Efficient history matching with dimensionality reduction methods for reservoir simulations

Abstract

Oil reservoir history matching is a well-known inverse problem for predicting production by optimizing enormous unknown parameters with numerical simulation. Typically it can be formulated in a Bayesian framework with geological priors. Instead of gradient-based optimization with the possibility of converging to a local minimum, evolutionary algorithms have been introduced to globally find optimal parameters. Due to the high-dimensional parameters, the optimization could become inefficient; therefore, many dimensionality reduction algorithms have been applied in history matching. However, these methods suffer from the linear assumption or the pre-image problem, which could affect the model optimization. In this paper, based on the evolutionary algorithm termed Multi-objective Evolutionary Algorithm Based on Decomposition, which is capable of simultaneously optimizing the parameters with respect to the data of several oil wells, we propose history matching with dimensionality reduction by explicitly utilizing the nonlinear dimensionality reduction model Auto-Encoder to reduce the number of unknown parameters, which can naturally handle the pre-image problem and then improve model performance in terms of precision and complexity. Experimental results based on PUNQ-S3 data verify the efficiency of the newly proposed methods.

Using Machine Learning to Reduce Ensembles of Geological Models for Oil and Gas Exploration

“…The triplet of alleles can be also be considered as a dictionary, where the first property, in this case saturation (sw), is defined by gene number 5, the second property, net-to-gross (ntg), is defined by gene number 7, and the third property, porosity (phi), is defined by gene number 13. Therefore, given In [3], the authors used a technique called oil reservoir history matching (HM) for estimating oil reservoir models parameters and making production forecasts. Whilst this is a different problem area from the one we are concerned with here, focused on existing oil reservoirs rather than exploration for new reservoirs, they had similar issues in terms of the size of the data.…”

“…In [3], the authors used a technique called oil reservoir history matching (HM) for estimating oil reservoir models parameters and making production forecasts. Whilst this is a different problem area from the one we are concerned with here, focused on existing oil reservoirs rather than exploration for new reservoirs, they had similar issues in terms of the size of the data.…”

Gaussian Processes Proxy Model with Latent Variable Models and Variogram-Based Sensitivity Analysis for Assisted History Matching

“…Additionally, the solutions are prone to be the local optimum to the initialization. To address the issues, global optimization techniques based on Evolutionary Algorithms (EAs) had been introduced for AHM such as genetic algorithms, differential evolution and particle swarm optimization, because of their easy application to various problems without special assumptions [15][16][17]. In the framework of EAs, the objective function can be formulated as a single-objective function or as a multi-objective function, which could be effectively optimized even when there are discrete reservoir parameters, highly non-Gaussian distributed data, or non-differentiable objective functions.…”