摘要
故障诊断是保障系统安全稳定运行的重要途径。在风力发电系统中,深度网络得到广泛应用,其自动特征提取能力得到了更好的发挥。然而,多时间尺度数据的不完备特性使得传统的诊断模型无法直接应用这类数据,因此本研究提出了基于数据融合的改进卷积神经网络/长期短记忆(CNN-LSTM)故障诊断模型。该方法通过统一向量化处理使多时间尺度数据变得一致和完备。同时,通过构造端对端的诊断模型,实现特征提取和故障诊断的同时优化。改进CNN-LSTM模型依靠CNN挖掘数据的隐含特征,然后LSTM基于CNN提取的特征进一步捕捉时间序列的时间依赖性。相比于手工提取特征的方法,该模型能够学习到更有效的数据特征。在风力发电故障诊断任务中,该模型的准确率、精度、召回率、综合分类率均达到了较高水平,准确率为99.41%,精度为99.87%,召回率为97.75%,综合分类率为98.80%。与对比模型相比,该模型取得了更好的效果。
Fault diagnosis is an important means to ensure safe and stable operation of systems.deep learning has been widely applied in wind power generation where its automatic feature extraction capability is better utilized.However,the incomplete nature of multi-timescale data makes it impossible to directly apply the traditional diagnosis models to this type of data.Therefore,CNN-LSTM fault diagnosis model is proposed based on data fusion.This method uses a unified vectorization process to make multi-timescale data consistent and complete.At the same time,by constructing an end-to-end diagnosis model,feature extraction and fault diagnosis optimization simultaneously can be achieved.The improved CNN-LSTM model relies on CNN to excavate the hidden features of data,and then LSTM further captures the time dependence of time series based on the features extracted by CNN.Compared with the methods that rely on handcrafted feature extraction,this model can learn more effective data features.In wind power generation fault diagnosis tasks,the model achieves extremely high accuracy,precision,recall,and comprehensive classification rates of 99.41%,99.87%,97.75%,and 98.80%,respectively.Compared with some comparative models,this model achieves the best results.
作者
刘大为
王承凯
王巍
刘敏
LIU Dawei;WANG Chengkai;WANG Wei;LIU Min(Guohua Energy Investment Co.,Ltd.,Beijing 100007,China;School of Automation,Beijing University of Technology,Beijing 100081,China)
出处
《系统仿真技术》
2023年第4期293-301,共9页
System Simulation Technology
基金
国家自然科学基金重点项目(71934004)
国家自然科学基金资助项目(42071229)
作者简介
通信作者:刘敏,E-mail:unt1869@126.com;刘大为,男(1979-),河北唐山人,硕士,主要研究方向为电力市场、新能源电力交易;王承凯,男(1976-),山东茌平人,硕士,主要研究方向为新能源生产管理,多目标优化。
