Abstract
We present a novel multi-objective optimization algorithm, Archived Multi-Objective Simulated Annealing (AMOSA), based on simulated annealing for transient electromagnetic (TEM) one-dimensional inversion. The data misfit and model-constraint are treated as two objectives rather than being assembled into a single objective function in this method to reduce uncertainties. We obtain a set of satisfactory solutions rather than a single 'global optimum' with Multi-Objective Optimization (MOO). Archive and Pareto domination are used for discontinuous fronts and to determine the acceptance of a new model with Pareto-optimal solutions. Temperature based on the Quasi-Cauchy distribution instead of the Gibbs distribution is selected to accelerate the inversion and stabilize model perturbation. We test the method using several 1D layered-earth models with noise and noise-free data. All synthetic model inversion results are in good agreement with true models. Finally, we test the method using a coincident loop TEM field data. The inverted profile shows a reasonable three layers of subsurface geology. A nearby well water table verifies the interpreted aquifer layer and the estimated aquifer's top surface.