Enhanced Kalman filtering for a 2D CFD Navier-Stokes wind farm model

Wind turbines are often grouped together for financial reasons, but due to wake development this usually results in decreased turbine lifetimes and power capture, and thereby an increased levelized cost of energy (LCOE). Wind farm control aims to minimize this cost by operating turbines at their optimal control settings. Most state-of-the-art control algorithms are open-loop and rely on a low fidelity, static flow model. Closed-loop control relying on a dynamic model and state observer has real potential to further decrease wind's LCOE, but is often too computationally expensive for practical use. In this work two time-efficient Kalman filter (KF) variants are outlined incorporating the medium fidelity, dynamic flow model "WindFarmSimulator" (WFSim). This model relies on a discretized set of Navier-Stokes equations in 2D to predict the flow in wind farms in a horizontal plane at hub height at low computational cost. The filters implemented are an Ensemble KF and an Approximate KF. Simulations in which a high fidelity simulation model represents the true wind farm show that these filters are typically several orders of magnitude faster than a regular KF with comparable or better performance, correcting for wake dynamics that are not modeled in WFSim (noticeably, wake meandering and turbine hub effects). This is a first big step towards real-time closed-loop control in wind farms.