Due to the influence of long-term waterflooding, the reservoir physical properties and percolation characteristics tend to change greatly in offshore unconsolidated sandstone reservoirs at ultrahigh water-cut stage, which can affect the remaining oil distribution. Remaining oil characterization and proper development strategy-making are of vital importance to achieve high-efficiency development of mature reservoirs. The present numerical simulation method is difficult to apply in reservoir development due to the problems of noncontinuous characterization and low computational efficiency. Based on the extended function of commercial numerical simulator, the time-varying equivalent numerical simulation method of reservoir physical properties was established, and the research of numerical simulation of X offshore oilfield with 350,000 effective grids was completed. The results show that the time-varying reservoir properties have a significant impact on the distribution of remaining oil in ultrahigh water-cut reservoir. Compared with the conventional numerical simulation, the remaining oil at the top of main thick reservoir in X oilfield has increased by 18.5% and the remaining oil in the low-permeability zone at the edge of the nonmain reservoir has increased by 27.3%. The data of coring well and the implementation effect of measures in the X oilfield are consistent with the recognition of numerical simulation, which proves the rationality of numerical simulation results. The new method is based on a mature commercial numerical simulator, which is easy to operate and has reliable results.

Summary History matching is a crucial process that enables the calibration of uncertain parameters of the numerical model to obtain an acceptable match between simulated and observed historical data. However, the implementation of the history-matching algorithm is usually based on iteration, which is a computationally expensive process due to the numerous runs of the simulation. To address this challenge, we propose a surrogate model for simulation based on an autoregressive model combined with a convolutional gated recurrent unit (ConvGRU). The proposed ConvGRU-based autoregressive neural network (ConvGRU-AR-Net) can accurately predict state maps (such as saturation maps) based on spatial and vector data (such as permeability and relative permeability, respectively) in an end-to-end fashion. Furthermore, history matching must be performed multiple times throughout the production cycle of the reservoir to fit the most recent production observations, making continual learning crucial. To enable the surrogate model to quickly learn recent data by transferring experience from previous tasks, an ensemble-based continual learning strategy is used. Together with the proposed neural network–based surrogate model, the randomized maximum likelihood (RML) is used to calibrate uncertain parameters. The proposed method is evaluated using 2D and 3D reservoir models. For both cases, the surrogate inversion framework successfully achieves a reasonable posterior distribution of reservoir parameters and provides a reliable assessment of the reservoir’s behaviors.

No abstract

No abstract