摘要
采用一种基于注意力机制的双向长短时记忆网络的深度学习结冰检测模型,引入欠采样和主成分析法方法对数据进行数据平衡以及降维处理,同时解决数据不平衡和数据冗余等造成的模型泛化能力差和检测精度低的问题。该模型在数据采集与监视控制系统的风场数据集上进行验证,其相关评价指标性能都有所提升,验证结果表明模型能有效检测风机叶片的结冰故障,对于风机的维护具有重要的参考意义。
In this paper,a deep learning icing detection model based on the bidirectional short and long duration memory network is adopted. Under sampling and principal component analysis methods are introduced to balance and reduce the dimension of the data. Meanwhile,the problem of poor generalization ability and low detection accuracy caused by data imbalance and data redundancy are solved. The model was verified on the wind field data set of data acquisition and monitoring control system,and the performance of related evaluation indexes was improved. The verification results show that the model can effectively detect the icing fault of fan blades,which has important reference significance for fan maintenance.
作者
马先超
廖宇
耿家豪
王港
MA Xianchao;LIAO Yu;GENG Jiahao;WANG Gang(Hubei Minzu University,Enshi 445000,China)
出处
《通信与信息技术》
2023年第1期58-63,共6页
Communication & Information Technology
基金
教育部2021年产学合作协同育人(项目编号:202102648020)资助项目
博士启动基金(项目编号:MD2019B006)资助项目
湖北民族大学高水平培育(项目编号:PY22012)资助项目。
关键词
风机叶片结冰
注意力机制
双向长短时记忆网络
欠采样
主成分析法
故障检测
Fan blade icing
Attention mechanism
Bi-directional long and short-term memory network
Under sampling
Principal component analysis
Fault detection
