Particle Swarm Optimization Algorithm-Based Nonlinear Inversion Approach Research

摘要

A new 1D MT nonlinear inversion approach based on PSO (Particle Swarm Optimization) algorithm is proposed, to solve these problems of the low convergence speed and the low precision of the traditional nonlinear inversion methods. This approach discussed here has the advantage of less relying on the aim function information and easily carrying out. Moreover, both of the relationship of particles and behavior of swarm have been considered in the iterative process, which can improve the convergence speed and the precision. DLL (Dynamic Link Library) technique has been used for mixed language programming between C# and Fortran, the 1D MT inversion code in Fortran and the PSO algorithm and graphics interface code in C#. The new approach has been applied to the inversion with some theory models of A, Q, H, K and more complex ones and real data processing. Meanwhile, to get the convergence regions and influence factors for the case of 1D MT inversion based on PSO algorithm,some experiments are carried out in different groups of variable PSO parameters, different theory model parameters. Therefore, the criterion of parameter selection for real data processing can be gained through the statistics of experiment results. Experiments results showed that this method had a good performance in high convergence speed and accuracy.