基于时空特征融合的剩余油快速预测模型研究OA

Substantial amounts of remaining oil exist within formations,awaiting further recovery during oilfield development.Accurate prediction of remaining oil distribution is of great significance for optimizing oilfield production and guiding subsequent development strategies.However,remaining oil distribution prediction faces challenges such as complex production history and numerous influencing factors,making it difficult for traditional prediction methods to meet the dual demands of accuracy and computational efficiency.To address these issues,this study innovatively proposed a rapid remaining oil prediction model that integrated spatiotemporal multi-scale features,aiming to efficiently and accurately predict the spatiotemporal distribution characteristics of remaining oil during reservoir development.The model was based on the convolutional long short-term memory neural network(ConvLSTM)framework and incorporated a spatiotemporal attention mechanism to achieve dynamic allocation of spatiotemporal feature weights,enhancing the model's ability to capture key spatiotemporal information.Meanwhile,the model integrated multi-scale convolutional networks,utilizing convolution kernels of different sizes to effectively extract multi-scale feature information,thereby improving the model's global and local capture capabilities for remaining oil distribution characteristics.Experimental results on the Egg reservoir model dataset demonstrate that the model can effectively predict the spatiotemporal dynamic changes of remaining oil.Comparison with traditional deep learning models also indicates that the model significantly outperforms traditional deep learning models in terms of root mean square error,mean absolute error,and mean absolute percentage error evaluation metrics,exhibiting stronger generalization capabilities in complex scenarios.Through analysis of activation intensity maps generated by the model,the study further validates the model's high attention capability to oil-water front positions,thereby significantly improving the accuracy of remaining oil distribution prediction.