Abstract Excessive fore-aft vibrations of wind turbine tower are the major reason for tower collapsing, which can be evaluated using the equivalent fatigue load (EFL). Wake effects are generated by the former wind turbines on the latter ones, and can greatly increase EFL and reduce lifetime of the latter ones, but they were seldom considered. Consequently, this study calculated EFL countering the wake effects under real wind distributions. The effects of wind turbine spacing on EFL indicate that the wake effects between wind turbines with spacing of approximately 2 times of rotor radius should be considered. Subsequently, tuned mass damper (TMD) was placed in the nacelle to passively control the fore-aft vibrations. An optimization tool for TMDs was developed based on the radial basis function neural network and genetic algorithm, which was compared with the theoretical equations and parametric analysis. GPU acceleration technology was utilized. Numerical results show that, under real wind distributions, the TMDs obtain at least 40.1% EFL reduction, which is 8.9% higher than the theoretically optimized one. Importantly, the GPU-based codes can run 2001 times faster than the CPU-based ones, and the optimization tool can reduce 44.9% computational time further. The codes are made available for other researchers.

中文翻译：

---

使用GPU加速和机器学习技术在真实风分布下优化风力涡轮机TMD对抗尾流效应

摘要 风电塔架前后振动过大是塔架倒塌的主要原因，可采用等效疲劳载荷（EFL）进行评价。尾流效应是由前者的风力机对后者的风力机产生的，可以大大增加 EFL 并降低后者的寿命，但很少被考虑。因此，这项研究计算了 EFL，以抵消真实风分布下的尾流效应。风机间距对 EFL 的影响表明，应考虑间距约为转子半径 2 倍的风机之间的尾流效应。随后，调谐质量阻尼器 (TMD) 被放置在机舱中以被动控制前后振动。基于径向基函数神经网络和遗传算法开发了TMDs优化工具，并与理论方程和参数分析进行了比较。使用了GPU加速技术。数值结果表明，在真实风分布下，TMD 至少获得 40.1% 的 EFL 降低，比理论上优化的高 8.9%。重要的是，基于 GPU 的代码可以比基于 CPU 的代码运行速度快 2001 倍，并且优化工具可以进一步减少 44.9% 的计算时间。这些代码可供其他研究人员使用。优化工具可以进一步减少 44.9% 的计算时间。这些代码可供其他研究人员使用。优化工具可以进一步减少 44.9% 的计算时间。这些代码可供其他研究人员使用。