摘要
恶劣的气候条件会增加风机叶片结冰的风险.随着传感器技术在风电系统中的广泛应用,数据驱动的叶片覆冰检测方法引起了广泛关注.与传统方法相比,数据驱动方法可以避免专业知识的限制,降低安装检测设备引起的额外成本.然而,传统数据驱动模型挖掘数据信息的能力有限.同时,深度学习方法存在难以调整超参数的问题.为了解决上述问题,本文提出了蜻蜓算法(DA)和Transformer的风机叶片覆冰检测混合模型.其中,Transformer中的自注意力机制可以挖掘时间序列的局部和全局特征信息,蜻蜓算法可以智能优化Transformer的超参数.实验结果表明,相比已有的模型及Transformer而言,提出的混合模型具有更好的覆冰检测效果.
The risk of wind turbine blade icing is increased on severe weather conditions.The methods of data-driven blade icing detection have attracted extensive attention with the wide application of sensor technology in wind power systems.It can avoid the limitation of professional knowledge and reduce the additional cost caused by the installation of detection equipment compared with traditional methods.However,the ability of mining the data information for the traditional data-driven models is limited.At the same time,the method of deep learning is difficult to adjust the hyper-parameters.In order to solve the above problems,a hybrid model of dragonfly algorithm and Transformer for wind turbine blade icing detection is proposed in this paper.Furthermore,the self-attention mechanism in Transformer can mine the local and global feature information of time series,and the dragonfly algorithm(DA)can intelligently optimize the hyper-parameters of Transformer.The experimental results show that the proposed hybrid model has better effect of icing detection than the existing models and Transformer.
作者
汪磊
何怡刚
谭畅
WANG Lei;HE Yigang;TAN Chang(School of Electrical Engineering&Automation,Wuhan Univ.,Wuhan 430072,China;State Grid Hubei EHV Company,Wuhan 430000,China)
出处
《三峡大学学报（自然科学版）》
CAS
2022年第5期1-8,共8页
Journal of China Three Gorges University：Natural Sciences
基金
国家重点研发计划“智能电网技术与装备”专项“电力物联网关键技术”项目(2020YFB0905900)
国家重点研发计划“重大科学仪器设备开发”项目(2016YFF0102200)
国家自然科学基金(51977153、51977161、51577046)
中央高校基本科研业务费专项资金(2042021kf0233)
国家自然科学基金重点项目(51637004)
装备预先研究重点项目(41402040301)
湖北省重点研发计划项目(2021BEA162)
武汉市局科技计划项目(20201G01)。
