Nonlinear State Estimation and Noise Adaptive Kalman Filter Design for Wind Turbines

Abstract

Modern wind turbines have grown significantly in nacelle heights and rotor diameters over the last three decades and will probably do so in the future. Additionally, more and more turbines are mounted in shallow waters on offshore monopiles and also on floating platforms in the deep sea. These technological developments implicate that the complex structural composition of the turbines becomes more flexible and simultaneously more and competing control objectives appear above the horizon. In this context, advanced state-feedback based control schemes have emerged in the wind sector in order to tackle these new challenges effectively though requiring the mostly hidden and immeasurable information about the dynamic state of the wind turbine. In order to obtain this valuable information without additional measurements and sensors, the present thesis bridges the scientific gap between the nonlinear estimation theory on the one hand and the practical application to wind turbine control systems on the other hand. This approach includes the investigation of the nonlinear filter algorithms, the control-oriented physical models and the design methodology needed to make nonlinear state estimation techniques ready for wind turbine application. The results of this approach are so-called virtual (model-based) sensors that are employed for multiple estimation tasks, such as the observation of unknown disturbances and the online reconstruction of mechanical component loads. These sensors are applicable whenever it is impossible or too expensive to measure the desired quantities directly. The thesis explores first the suitable nonlinear algorithms to solve the estimation problems. The focus is here laid upon the sigma-point Kalman filters (SPKF) where classical and adaptive versions are presented. As widely known, the free design parameters of these filters have a significant influence on the expected estimation accuracy. An unfortunate filter parameter design leads to weak filter performance or (even worse) filter divergence. This is very critical in closed-loop systems where the state estimates are essential for the feedback controller. For this reason, the dissertation investigates two approaches to address this problem. The first step is the optimal design of the filter parameters based on numerical optimization. Therewith, all the relevant information about the wind turbine are exploited prior to application in order to find the best initial design parameters. The second step is the selective noise adaptation of certain filter parameters. This approach improves on the filter performance when the previous knowledge is insufficient for a proper initial design or some critical filter parameters are unknown and time-varying. Moreover, a comprehensive engineering suite is developed in order to integrate the necessary functionality to perform an automated filter performance assessment. Finally, the strengths of these techniques are demonstrated illustratively for a variety of test scenarios, with extensive simulation results, for different estimation problems and different filter types in order to investigate all relevant practical aspects. In a nutshell, this thesis provides the theoretical foundations, the practical application and also the simulative proof of concept in order to realize the wind turbine state estimation effectively and to bring it successfully into practice. Thereby, these virtual sensors shall level the ground not only for advanced state-feedback control, but also provide further insight into the system’s internal behavior which can be exploited in future, for instance, for remaining useful life-time assessment based on reconstructed, experienced loads.