Exploring the utility of nonlinear hybrid optimization algorithms in seismic inversion: A comparative analysis

Abstract

The present study integrates various local and global optimization techniques together to estimate subsurface properties from post-stack seismic data and compare their efficacy qualitatively and quantitatively. Specifically, a local gradient-based optimization method, the quasi-newton method (QNM), is combined with global techniques such as simulated annealing (SA), genetic algorithms (GA), and particle swarm optimization (PSO). These are well-established methods in geophysics. The research compares three global optimization methods (SA, GA, and PSO), their hybrid variants, and QNM for estimating subsurface acoustic impedance. The goal is to assess the trade-offs between solution accuracy and convergence efficiency, offering insights into the strengths and weaknesses of each approach. The objective is to guide the selection of the most effective optimization technique for seismic inversion, balancing quality and computational performance. Both synthetic and real seismic datasets are used to validate the proposed methodology, demonstrating its robust performance across various geological scenarios. Comparative analyses with single global inversion approaches reveal that hybrid optimization methods offer greater accuracy and reliability, positioning them as versatile tools for subsurface characterization. The results indicate that while the hybrid PSO method does not provide significant improvements over single PSO, it extends the convergence time. On the other hand, SA and GA perform adequately, but their hybrid versions considerably enhance solution quality at the cost of longer convergence times. Among the methods, SA shows the fastest convergence to the global solution, followed by GA and PSO. Hybrid SA stands out, delivering superior resolution and faster convergence compared to hybrid PSO and GA.