Resumen
The initial model plays an important role in seismic inversion. Generally, the initial model is constructed by lateral extrapolation of parameters under horizons constraints. However, without horizon data, initial modeling becomes a challenging task. Velocity spectrum is a 2D image that can reflect the characteristics of the formations. We regard the problem of establishing the initial model as the problem of similarity analysis of seismic lateral characteristics and propose a method of establishing the initial inversion model based on velocity spectrum and Siamese network. Firstly, the lateral variation of formation characteristics is tracked on velocity spectra generated by common depth point (CDP) gathers. Then, the target tracking results at different CDP positions are obtained with the triple Siamese network. Finally, the discrete inversion parameters are extrapolated along the tracking paths to obtain the initial inversion model. The Siamese network can quickly obtain the similarity of 2D images and does not need manual labels. The theoretical and practical results show that our method can efficiently generate the initial model that conforms to the seismic structure and stratigraphic characteristics without the constraint of interpreted horizon data.