Parameter estimations from gravity and magnetic anomalies due to deep-seated faults: differential evolution versus particle swarm optimization

Abstract

Estimation of causative source parameters is an essential tool in exploration geophysics and is frequently applied usingpotential field datasets. Naturally inspired metaheuristic optimization algorithms based on some stochastic procedures have attractedmore attention during the last decade due to their capability in finding the optimal solution of the model parameters from the parameterspace via direct search routines. In this study, the solutions obtained through differential evolution algorithm, a rarely used metaheuristicalgorithm in geophysics, and particle swarm optimization, which is one of the most used global optimization algorithms in geophysics,have been compared in terms of robustness, consistency, computational cost, and convergence rate for the first time. Applications havebeen performed using both synthetic and real gravity and magnetic anomalies due to deep-seated fault structures. Before the parameterestimation studies, resolvability of the fault parameters have been examined by producing cost function/error energy topography mapsto understand the suitability of the problem and also the mathematical nature of the inversion procedure. Optimum control parametersof both algorithms have also been determined via some parameter tuning studies performed on synthetic anomalies. Consequently,the tuned parameters clearly improved the effectiveness of both metaheuristics on the solution of the optimization problems underconsideration. Moreover, reliabilities of the obtained solutions and also the possible uncertainties have been investigated using probabilitydensity function analyses. Real data applications have been performed using a residual gravity anomaly observed over the Graber oil field(Oklahoma, USA) and an airborne total field magnetic anomaly observed over the Perth Basin (Australia). Applications have shown thatalthough both algorithms provided close results in both synthetic and real data experiments, the differential evolution algorithm yieldedslightly better solutions in terms of robustness, consistency, computational cost, and convergence rate. Thus, the differential evolutionalgorithm is worth paying more attention to and is suggested as a powerful alternative to particle swarm optimization for the inversionof potential field anomalies.