Model-data-driven AVO inversion method based on multiple objective functions

The model-driven inversion method and data-driven prediction method are effective to obtain velocity and density from seismic data. The former necessitates initial models and cannot provide high-resolution inverted parameters because it primarily employs medium-frequency information from seismic data. The latter can predict parameters with high resolution, but it require a significant number of accurate training samples, which are typically in limited supply. To solve the problems mentioned for these two methods, we propose a model-data-driven AVO inversion method based on multiple objective functions. The proposed method implements network training, network optimization, and network inversion by using three independent objective functions. Tests on synthetic and field data show that the proposed method can invert high-accuracy and high-resolution velocity and density with a few training samples.