基于注意力机制的CNN-BiLSTM的IGBT剩余使用寿命预测  

CNN-BiLSTM Based on Attention Mechanism for Prediction of IGBT Remaining Useful Life

作　　者：张金萍[1] 薛治伦 陈航 孙培奇 高策 段宜征 Zhang Jinping;Xue Zhilun;Chen Hang;Sun Peiqi;Gao Ce;Duan Yizheng(School of Mechanical and Power Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学机械与动力工程学院,沈阳110142

出　　处：《半导体技术》2024年第4期373-379,共7页Semiconductor Technology

摘　　要：针对绝缘栅双极型晶体管(IGBT)可靠性问题,提出了一种融合卷积神经网络(CNN)、双向长短期记忆(BiLSTM)网络和注意力机制的剩余使用寿命(RUL)预测模型,可用于IGBT的寿命预测。模型中使用CNN提取特征参数,BiLSTM提取时序信息,注意力机制加权处理特征参数。使用IGBT加速老化数据集对提出的模型进行验证。结果表明,对比自回归差分移动平均(ARIMA)、长短期记忆(LSTM)、多层LSTM(Multi-LSTM)、 BiLSTM预测模型,在均方根误差和决定系数等评价指标方面该模型的性能最优。验证了提出的寿命预测模型对IGBT失效预测是有效的。Aiming at the reliability problem of insulated gate bipolar transistors(IGBTs),a remai-ning useful life(RUL)prediction model was proposed by integrating the convolutional neural network(CNN),the bidirectional long-short term memory(BiLSTM)network and the attention mechanism,which can be used for lifetime prediction of IGBTs.In the model,CNN was used to extract characteristic parameters,BiLSTM was used to extract temporal information,and attention mechanism was used to weight the characteristic parameters.The proposed model was verified using the IGBT accelerated aging dataset.The results show that compared with autoregressive integrated moving average(ARIMA),long-short term memory(LSTM),multi-layer LSTM(Multi-LSTM)and BiLSTM prediction models,the proposed model has the best performance in terms of evaluation indicators,such as root mean square error,determination coefficient,etc.It is verified that the proposed lifetime prediction model is effective for prediction of IGBT failure.

关 键 词：绝缘栅双极型晶体管(IGBT) 失效预测 加速老化 长短期记忆(LSTM) 注意力机制 卷积神经网络(CNN) 

分 类 号：TN32[电子电信—物理电子学] TN306