ABSTRACT

The airfoil design optimizations are often limited due to the high computational cost and complex algorithm selections. However, such computational requirement has not been fulfilled well because some major unpaired performances exist between the mathematical benchmarks and airfoil designs. In this study, we compared three heuristic optimization algorithms: hill-climbing algorithm (HC), simulated annealing algorithm (SA), and genetic algorithm (GA), using three test functions in the airfoil design optimization of the wind turbines. The results show that with functions representing relatively flat shape, the HC and the SA are faster to find the global optimum than the GA. Tested with multimodal functions such as Shubert function, however, it is found that the HC and SA failed to find the global optimum although the SA has better possibilities to jump out the local extremum than the HC. For the airfoil optimization of S809 and NACA64418, it is found that the SA is more efficient than the GA. After the optimization, the S809 airfoil noise is decreased more than 0.9 dB, and the lift-drag ratio is improved 6.96% compared to its baseline in the given working condition. Similarly, the performance of NACA64418 is also improved with the proposed optimization algorithms. However, when the computational cost is taken into account, the performance of the three widely used algorithms is different from that in the benchmark tests. Therefore, our findings provide important insights of the airfoil design optimization of wind turbines on how to select algorithms and trade-off between computational cost and efficiency.

摘要

由于高计算成本和复杂的算法选择，翼型设计优化通常受到限制。然而，由于数学基准和翼型设计之间存在一些主要的未配对性能，因此这种计算要求并未得到很好的满足。在这项研究中，我们在风力涡轮机的翼型设计优化中使用三个测试函数比较了三种启发式优化算法：爬山算法 (HC)、模拟退火算法 (SA) 和遗传算法 (GA)。结果表明，对于表示相对扁平形状的函数，HC 和 SA 比 GA 更快地找到全局最优值。但是，使用 Shubert 函数等多模态函数进行了测试，发现虽然SA比HC有更好的跳出局部极值的可能性，但HC和SA未能找到全局最优值。对于 S809 和 NACA64418 的翼型优化，发现 SA 比 GA 更有效。优化后的S809翼型噪声降低0.9dB以上，升阻比在给定工况下较基线提升6.96%。同样，NACA64418 的性能也通过所提出的优化算法得到改善。然而，当考虑到计算成本时，三种广泛使用的算法的性能与基准测试中的不同。所以，