Data driven analysis in oil and gas operations - Datadrevne analysemetoder i olje- og gassproduksjon

Abstract

Knowledge about the production system and models of relevant parts of the production network can improve the decision-making process in offshore oil and gas production. This thesis investigates how multivariate projection methods may be used to analyze historical production data for monitoring and production optimization purposes. Two multivariate projection methods, principal component analysis and partial least squares regression, are used to analyze a data set from the process historian of an offshore production system on the Norwegian continental shelf. The results presented in this thesis show that the methods may assist production optimization in several ways. Both methods are robust with respect to correlated variables, noise and missing values, and the methods are well suited for exploratory analysis of historical production data. Furthermore, principal component analysis may be used as a monitoring tool for detecting abnormal operating conditions, and partial least squares regression may be used to predict individual flow rates from the wells and the total flow rates from the platform with choke measurements and pressure measurement as explanatory variables. Examples of relevant applications are illustrated and discussed.