A New Approach to Model Parameter Determination of Self-Potential Data using Memory-based Hybrid Dragonfly Algorithm

Abstract

A new approach based on global optimization technique is applied to invert Self-Potential (SP) data which is a highly nonlinear inversion problem. This technique is called Memory-based Hybrid Dragonfly Algorithm (MHDA). This algorithm is proposed to balance out the high exploration behavior of Dragonfly Algorithm (DA), which causes a low convergence rate and often leads to the local optimum solution. MHDA was developed by adding internal memory and iterative level hybridization into DA which successfully balanced the exploration and exploitation behaviors of DA. In order to assess the performance of MHDA, it is firstly implemented to invert the single and multiple noises contaminated in synthetic SP data, which were caused by several simple geometries of buried anomalies: sphere and inclined sheet. MHDA is subsequently implemented to invert the field SP data for several cases: buried metallic drum, landslide, and Lumpur Sidoarjo (LUSI) embankment anomalies. As a stochastic method, MHDA is able to provide Posterior Distribution Model (PDM), which contains possible solutions of the SP data inversion. PDM is obtained from the exploration behavior of MHDA. All accepted models as PDM have a lower misfit value than the specified tolerance value of the objective function in the inversion process. In this research, solutions of the synthetic and field SP data inversions are estimated by the median value of PDM. Furthermore, the uncertainty value of obtained solutions can be estimated by the standard deviation value of PDM. The inversion results of synthetic and field SP data show that MHDA is able to estimate the solutions and the uncertainty values of solutions well. It indicates that MHDA is a good and an innovative technique to be implemented in solving the SP data inversion problem.

Combined Gravity or Self-Potential Anomaly Formula for Mineral Exploration

“…The interpretation of gravity and self-potential data falls on the main two categories as follows: the first category depends on threedimensional and two-dimensional data elucidation [8][9][10][11][12][13], the second category is depending using the simple geometric-shaped model such as spheres, cylinders, and sheets which are playing a vital role in interpreting the subsurface structures to reach the priors information that help in more investigations [14][15][16][17][18][19][20]. In addition, methods depend on the global optimization algorithms such as genetic algorithm [21][22][23][24], particle swarm [25,26], simulated annealing [27][28][29][30][31][32], flower pollination [33], memory-based hybrid dragonfly [34], differential evolution [35,36].…”