| dc.description.abstract | Gas lift optimization is a common artificial lift method, where gas is injected into the well to lower the hydrostatic pressure column of the fluid in the tubing. Subsequently, the bottom hole pressure will also be lowered, which will allow for gain in production rate. However, at a certainty gas lift rate, the gain from lowering the hydrostatic pressure loss will be evened out by the increase in frictional pressure loss. Increasing the gas lift further from this point will then also lower the production rate. Plotting the oil rate versus the gas lift rate, know as the gas lift performance curve (GLPC), will theoretically create a concave function. In this thesis, extremum seeking control, which is an adaptive control method, has been applied to a gas lift system to optimize the performance
Two different extremum-seeking controllers have been implemented in Simulink. One controller using a classical perturbation-based approach, and one controller where the nominal prat of the signal is included in the gradient estimation. Including the nominal part of the signal in the gradient estimation should increase the accuracy for small values in the perturbation parameters. The controllers were applied to a static model, GLPCs, which displayed good convergence and capability of handling constraints on the gas lift rate using a logarithmic barrier function.
The controllers were also applied to three dynamic models created in the industry standard dynamic simulator OLGA. One case was with a vertical well, while two cases were with a horizontal well created with data provided by AkerBP. An OPC server was used to connect OLGA with Simulink. For the vertical well, both controllers demonstrated practical convergence to the optimal, and a tuning sensitivity analysis showed that correctly tuning the controller can lead to a practical convergence arbitrarily close to the theoretical optimum. Simulations with changing well head pressure and water cut was successfully run. Both horizontal well cases displayed a GLPC which was not strictly concave. This made applying extremum-seeking control more challenging, but practical convergence was still reached. For the vertical well model and one of the horizontal well cases, simulations with two wells were run using a synchronization based optimizer. | en |
