基于PSO-BP神经网络模型的IGBT老化预测Ageing Prediction of IGBT Based on PSO-BP Neural Network Model
边少聪,王宇
BIAN Shaocong,WANG Yu
摘要(Abstract):
IGBT(绝缘栅双极型晶体管)在使用过程中会产生热量,在其内部产生疲劳累积,最终老化失效。而在IGBT老化过程中,其输出电气参数会发生变化,可据此判断IGBT的老化状态。为此,基于NASA PCoE研究中心提供的IGBT加速老化数据,选取IGBT关断过程中集射极电压尖峰值为失效特征参数;采用卡尔曼滤波对该参数进行预处理后,应用BP神经网络建立IGBT老化预测模型;分别使用遗传算法和粒子群优化算法对BP神经网络初始权值及阈值进行寻优,解决其易陷入局部最优的缺陷;应用均方误差、平均绝对误差、相关系数作为评价指标,对网络性能进行评估。结果表明,经优化后的网络效果均优于BP神经网络,而其中以PSO-BP算法所构建的IGBT老化预测模型为最优模型,可以更准确地实现IGBT的老化预测。
IGBT(insulated gate bipolar transistor) generates heat during use, causing inside fatigue buildup and ultimately ageing failure. In the ageing process of the IGBT, the output electrical parameters will change,and thus the ageing state of the IGBT is judged. Therefore, based on the IGBT accelerated aging data provid ed by the NASA PCoE research center, the peak value of the collector voltage(Vce) in the IGBT turn-off process is selected as the failure characteristic parameter; after preprocessing with Kalman filter, BP neural network is used to establish IGBT ageing prediction model; the GA and PSO algorithms are used to optimize the initial weights and thresholds of BP network respectively to overcome its proneness to local optimal solution;the mean square error(MSE), the mean absolute error(MAE), and the correlation coefficient are used as evaluation indicators to evaluate the network performance. The results show that the optimized network effect is better than the BP network, and the IGBT ageing prediction model constructed by PSO-BP algorithm is the optimal model that can accurately predict IGBT ageing.
关键词(KeyWords):
IGBT;BP神经网络;GA-BP;PSO-BP;老化预测
IGBT;BP neural network;GA-BP;PSO-BP;ageing prediction
基金项目(Foundation): 西安市科技计划资助项目(2017074CG/RC037(XAGC009))
作者(Author):
边少聪,王宇
BIAN Shaocong,WANG Yu
DOI: 10.19585/j.zjdl.201911010
参考文献(References):
- [1]刘宾礼,刘德志,唐勇,等.基于加速寿命试验的IGBT模块寿命预测和失效分析[J].江苏大学学报(自然科学版),2013,34(5):556-563.
- [2]方鑫,周雒维,姚丹,等.IGBT模块寿命预测模型综述[J].电源学报,2014,12(3):14-21.
- [3]李志刚,李雄,孔梅娟,等.IGBT寿命评估中解析预测模型综述[J].微电子学,2017,47(3):396-400.
- [4]YANG L,AGYAKWA P A,JOHNSON C M.Physics-offailure lifetime prediction models for wire bond interconnects in power electronic modules[J].IEEE Transactions on Device&Materials Reliability,2013,13(1):9-17.
- [5]孟苓辉.牵引变流器的故障预测与健康管理(PHM)及可靠性评估技术研究[D].北京:北京交通大学,2017.
- [6]莫思雨,周晴.基于MEA-BP神经网络的图像复原方法[J].电子设计工程,2018,26(21):185-189.
- [7]王小川.MATLAB神经网络43个案例分析[M].北京:北京航空航天大学出版社,2013.
- [8]程换新,刘军亮.基于GA-BP神经网络在手写数字识别中的应用研究[J].电子测量技术,2019,42(9):89-92.
- [9]禹健,郭天星,高超.基于GA-BP算法的IGBT结温预测模型[J].自动化与仪表,2019,34(1):79-83.
- [10]陈良波,郑亚青.粒子群算法在多峰值函数优化中的应用[J].贵州师范学院学报,2012,28(12):23-26.
- [11]杨杰,赵敏,苏浩,等.基于粒子群矢量搜索融合的射流轨迹识别方法[J].电子测量与仪器学报,2016,30(5):803-809.
- [12]朱奇光,王洪瑞.IC-PSO算法的收敛性分析及应用研究[J].光电工程,2010,37(4):108-112.
- [13]李孝全,黄超,徐晨洋,等.基于改进PSO-LSSVM和蒙特卡洛法的电力系统可靠性评估[J].河海大学学报(自然科学版),2016,44(5):458-464.
- [14]陈远鸣,常建华,沈婉,等.基于改进型BP神经网络的SF6气体传感器[J].电子测量与仪器学报,2017,31(10):1582-1588.
- [15]陈柏元,张红民,罗永涛,等.一种基于GA-BP的高压混合输电线路故障分类方法[J].重庆理工大学学报(自然科学),2018,32(2):199-206.
- [16]崔月生.基于PSO-BP算法的雷电流概率预测研究[J].国外电子测量技术,2018,37(10):118-123.