Testing Different Stochastic Methods to Model Direct Current Resistivity and Seismic Refraction Geophysical Data Using a Combined Local and Global Optimization Algorithm

Global optimization methods have recently become an essential option in the processing and interpretation of geophysical data sets. In this study, we compared the optimization capability of five global optimization techniques in terms of their ability to localize a global solution and computation cost using objective functions for direct current resistivity (DCR), and seismic refraction (SR) data processing and interpretation. The genetic algorithm (GA), particle swarm optimization (PSO), simulated annealing (SA), surrogate optimization (SO), and pattern search (PS) optimization techniques were tested. The recently developed combined (global and local) optimization algorithm that uses the local optimization’s output to minimize the search space for the global search method and achieve faster convergence is applied to solve the optimization problem for comparison. Because all the optimization algorithms try to minimize the same objective function, we benchmarked the comparison with the result of the local optimization algorithm using the Gauss–Newton (GN) method. The combined optimization algorithms run for 500 iterations for both synthetic and real data. All the optimization algorithms satisfactorily reconstructed the positive anomalies (synthetic data) and defined a suspected fault (real data). However, considering the result of the local optimization as a benchmark (in terms of misfit and run time), only the GA and PSO satisfactorily reduce the misfit obtained from both DCR and SR inversion using the GN method. Therefore, the GA and PSO should be considered as the proper techniques for inverting the real DCR and SR data sets, with sequential use of local and global optimization methods.