Data-enhanced model predictive wind turbine control

RWTH Publications (RWTH Aachen)

Abstract

Generating renewable energy through Wind Turbines (WTs) is a key technology in the current energy transition. In recent years, both the absolute amount of energy generated by wind and the share of wind energy in global electricity generation increased. This imposes increasingly complex requirements on the dynamic WT operation. Current and future challenges include not only maximizing the power output, but also reducing mechanical loads, among others. For the dynamic operation of WTs, their control plays a major role.Model-based Predictive Control (MPC) of WTs in combination with model-based state estimation shows the potential to integrate multiple requirements in a single optimizationbased controller and optimally resolve conflicting objectives within constraints.At the same time, Machine Learning can be used for data-based modeling of complex relationships that are difficult to model physically. Combining this with the model-based control algorithms ultimately shows the potential for addressing complex WT operation requirements by incorporating relevant submodels.Additionally, as a preliminary step towards experimental testing of WT controllers, WTemulators in the form of nacelle test benches demonstrate the potential for reproducibly testing and validating dynamic operations on quasi-final prototypes.This work demonstrates how data-driven state estimation can estimate mechanical WT loads at runtime. The accuracy of these estimations appears sufficiently precise for load-reducing control. As part of an experiment, an MPC including state estimation of a real 3MW-WT is presented. Stable operation was achieved over three hours in the field with continuous full access. Additionally, preliminary results of a data-based enhancement of this MPC to include a prediction of mechanical loads are shown in an experiment as well. Simulation results of this extension indicate a reduction in the dynamics of the thrust force in the low-frequency range. Finally, this work investigates WT-emulators in the form of nacelle test benches for testing and validating WT controllers. This identifies a dynamic distortion in the test results due to the interaction between the test bench controller and the WT controller.

Authors 0

1. Author list not loaded yet.