Estimation of Formation Water rate using Unscented Kalman filtering with application to the Snøhvit Gas/Condensate Field.

• Main Author: Mørk, Anders Jordtveit
• Other Authors: Imsland, Lars, Norges teknisk-naturvitenskapelige universitet, Fakultet for informasjonsteknologi, matematikk og elektroteknikk, Institutt for teknisk kybernetikk
• Format: Master Thesis
• Language: English
• Published: Institutt for teknisk kybernetikk 2013
• Subjects: Snøhvit
• Online Access: http://hdl.handle.net/11250/260990

In this thesis, a simplified model of the MEG loop at the Snøhvit field is established and implemented, together with an unscented Kalman filter. The model is tested with and without the Kalman filter. The model is tested with test data, while the model included the Kalman filter is tested with both test data and real production data.It is found that the model is some part away from the real process. The most significant difference is the rate of rich MEG into the MEG-regenerator onshore, which in the real process has severe oscillations relative to the rate of lean MEG rate out of the MEG-regenerator. In the model the rich MEG is modeled primarily to follow the lean MEG rate, and is therefore fairly constant.For many data sets of real production data, the system is not able to predict the formation water, because the error covariance matrix in the Kalman filter becomes negative semi definite. This can be caused by the Kalman filter not being robust enough or/and inconsistent data from Snøhvit. Still, when a data set is found where the system is able to predict the formation water is found, the system does predict a formation water rate that is located in the region that is expected at the Snøhvit field.