Research on reservoir lithology prediction method based on convolutional recurrent neural network

Abstract

Considering that conventional reservoir prediction methods cannot fully explore the implicit relationship between seismic attributes and reservoir lithology, a deep learning lithology prediction model combining convolutional neural network and Long Short-Term Memory recurrent neural network is proposed to improve the classification prediction accuracy of reservoir lithology. This model is built and trained by seismic attribute data and lithology data of Shengli Oilfield to establish the nonlinear mapping relationships between seismic attributes and lithology labels. The experimental results show that the proposed method can significantly improve the effect of reservoir lithology prediction, whose prediction accuracy for complex reservoirs is close to 70%.