Productivity "sweet spot" prediction method combining intelligent estimation and conventional seismic attributes

The productivity prediction of tight sand is critical for oil and gas exploration in China.Owing to the complex structure and limited number of vertical wells in the study area of the Ordos Basin, it is difficult to characterize "sweet spots" of productivity using single traditional prediction methods.We proposed a method that combines seismic attributes and deep learning results to calculate "sweet spot" productivity using seismic data.In this study, a convolutional neural network was used to characterize the sedimentary microfacies of sandy debris flows in the study area by combining geological, logging, and seismic data to obtain the dominant facies distribution.Considering the impact of the local structure on the study area, the three-dimensional distribution of productivity "sweet spots" was obtained by integrating the curvature properties and sandy debris flow microfacies.This method was used to extract and optimize productivity-sensitive attributes from the seismic data.Simultaneously, deep learning was used to predict the reservoir properties and sedimentary facies from seismic data.Finally, the productivity "sweet spot" distribution was obtained by integrating the sensitive attributes and projected sedimentary microfacies.The application results showed that the productivity "sweet spot" distribution obtained by the above fusion was significantly improved and correlated well with the production data.The "sweet spot" area was consistent with the drilled horizontal well results.These results provide a reference for the future deployment of high-yield well locations and more effective exploitation of tight sand reservoirs.