Variable neighborhood search for large offshore wind farm layout optimization

• Published in: Computers & operations research 2022-02, Vol.138, p.105588, Article 105588
• Main Authors: Cazzaro, Davide, Pisinger, David
• Format: Article
• Language: English
• Subjects: Algorithms; Heuristic; Heuristic methods; Layouts; Literature reviews; Metaheuristic; Offshore energy sources; Operations research; Optimization; Optimization techniques; P-Dispersion Sum Problem; Variable Neighborhood Search; Wind farm layout; Wind farms; Wind power; Wind turbines
• ISSN: 0305-0548; 0305-0548
• DOI: 10.1016/j.cor.2021.105588

The task of the offshore Wind Farm Layout Optimization Problem (WFLOP) is to place n wind turbines at a subset of k available positions to maximize power production and minimize wake losses. We start with a comprehensive literature review, where we highlight trends and analyze strategies used to solve WFLOP. Due to the increasing size of real-life wind farm projects, a more scalable approach is needed than those proposed in literature, to place hundreds of turbines at a subset of thousands of available positions. For this reason, we develop an advanced heuristic to solve WFLOP, based on the Variable Neighborhood Search meta-heuristic, considering in addition the minimum distance constraint and foundation costs. We also develop a novel initial solution algorithm, inspired by the P-Dispersion Sum Problem, which proves to be fast and effective. In the results, we compare and choose the best diversification strategy for this problem. Then, we test our heuristic on the most used set of instances in literature involving 100 positions, showing that our optimizer achieves higher production in less than 30 s than the original paper. These instances do not reflect the complexity of real problems, therefore we generate, and make available, a set of the 10 instances that mimic real cases, with over 20 thousand available positions, obstacles, foundation costs, wind distributions, and wind turbine characteristics. We benchmark cases with high, low and no foundation costs on this set. Our optimization technique is able to handle these large cases and reaches in 1 h objective values that are within 0.05% from the ones obtained in 10 h. Finally, we provide results of our heuristic in all these cases, to encourage future comparison of techniques on a more realistic and challenging dataset than currently used in literature. •Extensive literature review.•A new initialization heuristic based on the P-Dispersion Sum Problem.•Metaheuristic based on Variable Neighborhood Search to handle large number of available positions, considering parameters and constraints used in industry: wake effects, power curves, thrust curves, foundation costs, minimum distance.•Perform better than the most used instances in literature.•Provide a new set of synthetic instances that are realistic, and a benchmark on those.