融合注意力机制的GCN-BiGRU剩余油预测方法

打开文本图片集

关键词：剩余油预测；图卷积神经网络；双向门控循环神经网络；克里金插值法中图分类号：TQ514；TE328 文献标志码：A 文章编号：1671-0460（2026）01-0128-06

GCN-BiGRUResidual Oil PredictionMethod Integrating Attention Mechanism

WANG Mei1，LOU Jinxiang1， GUO Junhui²，DONG Chi³ （1.SchoolofComputerandInformation TechnologyNortheast Petroleum University，Daqing Heilongjiang 163318，China; 2.Exploration and Development Research Institute of Daqing Oilfield Co.，Ltd.，Daqing Heilongjiang 163318， China; 3.Northeast Petroleum University Sanya Offshore Oil and Gas Research Institute，Sanya Hainan 572024， China）

Abstract：The influence factors ofresidual oil distributionarecomplex，the injection-production wellis influenced by the development historyof the welland the injection-production well around it. In order to solve these problems，an adaptive GCN-BiGRU residual oil prediction model basedon atention mechanism was proposed in this paper， adaptive Graph Convolutional Networks （GCN）module was usedto extract spatial dependency relationships between each njection-production well and surrounding injection-production wels in each layer，bidirectional Gated Recurrent Unit （BiGRU） neural network integrating the attention mechanismcanbeterlearn the timing dependenceof injectionproduction wels.The experimental resultsshowed that the performance of the model was significantly improved compared with CNN-LSTM， GCN-LSTMand CNN-GRU.Using this model，the predicted watersaturation ofeach wel point in each layer can be obtained，andcombined with Kriging interpolation method，the watersaturation fieldof each layer can be obtained， which can effectively predict the favorable areas of remaining oil.

Key Words：Residual oil prediction; Graph convolutional neural network; Bidirectional gated cyclic neural network; Kriging interpolation method

剩余油是指经历各种方法的生产开发之后，保留在不同地质环境油藏中的原油[1]。（剩余8193字）