Two-Dimensional Magnetotelluric Inversion Using Particle Swarm Optimization Algorithm

Magnetotelluric (MT) is a geophysics method which uses the nature alternating electromagnetic field for the exploration of electrical structure underground. The method has many advantages. It can explore very deep. It is not shielded by the high resistance layer and sensitive to the low resistance layer. It has an extend application to the fields such as the geosciences of depth, the exploration of oil and gas, and the exploration of mineral and so on. As an important part of MT, MT inversion has been widely researched over the world.The MT sounding data inversion is using some kinds of numerical calculation methods to calculate the apparent resistivity and phase which change with the frequency to get the resistivities underground on different deepth. We also call it quantitative interpretation. First, we make a tentative electricity model based on the geology data, geophysic detection data or the drill data. And then calculate it's apparent resistivity by the forward modeling method. Compare the difference between the observed resistivity and calculated resistivity, and then revise the electricity model until the difference is small enough to what we want. And that final electricity model will be the inversion result.Particle swarm optimization algorithm (PSO) is a global optimization strategy that simulates the social behavior observed in a flock (swarm) of birds searching for food. It has many thing in common with genetic algorithm, they are all swarm intelligence algorithms, PSO is easy to realize and it's a swarm intelligence method, so it adapt not only to the science research, but also the engineering application. Nowadays, PSO has been widely used in electricity system optimization, TSP problem, neural network training, digital circuit's optimization, function optimization and so on.In this paper, I try to use PSO algorithm in two-dimensional magnetotelluric inversion, and try to work out the feasibility of this inversion, the quality of inversion results and its efficiency. In the first chapter, there are research actualities of MT data inversion and PSO algorithm.In chapter 2, it shows the basic knowledge of PSO algorithm, and then introduces a new improvement about PSO algorithm based on the research results people have already done. At the end of this chapter, there is a test case of PSO with this new improvement.Chapter 3 is the basic knowledge of 2-D MT and the forward method of finite element (FE) with rectangle element. And then shows the forward results of some typical 2-D models.And then the most important chapter, chapter 4 is the research result of this paper. The first part of this chapter is 1-D MT inversion theory and one inversion case. The second part is 2-D MT inversion theory and some details of PSO algorithm. Then inversion cases of three typical models.The Mesh size of a model is 34×28, but the inverse mesh size is 24×14, because the air elements and the extends area have no need to be inversed, so the total elements that need to be inversed are 336 elements. In this paper, the step number of al12-D MT inversions is 50. But the simple 2-D model inversion has 20 particles, and cost about 2.8 hours; the complicated 2-D model inversion has 40 particles arid cost about 5.7hours. The tool that we use to programme is MATLAB. The computer's CPU that we used to calculate is Genuine Intel CPU T2130, 1.86GHz&1.86GHz, its memory size is 1G.The last chapter is conclusions and suggestions. Based on all the things that we have done in this paper, we can conclude that:(1) The PSO algorithm is simple, easy to realize; is a fast and effective nonlinear inverse algorithm. After the testing and inversion case of one-dimensional MT data. We proved that PSO algorithm is better than Monte Carlo method and simulated annealing method.(2) We put forward a new strategy toω, named damped PSO algorithm. This improvement is better than the basic PSO algorithm through the results of the test and 1-D MT inverse case.(3) We applied the new damped PSO algorithm in 1-D MT data inversion, through simulated inversion of the theoretical model; we find that PSO algorithm has better optimization ability and antinoise ability.(4) Through the application of simple 2-D MT model inversion, the inverse results show that PSO algorithm can find the general location and boundary of the abnormity.(5). For the complicated 2-D geoelectric model inversion, PSO algorithm can find the abnormities'center position, but can hardly find the real boundary of the abnormities; there are some errors between the inversion result and the theoretical model. The algorithm slows down into the local minima, so we still need to improve the algorithm.As an algorithm of swarm intelligence, PSO algorithm is slower than linear algorithms in calculation, but it has more potential to be a better algorithm for inversion problems. In this paper, we just improve PSO can be a good way to inversion, and don't compare with OCCAM, RRI and REBOCC. It is just a basement for the next research.