基于Bo-BiLSTM网络的IGBT老化失效预测方法  

IGBT aging failure prediction method based on Bo-BiLSTM network

作　　者：万庆祝[1] 于佳松 佟庆彬[2] 闵现娟 WAN Qingzhu;YU Jiasong;TONG Qingbin;MIN Xianjuan(School of Electric and Control Engineering,North China University of Technology,Beijing 100144;School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044)

机构地区：[1]北方工业大学电气与控制工程学院,北京100144 [2]北京交通大学电气工程学院,北京100044

出　　处：《电气技术》2024年第3期1-10,共10页Electrical Engineering

基　　金：北京市教育委员会基金项目(110052972027/067);北京市自然科学基金项目(21C30037)。

摘　　要：针对绝缘栅双极型晶体管(IGBT)受热应力冲击后对其进行老化失效预测精度不高的情况,提出一种基于贝叶斯优化(Bo)-双向长短期记忆(BiLSTM)网络的IGBT老化失效预测方法。首先分析IGBT模块老化失效原理,然后基于NASA老化实验数据集建立失效特征数据库,最后利用Matlab软件构造Bo-BiLSTM网络预测失效特征参数数据。选取常用回归预测性能评估指标将长短期记忆(LSTM)网络模型、BiLSTM网络模型与Bo-BiLSTM网络模型的预测结果进行对比分析。结果表明,Bo-BiLSTM网络的模型拟合精度更高,基于Bo-BiLSTM网络的IGBT老化失效预测方法具有较好的预测效果,能够应用于IGBT的失效预测。Aiming at the low accuracy of aging failure prediction for insulated gate bipolar transistor(IGBT)after thermal stress impact,a bi-directional long short term memory(BiLSTM)network based on Bayesian optimization(Bo)is proposed to predict the aging failure of IGBTs.Firstly,the aging failure principle of IGBT module is analyzed,the failure characteristic database is established based on NASA aging experiment data set,and finally the Bo-BiLSTM network is constructed to predict the failure characteristic parameters by using Matlab software.Commonly used regression prediction performance evaluation indexes are selected to compare and analyze the prediction results of long short term memory(LSTM)network model,BiLSTM network model and Bo-BiLSTM network model.The results show that the model fitting accuracy of Bo-BiLSTM network is higher,so the IGBT aging failure prediction method based on Bo-BiLSTM network has better prediction effect and can be applied to IGBT failure prediction.

关 键 词：绝缘栅双极型晶体管(IGBT) 贝叶斯优化 双向长短期记忆(BiLSTM)网络 老化失效预测 

分 类 号：TN322.8[电子电信—物理电子学] TP183[自动化与计算机技术—控制理论与控制工程]