摘要
为了提高风机叶片振动异常预测的准确性,解决传统预测方法中的计算时间过长、局限于特定类型等问题,提出了一种基于时频分析特征数据的增量深度学习的预测方法.通过利用采集风机叶片的振动信号,采用时频分析方法提取振动信号的特征参数,然后建立增量深度学习模型以及CNN异常诊断器,从而实现风力涡轮机叶片的振动异常预测,同时保证模型具备自学习和自更新能力.基于风机叶片实际采集数据进行模型对比实验,验证了所提出方法的有效性.研究结果表明所提出方法均方根误差小于0.142,有效预测了风机叶片振动异常.
In order to enhance the accuracy of wind turbine blade vibration anomaly prediction and address the issues of extended computation time and applicability limitations associated with traditional prediction methods,this study proposes an incremental deep learning prediction approach based on time-frequency analysis feature data.By collecting vibration signals from wind turbine blades and employing time-frequency analysis techniques to extract characteristic vibration signal parameters,we establish an incremental deep learning model and a CNN-based anomaly diagnosis system to predict vibration anomalies in wind turbine blades.Importantly,the model possesses self-learning and self-updating capabilities.Model comparison experiments using actual data collected from wind turbine blades validate the effectiveness of the proposed method.The research results demonstrate that the proposed method achieves a root mean square error of less than 0.142,effectively predicting wind turbine blade vibration anomalies.
作者
林志灿
彭清和
LIN Zhi-can;PENG Qing-he(Practice Teaching Center,Minnan University of Science and Technology,Quanzhou Fujian 362700,China;School of Optoelectronics and Electromechanical Engineering,Minnan University of Science and Technology,Quanzhou Fujian 362700,China)
出处
《菏泽学院学报》
2023年第5期45-49,共5页
Journal of Heze University
基金
福建省教育厅中青年科研项目(JAT190886)。
关键词
深度学习
风机能源
异常预测
增量学习
时频特征
deep learning
wind energy
anomaly prediction
incremental learning
time-frequency features
