A digital approach to the management of brownfields

Abstract

This thesis investigates analytic and data-mining methods that can be used for the management of petroleum brownfields, specifically as it applies to the surveillance, analysis, & optimization of gas lifted oil wells. Building on the output of validated physics-based models, this thesis investigates a range of analytic methods which may be used to determine a probable depth of gas lift injection of wells without pressure gauges, and finds that the Random Forest method coupled with a k-means clustering algorithm can offer good results. Additionally, this thesis shows how a pan matrix profile may be used to efficiently identify patterns (motifs) in the real time pressure signatures of wells. Understanding of the motifs are assessed through a physics-based model, providing a useful tool for engineers to perform surveillance of large well count areas, which are typical for brownfields.