Multi-parameter neural network deep learning with a novel amplitude-versus-offset attribute for hydrocarbon detection in the X gas field of the Xihu Sag

Abstract Traditional hydrocarbon detection methods cannot effectively identify the reservoir fluid properties in the X gas field of the Xihu Sag in the East China Sea Shelf Basin, where the number of exploration wells is limited and the amplitude variation with offset (AVO) data are scarce. Therefore, an effective hydrocarbon detection approach must be specifically tailored to this area. A fluid-identification intercept-gradient crossplot template of the study area was created through pre-stack AVO forward modeling at a single well drilled in the X gas field. As gas-bearing sandstone exhibits distinct Class II to III AVO characteristics, it can be differentiated at varying burial depths from the AVO intercept and gradient crossplots. Based on these findings and the Poisson impedance concept, this paper derives the scalar Poisson’s ratio to enhance the gas-identification sensitivity in the study area. The proposed method has successfully achieved detecting hydrocarbons at intermediate and deep layers within the Xihu Depression. Furthermore, the paper explores multi-parameter neural network deep-learning techniques to minimize the uncertainties in hydrocarbon detection. The sensitivity of gas identification in the study area can be further improved by the attributes that are predicted by leveraging a convolution neural network, achieving promising hydrocarbon detection outcomes in medium-deep layers of the Xihu Sag.