Comparative Study of Parsimonious NARX Models for Three Phase Separator

摘要

The three phase separator is an important unit in oil and gas production facilities to separate gas, water and condensate from the fluid (raw gas) generated from gas wells. A dynamic model of the three phase separator is essential for process optimization and control design. Since first principles modelling of the three phase separator is complex, a data based model (Wavelet Network based Nonlinear AutoRegressive eXogenous model (WN-NARX)) is used to capture the dynamics of the process. As most of the terms in the WN-NARX expansion are redundant, this identification problem is over parameterized. In order to handle this issue, a sparsity constraint on the parameter vector is considered and sparse estimation algorithms such as Orthogonal Matching Pursuit (OMP) and Least Angle Regression (LAR) are used for identification of the WA-NARX model. The application of these sparse estimation methods for identification of an industrial three phase separator process is demonstrated.