion calculations only change the resistivity models, thus reducing the inverse calculation parameters, improving the speed of inversion. Due to the characteristics of genetic algorithm, search range can't be infinite, based on the extremes of the inversion data or other inversion results, expand search range on the basis of extremum, so as to limit the search scope of genetic algorithm, the main search calculation is focused on the effective area to improve the search efficiency. Adaptation of the model is determined based on the variance of the forward model results or the original data, in select operations, directly out of the individual of the larger variance, to retain the individual of the small variance, the individual of minimum variance in the genetic is always preserved. Coding have used real-coded to avoid parameter mutation in crossover operation and mutation operation. Crossover mode is random multi-point crossover. The mutation operation design is different from the binary encoding, is on the basis of the original data to plus or minus a certain numerical value according to the different inversion data. In this paper, design two simple model trial Thesis has designed two simple models on genetic algorithm, the trial results show that genetic algorithm can better reflect the characteristics of the model. But in the inversion results there are insufficient.In many inversion method comparison results, the results of genetic algorithm in the structure of large effect is good, inversion results of Marquardt method, the layered structure has good performance, but prone to distortion, cause a lot of false anomaly, inversion calculation is also prone to failure. The results of Bostick transform are relatively stable, has fast calculation speed, and the request for quality of the data is low. The RRI inversion has fast calculation speed, but its results in the lack of smoothness. The inversion results of 2D NLCG and 2D OCCAM are greatly influenced by the initial model. Different data will show different inversion results. Different model of the initial 2D inversion contrast can be seen in homogeneous half space as a result of the initial model, can reflect the most essential structure model, the rough, the emergence of false anomaly is the smallest, and the practical data processing can't be ignored, also constrained inversion results is fine and improve the accuracy of the inversion, can achieve better retrieval results.