Geothermal model calibration using a global minimization algorithm based on finding saddle points and minima of the objective function

The objective function used when determining parameters in models for multiphase flow in porous media can have multiple local minima. The challenge is then to find the global minimum and also to determine the uniqueness of the optimized parameter values. A method for mapping out local minima to search for the global minimum by traversing regions of first order saddle points on the objective function surface is presented. This approach has been implemented with the iTOUGH2 software for estimation of models parameters. The methods applicability is illustrated here with two examples: a test problem mimicking a steady-state Darcy experiment and a simplified model of the Laugarnes geothermal area in Reykjavík, Iceland. A brief comparison with other global optimization techniques, in particular simulated annealing, differential evolution and harmony search algorithms is presented.