Support Vector Machine (SVM) based prestack AVO inversion and its applications

• SVM popularly used in statistical learning has been used to solve AVO inversion problems successfully.
• We quantitatively compare the inversion results with the traditional Bayesian method.
• We analyze the difference behind them and give our comprehension.
• The resolution of the density term has large improvement using the novel method.

AVO inversion can be used to estimate P-wave velocity, S-wave velocity, and density perturbations from reflection seismic data. The inversion of the density term, however, due to its little sensitivity to amplitudes and the paucity of large angle incident information, is usually difficult and unstable. The conventional method of linearized approximation is usually not accurate enough and tends to be affected by the background information, while the accurate method is more likely to be trapped in a local minimum and more computationally intensive. This paper delineates a novel method of AVO inversion based on Support Vector Machine (SVM). First, we describe the basic principle of SVM, and then we investigate an SVM procedure for the three-term AVO inversion problem. To demonstrate its performance, we compare it with the conventional Bayesian method. From the inversion results of both the synthetic and real data, we conclude that the algorithm of SVM leads to high-resolution P-wave velocity, S-wave velocity, and density perturbation, moreover, the resolution of the density term has large improvement, compared to the Bayesian method. They all demonstrate the feasibility and application of SVM on both synthetic and real data.