摘要
传统神经网络模型无法有效提取信号时序特征。为充分利用光伏阵列信号的时序相关性特征、增强神经网络模型对数据的挖掘能力、更好地提升卷积神经网络(CNN)分类精度,把CNN和门控循环单元(GRU)相结合,提出了新的光伏阵列故障诊断方法。首先,利用CNN对数据进行标签处理。然后,将提取的标签作为GRU的输入,进一步提取时空特征。为加强该方法分类优势,添加了自适应批量归一化(AdaBN)函数。试验结果表明,该方法较传统神经网络分类方法更具优越性,提高了光伏阵列故障分类的准确率。最后,通过增加噪声模拟现场环境,验证了结果的正确性。
The traditional neural network model cannot effectively extract the signal temporal features.To make full use of the timing correlation features of photovoltaic array signals,enhance the data mining ability of neural network models,and better improve the classification accuracy of convolutional neural networks(CNN),the new fault diagnosis method for photovoltaic arrays is proposed by combining CNN and gated recurrent unit(GRU).Firstly,the CNN is used to label the data.Then,the extracted labels are used as the input of GRU to further extract spatial-temporal features.To enhance the classification advantage of the method,the adaptive batch normalization(AdaBN)function is added.The experimental results show that the method is superior to the traditional neural network classification method and improves the accuracy of photovoltaic array fault classification.Finally,the correctness of the results is verified by adding noise to simulate the field environment.
作者
陈伟
陈克松
纪青春
裴婷婷
王忠飞
何峰
CHEN Wei;CHEN Kesong;JI Qingchun;PEI Tingting;WANG Zhongfei;HE Feng(College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;State Grid Lanzhou Power Supply Company,Lanzhou 730070,China)
出处
《自动化仪表》
CAS
2022年第6期13-17,共5页
Process Automation Instrumentation
基金
国家自然科学基金资助项目(51767017)。
关键词
光伏阵列
故障诊断
卷积神经网络
循环神经网络
时序特征
Photovoltaic array
Fault diagnosis
Convolutional neural network(CNN)
Recurrent neural network(RNN)
Temporal features