Establishment of sample pool and prediction of saturation field in complex fault block reservoir based on deep learning algorithm

Abstract

The remaining oil prediction of complex fault block reservoir is the key to guide the well deployment in the late high water cut stage. We formed the construction methods of 20,000 numerical simulation forward models, including the dynamic and static attributes of the reservoir, such as structural depth, reservoir thickness, permeability, porosity, injection and production fluid volume, etc. The saturation field distribution of each model is obtained through numerical simulation, so as to form the saturation field characteristic data sample database. Through the deep convolutional generative adversative neural network model, the data of sample pool was trained, in which 70% of the data was randomly selected as the training set and 30% of the data as the test set. Finally, the saturation field prediction method that could be applied to an actual block was established. The result showed that: the new method did not need to carry out the numerical simulation research on the actual region, the remaining oil saturation distribution at different time step that obtained by the deep convolutional generative adversative neural network model only need to input the reservoir physical parameters, well location coordinates, injection production and other information of the actual block, and the prediction accuracy of the method was more than 90%. The test showed that the deep convolutional generative adversative neural network model had good generalization ability, and the model could be widely used in the prediction of residual oil saturation field of complex fault block reservoirs, which greatly improved the efficiency of reservoir research.