1-D MAGNETOTELLURIC DATA INVERSION USING MODIFIED VERY FAST SIMULATED ANNEALING (MVFSA)

• Main Author: Heriyanto, Mohammad
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/35195
• Institution: Institut Teknologi Bandung

The early steps of natural resources exploration such as geothermal, groundwater, oil, gas and minerals were calculated using geophysical measurements. The magnetotelluric method measures the physical properties of resistivity subsurface by utilizing measurement data in electric and magnetic field. The field data obtained, then be directly interpretation or be inverted to getting subsurface resistivity models. Resistivity model obtained then associated with subsurface lithology, so we can know the location of the resource requested. Inversion of the magnetotelluric data into a subsurface resistivity model is not easy, because the data and models connected in non-linear. In this final project, we apply a modified very fast simulated annealing (MVFSA) method to inverse the magnetotelluric data into the subsurface resistivity model in one-dimensional (1-D). This method is global approach inversion and a modified version of very fast simulated annealing (VFSA) algorithm, commonly known in optimization discussion. The robust of MVFSA method tested by comparing with the Levenberg-Marquardt (LM), singular value decomposition (SVD), simulated annealing (SA), and VFSA, then statistically analyzed with a Gaussian probability density function (PDF). The result MVFSA is excellent method for modeling the 1-D resistivity structure of synthetic data, and be obtained rms error 2.97e-4% for 42 layers of earth model and 1.13e-5% for 3 layers, 0.147% for 3 layers with noise, and 0.074 % for 5 layers. The MVFSA method is faster convergent and iterations required is less than SA and VFSA. If the objective function is a quite complex, MVFSA is less possibility trapped in a local minimum compared with LM and SVD, so that the inversion produces a very good solution. MVFSA method was implemented on 5 point field data and produce inversion solution model with rms-error is relatively small (under 1%).