GRAVITY INVERSION DEVELOPMENT MODELING METHOD USING NATURE-INSPIRED BAT ALGORITHM

• Main Author: Prasetyo, Nugroho
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76485
• Institution: Institut Teknologi Bandung

The deterministic inversion modeling technique commonly utilized in gravity inversion exhibits limitations in achieving an accurate initial model to represent subsurface density distribution aligned with field geological conditions. Stochastic inversion methods are often employed to mitigate ambiguity or local optima issues. Ambiguity arises due to the underdetermined nature of gravity data modeling, where the sought model parameters outnumber the measured data parameters. The Bat Algorithm (BA) method, a meta-heuristic approach, is applied to stochastic inversion to address this challenge. This technique emulates the hunting behavior of microbats in a population, exploring the model space randomly and preventing model solutions from getting trapped in local optima. This study aims to establish a novel inversion approach to model gravity responses within a 3D framework. The density model employed adopts a 3D and 2.5D grid arrangement in the form of rectangular prisms. The implementation of BA in the gravity parameter search encompasses geometry and density, perturbing the model by manipulating the 3D array indices to represent density grid modifications. The development of BA, through modified individual velocity components, influences perturbation of the best solution in each iteration. Testing of the modified BA is conducted using simple synthetic models, including single, double, and triple density bodies with varying distances between bodies. Geological models encompass intrusion, anticline folding, basin, and normal fault with varied fault zone dip angles. The success rate parameter is determined by calculating the percentage match between the inversion model grids and reference models. Test results demonstrate a 15% enhancement in success rates. The inversion outcome models also exhibit accurate identification of rock body anomaly sources, with a success rate exceeding 85% for both singlelayer and multi-layer array models within the 3D configuration, featuring a centrally positioned single-density body amid the mesh grid. Anomaly source locations for rock density anomalies are effectively identified in the 2.5D models, particularly for simple geological structures like intrusion, folding, and inclined layers. However, limitations emerge during the identification of 3D density model distribution and depth variations in the 2.5D models, such as in the case of normal faults and inclined layers. The 2.5D model testing yields a success rate exceeding 75% for all synthetic geological models. The algorithm's potential development in gravity inversion is promising based on the achieved outcomes. Future developmentv involves the utilization of irregular polygon prism geometries, aiming to attain more optimal results