Dynamic Wind Farm Control using the WFSim flow model

Abstract

Renewable energy is becoming more and more important in today's society. Wind energy plays an important role in the production of renewable energy. Due to economic advantages, wind turbines are often sited close together, creating wind farms. As a result, the wind turbines in the farm become interconnected due to the wakes of the turbines. In conventional wind farm control, this interconnection is ignored, and wind turbines are operated at their own local optimum. This control strategy is referred to as greedy con- trol. In the wake of a turbine, the wind speed is reduced. As a result, the power production of a wind turbine that is situated in a wake is also significantly lower. The greedy control strategy may therefore not be optimal for wind farm control, which subsequently has been receiving an increasing amount of attention recently. Most research on wind farm control focusses on steady- state optimization, i.e., finding the optimal steady-state given certain con- stant conditions. Since wind flow is always subject to change, the potential gain of this procedure is limited. In this thesis, a closed-loop dynamic wind farm controller is presented us- ing a Model Predictive Control (MPC) framework. The control objective is to maximize the power production of the wind farm, resulting in an eco- nomic MPC problem. As the optimal input is time-dependent, the number of control inputs and states involved increase as the prediction horizon of the MPC problem increases, resulting in a more complex problem than the steady-state optimization problem. As a model, WFSim is used, which is based on 2-dimensional Navier-Stokes equations. This results in a high-dimensional problem from which the opti- mum is unknown. The adjoint method is applied to determine the gradient of the objective function in a time-efficient manner. The controller developed in this thesis increases the power production of a wind farm with respect to the conventional greedy control stategy, and is able to adapt to changes in the atmospheric conditions. The adjoint-based MPC algorithm therefore shows real potential to perform real-time dynamic control on wind farms.