Browse by author
Lookup NU author(s): Mike Diessner, Dr Xiaonan Chen, Professor Kevin WilsonORCiD, Dr Richard Whalley
Full text for this publication is not currently held within this repository. Alternative links are provided below where available.
© TSFP 2024. All rights reserved. Most simulations and experiments that control the turbulent boundary layer to reduce the skin-friction drag, assume that all influential variables, such as the free-stream velocity and the blowing amplitude, can be controlled. In a real-world application, it is likely that some variables are given externally by the environment—such as the wind speed—and blowing strategies must be selected accordingly. This study brings the optimisation of blowing actuators closer to real-life conditions by enabling optimisation of randomly varying free-stream velocity. Bayesian optimisation is extended to dynamic environments with controllable and uncontrollable variables by fitting a global surrogate model over all variables but optimising only the controllable variables conditional on the environmental variables. By conditioning on measurements of the uncontrollable variables, optima for their full domain can be predicted. This is in contrast to keeping environmental variables fixed to a single value, where only one optimum is found and the experiment must be repeated multiple times for different values to achieve similar results. The presented approach increases the available information within a single optimisation run and results in a more sample-efficient and cost-effective algorithm. As an example application, the method is applied to a 5-dimensional wind farm simulator to maximise the energy production conditional on the wind speed by controlling the derating of five wind turbines. The new method outperforms the Nelder-Mead algorithm by 2.2–60.0% and performs comparably to standard Bayesian optimisation for five selected wind speeds while allowing predictions of optimal derating levels for the full range of wind speeds.
Author(s): Diessner M, Chen X, Wilson KJ, Whalley RD
Publication type: Conference Proceedings (inc. Abstract)
Publication status: Published
Conference Name: 13th International Symposium on Turbulence and Shear Flow Phenomena (TSFP13)
Year of Conference: 2024
Online publication date: 28/06/2024
Acceptance date: 02/04/2018
URL: http://www.tsfp-conference.org/proceedings/2023/52.pdf
