This research aims to develop a method to create a wind farm layout that is robust against the uncertainty source, the inter-year variation of Weibull parameters and wind direction sector probabilities. A wind farm layout optimisation problem under uncertainty corresponds to optimisation under uncertainty (OUU), which is computationally expensive. It is, therefore, proposed to solve this issue by applying surrogate modelling. The thesis considers and compares two different surrogate models, polynomial chaos expansion (PCE) and Kriging, on two different applications. First, the surrogate model is built for the wind farm power model to emulate the medium-fidelity model that requires to evaluate the wake effect in every wind speed and wind direction. Another application of the surrogate model is for the uncertainty quantification (UQ). As the computational expensive Monte Carlo method is traditionally used to propagate the uncertain variables, the use of surrogate models can further reduce the number of uncertain samples to obtain the statistic output response.
Once the statistic of annual energy production (AEP) of the wind farm can be computed efficiently, the optimisation under uncertainty problem is performed with the genetic algorithms. The computational time of optimisation under uncertainty is reduced by using the surrogate UQ model, and a layout that may be more robust to the inter-year variation is found as indicated by a higher P90 value of AEP.