| Abstract Submission No. | ABS-4.1-0085 |
| Title of Abstract | Gravity Inversion of Basement Relief Using Particle Swarm Optimisation with Physics-Informed Gravity Response Calculation, Data-Driven Fourier Coefficient Selection, and Adaptive Parameter Tuning |
| Authors | Jainendra Tripathy* |
| Organisation | Indian Institute of Technology (Indian School of Mines) Dhanbad |
| Address | Near Chakeri Railway Station, Gully No. 1, Plot No. 12 Kanpur, UP, India Pincode: 208008 E-mail: jainendratripathy@gmail.com |
| Country | India |
| Presentation | Poster |
| Abstract | Gravity inversion is a central technique in applied geophysics for estimating basement relief in heterogeneous sedimentary basins, with applications in hydrocarbon, groundwater, and mineral exploration. Conventional approaches combine Particle Swarm Optimization (PSO) with Fourier series to reduce model dimensionality. However, these methods suffer from resource-intensive gravity response calculations, fixed PSO parameters, and heuristic Fourier coefficient selection, which often limit performance in noisy or geologically complex settings. To address these challenges, we propose an enhanced PSO-based inversion framework with three innovations. First, the iterative calculation of gravity anomalies from PSO-proposed basement relief models is accelerated using a physics-consistent neural network. Second, PSO control parameters are tuned adaptively during the optimization process by a reinforcement learning agent, improving convergence stability and reducing the risk of trapping in local minima. Third, the number of Fourier coefficients is estimated automatically from observed gravity data using a one-dimensional Convolutional Neural Network, offering a robust, data-driven alternative to heuristic selection. The framework is validated on synthetic datasets contaminated with white Gaussian noise and on field gravity profiles from the Godavari Basin, India, and the Sayula Basin, Mexico. Performance is assessed in terms of computational efficiency, inversion accuracy, robustness to noise, and uncertainty quantification. Results show marked improvements over conventional PSO-Fourier methods, establishing a fully automated, scalable, and reliable approach for basement relief inversion. More broadly, this methodology provides a modular blueprint for integrating Artificial intelligence with physics-based models across diverse geophysical inverse problems. |
| Keywords | Gravity Inversion, Basement Relief, Particle Swarm Optimization (PSO), Physics-Informed Neural Networks (PINN), Reinforcement Learning (RL), Convolutional Neural Networks (CNN), Fourier Series, Adaptive Parameter Tuning, Geophysical Inverse Problems, Sedimentary Basins. |