ORCID

Abstract

The design of offshore wind farms is computationally challenging, requiring the simultaneous optimisation of many conflicting objectives. Solving this problem is of paramount importance if society is to meet ambitious net zero goals. The problem is solved by identifying an optimal arrangement of individual turbines such that all objectives are optimised. However, a single solution does not exist due to the inherent conflict between objectives, and a set of solutions must be identified. As well as the challenge in generating optimal solution sets, there exists a decision support task if the solutions are to be effectively presented to a decision maker. This study focuses on six key objectives: wind farm efficiency, annual energy production, electric cable length, number of wind turbines, levelised cost of energy, and total area. Two evolutionary algorithms, NSGA-II and NSGA-III, were employed to explore the solution space efficiently. Performance evaluation was performed using spacing, generational distance, and hypervolume metrics. The aforementioned algorithms and metrics were applied to three wind farm layouts: a discrete layout and two continuous layouts. The NSGA-III algorithm was shown to perform better than its predecessor (NSGA-II). The difference was small, albeit significant. Previous works (e.g., Rodrigues et al. (2016); Mytilinou and Kolios (2017)) in which many-objective optimisations were discussed provided little insight into the visualisation and interpretation of the results. While the mentioned work used parallel coordinate plots, this work provides a deeper insight by presenting the results via Principal Component Analysis (PCA) and Multi Dimensional Scaling (MDS) plots. The best solution, containing 6188 wind farm layouts, was found by the NSGA-III algorithm on a continuous wind farm layout with repair mechanism. From the best solution, the wind farms containing 27, 102 and 160 wind turbines were selected and compared with the real wind farms located around the UK. It was demonstrated that the optimiser could identify better wind farm layouts concerning annual energy production, efficiency, and LCOE than the real wind farm layouts of Rhyl Flats and Greater Gabbard.

DOI

10.1016/j.asoc.2025.112879

Publication Date

2025-02-27

Publication Title

Applied Soft Computing

Volume

173

ISSN

1568-4946

Keywords

Wind farm optimisation, Many-objective optimisation, NSGA-II, NSGA-III, Hypervolume, Spacing, Generational distance, Parallel coordinate plot, Dimension reduction

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Download

Share

COinS