Attenuation-based deep learning for volcanic reservoirs characterization: A case study from Junggar Basin, China Available to Purchase

Seismic characterization of volcanic reservoirs is a difficult task considering the low quality of seismic data. The reflection of volcanic rocks is always characterized by high amplitude and relatively low frequency compared with reflections of clastic sediments which makes the identification of oil-bearing volcanic reservoirs more difficult. In order to depict volcanic reservoirs in Junggar Basin, we combine time frequency analysis and deep learning to predict the Carboniferous volcanic reservoirs. We first use time frequency analysis to determine the features of volcanic reservoirs’ reflection and the specific frequency for characterizing volcanic reservoirs. Second, we use low-, middle-, and high-frequency component and its corresponding attributes as Convolutional Neural Networks’ input to predict the impedance of volcanic rocks. The results are in line with the drilling data, and this method would give insights to characterize volcanic reservoirs in Junggar Basin.