Application of particle swarm optimization on self-potential data

Particle swarm optimization (PSO) is a global search method, which can be used for quantitative interpretation of self-potential data in geophysics. At the result of this process, parameters of a source model, e.g., the electrical dipole moment, the depth of the source, the distance from the origin, the polarization angle and regional coefficients are estimated. This study investigates the results and interpretation of a detailed numerical data of some simple body responses, contaminated and field data. The method is applied to three field examples from Turkey and the results are compared with the previous works. The statistics of particle swarm optimization and the corresponding model parameters are analyzed with respect to the number of generation. We also present the oscillations of the model parameters at the vicinity of the low misfit area. Further, we show how the model parameters and absolute frequencies are related to the total number of PSO iterations. Gaussian noise shifts the low misfit area region from the correct parameter values proportional to the level of errors, which directly affects the result of the PSO method. These effects also give some ambiguity of the model parameters. However, the statistical analyses help to decrease these ambiguities in order to find the correct values. Thus, the findings suggest that PSO can be used for quantitative interpretation of self-potential data.

Particle swarm optimization (PSO) has been investigated on self-potential data. PSO has been applied to some synthetic, noise added, and field data. A model response of a cylinder and its fitness can be seen in Fig. 1a–b. Fig. 2 shows the low misfit regions. Oscillations of the model parameters are displayed in Fig. 3. Fig. 4 illustrates frequency distribution of the cylinder model for all parameters.Figure optionsDownload as PowerPoint slideFigure optionsDownload as PowerPoint slideHighlights
► Particle swarm optimization.
► Effect of errors and low misfit region.
► Frequency distribution of model parameters.
► PSO application on SP data.