Reservoir Parameter Prediction Technology Based on Deep Learning and Its Application in the Panyu 4 Sag, Pearl River Mouth Bain

Li Yingwei ^* Zhu Yanhui Li Zhenshen Zhang Xiaozhao Cai Guofu

• Shenzhen Branch of CNOOC Ltd., Shenzhen, China

The continental deep strata in the Panyu 4 Sag of the Pearl River Mouth Basin in the South China Sea are characterized by complex lithology and tight sandstone reservoirs with low porosity and low permeability. Predicting porosity and lithology in this area has long been a challenge in seismic reservoir prediction. Traditional methods, which rely on linear mapping based on well data or probabilistic mapping through multi-attribute fusion, struggle to capture the complex nonlinear relationships between reservoir parameters and seismic attributes. To address this issue, this paper proposes a method using a convolutional neural network (CNN) for predicting porosity and facies distribution. Based on rock physics analysis and pre-stack elastic impedance inversion data, this approach first takes the effective porosity and shale content(VCL) from well-log interpretation as training targets. It then constructs training samples by simulating different lithologies and extracting the corresponding elastic parameters from well-log data. Through optimal evaluation, the model parameters of the deep learning network are determined, and a nonlinear mapping relationship between elastic parameters and reservoir parameters, such as porosity, is established. Finally, the trained deep learning model is applied to the elastic parameter bodies to obtain predictions of effective porosity and VCL, thereby achieving a quantitative characterization of high-quality deep sandstone reservoirs. The application of this method in the deltaic sediments of the Panyu 4 Sag in the Pearl River Mouth Basin shows that the deep learning-based predictions of facies distribution and porosity are consistent with well data and geological understanding. The fractured well, designed on the basis of the prediction results, achieved significant productivity enhancement following drilling, thereby demonstrating the efficacy of this method as a reservoir parameter prediction tool.