Fast and Reliable History Matching of Channel Reservoirs using Ensemble Kalman Filter with CAE and PCA

Abstract

Summary Ensemble Kalman filter (EnKF) has been utilized to characterize reservoirs with high uncertainty. However, it requires a large number of models and long simulation time for stable and reliable results. Therefore, the authors propose a new history matching scheme using convolutional auto encoder (CAE) and principal component analysis (PCA). Our method firstly performs PCA for the latent codes of CAE for channel reservoir information. Then, it chooses of the 45 models among a total of 200 models near the representative model, which gives the most similar behaviors with the reference model. This process can minimize computation time in EnKF as well as increase prediction quality on reservoir performances by using the small but reliable models instead of the whole models or the randomly selected 45 models. By applying the proposed scheme to 2D channelized field with 72 by 72 grids, we can clearly see improved assimilation results while saving simulation time.