PhD defence Hildo Bijl

17 October 2018 10:00 till 12:00 - Location: Aula Senaatszaal, Mekelweg 5 - By: DCSC

"LQG and Gaussian process techniques". Subtitle: "For fixed-structure wind turbine control".

Wind turbines are growing bigger to become more cost-efficient. To reduce the similarly growing vibrations, trailing-edge flaps can be installed to the turbine blades. Because of the variety of circumstances which the turbine should operate in, stochastic methods are needed to control these flaps. One such method is linear-quadratic-Gaussian control, where special attention is paid to the spread of the cost function. By reducing the variance of the cost, or by applying a discount exponent, the risk of damage may be reduced. Linear methods have their limitations though, which can be overcome by using nonlinear methods like Gaussian Process (GP) regression. In this case the challenge lies in efficiently incorporating large and noisy data sets. By making the right approximations, GP regression can be used for wind turbine applications, for example when using it for non-linear black-box modeling. Alternatively, a GP can be used to approximate a non-linear cost function like the damage equivalent load, as a function of for instance controller gains of a fixed-structure controller. When this is done, particle methods can be applied to find the optimum of the Gaussian process, automatically tuning said controller.


Promotor: J.W. van Wingerden