摘要
为了提高电力电子电路故障预测的准确性和快速性,对所采用的故障特征性能参数利用LIFT保真去噪法进行处理,并根据数据特征划分多个数据窗口,利用长短记忆神经网络LSTM对DC-DC电路特征性能参数达到自适应预测。预测时可自动调整窗口的大小,找到最优的训练集个数,使得预测精度最高。以DC-DC电路中的boost电路为例,以输出电压作为监测信号,并计算特征参数Upp和Uave。根据电路性能设定电压阈值,分别建立基于自适应数据窗的LSTM故障预测模型和基于自适应数据窗的LITF-LSTM故障预测模型,通过对比分析预测结果,充分说明LITF-LSTM预测模型的有效性,最后分析预测数据是否达到所设阈值来判断电路在未来时刻是否发生故障。
In order to improve the accuracy and rapidity of power electronic circuit fault prediction, In this paper, the fault characteristic performance parameters used are processed according to the LIFT fidelity denoising method, and divides multiple data windows according to data characteristics, using Long Short-Term Memory(LSTM) to adaptively predict the characteristic performance parameters of DC-DC circuits.The size of the window can be adjusted automatically during prediction, finds the optimal number of training sets, makes the prediction accuracy the highest.Taking the boost circuit in the DC-DC circuit as an example, Use the output voltage as the monitoring signal, calculates the characteristic parameters Upp and Uave to set the voltage threshold according to the circuit performance, and establish the LSTM fault prediction model based on the adaptive data window and the LITF-LSTM fault prediction model based on the adaptive data window, By comparing and analyzing the prediction results, explains the validity of the LITF-LSTM prediction model, Finally, analyzes whether the predicted data reaches the set threshold to determine whether the circuit is malfunctioning in the future.
作者
张强
战柯柯
ZHANG Qiang;ZHAN Keke(State Grid Anhui Electric Power Co.,Ltd.Nanling Power Supply Company,Wuhu 242400,China;State Grid Anhui Electric Power Co.,Ltd.Fanchang Power Supply Company,Wuhu 241200,China)
出处
《河北电力技术》
2021年第2期7-11,28,共6页
Hebei Electric Power
