Abstract
History matching can provide reliable numerical models for reservoir management and development by assimilating the historical production data into prior geological realizations. It is usually a typical inverse problem with multiple solutions. However, efficiently obtaining multiple posterior solutions is still challenging for most existing history matching algorithms. In this paper, we present a novel algorithm to tackle this problem, which integrates the niching technique and surrogate-assisted method into the particle swarm optimization (PSO), in which, the niching technique can improve the exploration ability and maintain the diversity of the population, while the surrogate-assisted method is focused on accelerating the convergence. Additionally, the convolutional variational autoencoder (CVAE), a deep learning model, is adopted to map the high-dimensional spatially uncertain parameters such as permeability and porosity to low-dimensional latent variables. Experimental results show that the proposed algorithm has good convergence and sampling ability for history matching problems.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have