Principal component analysis for reservoir uncertainty reduction

Abstract

Reservoir monitoring considering all measurements and simulator outcomes available nowadays can become a complex task. The data integration and mainly the proper use of the big datasets is a challenge, especially in full field studies. This scenario of increasing data availability is an ongoing process due to new measurement technologies, high computational power and the reservoir characterization complexity. We propose to identify reservoir measurements that best represent the overall reservoir behavior using the Principal Component Analysis mathematical procedure. In addition, this procedure allows a reduction of the dataset dimension for a faster and more efficient reservoir analysis. Latin Hypercube sampling is used to sample the reservoir attribute range and the principal component of the measurements are integrated to identify the attribute interval that minimizes the simulation mismatch. The methodology is applied to a reservoir simulation model with 20 uncertainty attributes. Three study tests were performed using different percentiles in the likelihood distribution, which can conservatively or severely reduce the attribute ranges. The method achieved a coverage of approximately 95 % of the problem variability using five out of fifteen original principal components. Reservoir uncertainties were reduced and most of the simulated measurements had a significant history matching improvement