Deep Learning Surrogate Model Based on Residual Bottleneck Block for History Matching

History matching is a key link to predict the development performance of oil field by using reservoir numerical simulation, which directly determines the reliability of numerical simulation results. In practical application, the historical fitting process calls the numerical simulator for many times, which makes it have some problems such as heavy workload, long time consumption and expensive calculation of the numerical simulator. In order to solve the above problems, this paper constructs an RRBB-U-Net proxy model based on residual bottleneck block (RRBB) and cyclic coding-decoding network (R-U-Net) for dynamic reservoir prediction. In the aspect of parameterization of geological model, this paper combines CVAE with principal component analysis (PCA) to construct a CVAE-PCA parameterized model for the realization of two-dimensional complex perme-ability field. Finally, this paper combines RRBB-U-Net with CVAE-PCA to solve the problem of historical fitting, and combines particle swarm optimization to carry out historical fitting. The results show that the combination of the two algorithms can improve the efficiency of historical fitting and further promote the related research of reservoir historical matching.