Stability-Improved Prestack Seismic Inversion Based on Orthogonal Learning Hybrid Particle Swarm Optimisation

Abstract

Summary Prestack seismic inversion quantitatively converts seismic data into multiple elastic properties, supporting the exploration of subsurface reservoirs in detail. Whereas, nonlinearity existed in the physical model causes the ill-posed inverse problem. To address the nonlinearity of the problem, global optimisation algorithm owns the advantages of achieving global optimal solution regardless of initial models. However, such algorithms may suffer from instabilities and premature convergence when applied to prestack inversion involving multiparameter and multimodal results, especially under complex geological condition. In this abstract, orthogonal learning hybrid particle swarm optimisation (OLHPSO) algorithm is proposed, and is incorporated into the prestack inversion based on Bayesian framework, with special intention of mitigating the instabilities and premature convergence. In particular, the stability of multiple elastic parameters is enhanced by the orthogonal learning scheme during model update/perturbation. In addition, the searching capability of the algorithm is improved by employing the probabilistic acceptance based on Metropolis criterion. Synthetic tests validate the stability and accuracy of the proposed method. Field data application demonstrates the method is capable of obtaining fine description of subsurface properties for better identifying potential reservoirs under complex geological condition.