随着电力电子器件的高速发展，各种电力电子器件被广泛的应用于各个领域。半导体器件作为电力电子系统中的核心器件，其性能关系着整个系统的可靠性，若不能及时发现老化或者失效会给整个系统造成很大损失，因此对半导体器件进行性能退化趋势和剩余寿命预测分析十分必要。鉴于此，以绝缘栅双极型晶体管IGBT(Insulated Gate Bipolar Transistor，IGBT)作为研究对象，根据其工作原理以及失效机理，构造融合多特征参数的健康度指标，基于健康度指标对IGBT的性能退化和剩余寿命进行预测。论文主要工作如下。

(1)提出一种包含多维特征参数中性能退化信息的健康度指标。在分析IGBT的结构及其失效机理的基础上，通过对比IGBT的失效参数和失效标准，选取了IGBT集电极的拖尾电流、栅极漏电流、集电极-发射极电压作为判断IGBT性能退化程度的三种特征参数，特征参数中包含IGBT剩余寿命和性能退化状态信息，基于三种特征参数构造了IGBT健康度指标，将健康度指标作为IGBT性能退化判别依据，为进一步研究提供了理论基础。

(2)采用IMHA(Independent Metropolis-Hastings Algorithm，IMHA)优化粒子滤波算法(IMHA-PF)对IGBT性能退化状态的预测模型。在采用粒子滤波算法(PF)对IGBT进行性能退化预测时，为了避免重采样过程中产生的样本贫化问题，采用IMHA对粒子滤波算法进行优化，建立IMHA-PF的预测模型。最后，采用多特征参数融合的健康度指标并基于IMHA-PF模型预测对IGBT的性能退化状态进行预测，预测结果表明，在不同预测起点下，优化后IMHA-PF算法与传统预测方法对比，其均方根误差和平均误差最小，预测精度更高，能更好地预测IGBT性能退化状态。

(3)建立粒子群优化CNN-BiLSTM网络的IGBT剩余寿命预测模型，即PSO-CNN-BiLSTM网络模型。在采用CNN-BiLSTM对IGBT进行剩余寿命预测时，基于粒子群算法(PSO)对CNN-BiLSTM参数进行寻优，优化后能大大提升剩余寿命预测的准确度。采用多特征参数融合的健康度指标并基于PSO-CNN-BiLSTM预测模型对IGBT的剩余寿命进行预测，将其与传统的预测模型进行对比，验证提出的预测模型在预测精度方面优于传统的预测模型，能够实现对IGBT剩余寿命的有效预测。

本文主要分析了IGBT多个特征参数，将其融合为健康度指标表征IGBT性能退化状态和建立寿命预测模型，可以对IGBT器件进行健康状态监测和安全评估，对于IGBT的可靠性研究具有重要意义。

With the rapid development of power electronic devices, various power electronic devices are widely used in various fields. As a core device in power electronic system, the performance of semiconductor devices is related to the reliability of the whole system, and if aging or failure is not detected in time, it will cause great loss to the whole system. In view of this, the insulated gate bipolar transistor IGBT is used as the research object, and based on its operating principle and failure mechanism, a health index incorporating multiple characteristic parameters is constructed to predict the remaining life and performance degradation of the IGBT based on this health index. The main work of the paper is as follows.

(1) To propose a health index that includes performance degradation information in multi-dimensional characteristic parameters. Based on the analysis of the structure of IGBT and its failure mechanism, the trailing current of IGBT collector, gate leakage current, and collector-emitter voltage are selected as three characteristic parameters to judge the degree of IGBT performance degradation by comparing the failure parameters and failure criteria of IGBT, and the characteristic parameters contain the information of remaining life and performance degradation state of IGBT, and the IGBT health index is constructed based on the three characteristic parameters. The IGBT health index is constructed based on the three characteristic parameters, and the health index is used as the basis for the discrimination of IGBT performance degradation, which provides a theoretical basis for further research.

(2) The IMHA optimized particle filter algorithm (IMHA-PF) is used to predict the degradation state of IGBT performance. When using particle filter algorithm (PF) for performance degradation prediction of IGBT, in order to avoid sample depletion problems during resampling, IMHA was used to optimize the particle filter algorithm and establish an IMHA-PF prediction model. Finally, the health indicators fused with multiple feature parameters were used to predict the performance degradation status of IGBT based on the IMHA-PF model. The prediction results showed that under different prediction starting points, the optimized IMHA-PF algorithm had the smallest root mean square error and mean square error compared to traditional prediction methods, and the prediction accuracy was higher, which can better predict the performance degradation status of IGBT.

(3) A particle swarm optimization CNN-BiLSTM network-based IGBT remaining life prediction model is established. In the remaining life prediction of IGBTs using CNN-BiLSTM, the optimization of CNN-BiLSTM parameter parameters based on particle swarm algorithm (PSO) can greatly improve the accuracy of remaining life prediction. The PSO-CNN-BiLSTM prediction model is used to predict the remaining life of IGBTs based on the health indexes of multi-feature parameter fusion and compared with the traditional prediction model to verify that the proposed prediction model is better than the traditional prediction model in terms of prediction accuracy and can achieve effective prediction of the remaining life of IGBTs.

This article mainly analyzes multiple characteristic parameters of IGBT, and integrates them into health indicators to characterize the performance degradation status of IGBT and establish a life prediction model. This can monitor the health status and safety evaluation of IGBT devices, which is of great significance for the reliability research of IGBT.

[1]陆剑峰, 徐煜昊, 夏路遥, 等. 数字孪生支持下的设备故障预测与健康管理方法综述[J]. 自动化仪表, 2022, 43(6):1-7.

[2]Xiao D, Qin C, Ge J, et al. Self-attention-based adaptive remaining useful life prediction for IGBT with Monte Carlo dropout[J]. Knowledge-based systems, 2022,5:239-245.

[3]陈兴玉, 张红旗. 面向预测性维修的工程装备故障预测方法研究综述[J]. 智能制造, 2022(2):50-55.

[4]Grall A, Bérenguer C, Dieulle L. A condition-based maintenance policy for stochastically deteriorating systems[J]. Reliability Engineering & System Safety, 2002, 76: 167-180.

[5]Jay L, Jun N, Dragan D. Intelligent prognostics tools and e-maintenance[J]. Computers in Industry, 2006, 57: 476-489.

[6]吕瑛， 王振宇， 赵岩， 张会平， 陈娟. 电子元器件加速退化寿命评估方法研究[J]. 计算机测量与控制, 2021, 29(11):230-234+239.

[7]朱慧军, 杨洪磊. 朱慧军,杨洪磊.机械设备故障规律及运行趋势预测方法综述[J].电子测试,2021(02):57-58+50.

[8]余萍,曹洁. 深度学习在故障诊断与预测中的应用[J]. 计算机工程与应用, 2020, 56(03):1-18.

[9]Liu Q, Zhang J, Li K, et al. The Remaining Useful Life Prediction by Using Electrochemical Model in the Particle Filter Framework for Lithium-Ion Batteries[J]. IEEE Access, 2020, 8:126661-70.

[10]Chen Y X, Wang Z Z, Cai Z Y. Optimal Maintenance Decision Based on Remaining Useful Lifetime Prediction for the Equipment Subject to Imperfect Maintenance[J]. IEEE Access, 2020, 8:6704-16.

[11]Lotfi S D, Jaouher B A, Mohamed B, et al. An integrated wind turbine failures prognostic approach implementing kalman smoother with confidence bounds[J]. Applied Acoustics, 2018, 138:199-208.

[12]Jay L, Wu F J, Zhao W Y, et al. Prognostics and health management design for rotary machinery systems-reviews, methodology and application[J]. Mechanical Systems and Signal Processing, 2014, 42(1):314-334.

[13]Cheng S J, Li W K, Chang T J, et al. Data-Driven Prognostics of the SOFC System Based on Dynamic Neural Network Models[J]. Energies, 2021, 14.

[14]Wang Y, Gogu C, Binaud N, et al. A model-based prognostics method for fatigue crack growth in fuselage panels[J]. Chinese Journal of Aeronautics, 2019, 32(02):186-198.

[15]张军, 张犁, 成瑜. IGBT模块寿命评估研究综述[J]. 电工技术学报, 2021, 36(12):16.

[16]Thirumarimurugan M, Bagyalakshmi N, Paarkavi P. Comparison of fault detection and isolation methods: A review[C]//International Conference on Intelligent Systems and Control, 2016, 1:1-6.

[17]Khoukhi A, Khalid H, Doraiswami R, et al. Fault detection and classification using kalman filter and hybrid neuro-fuzzy systems[J]. International Journal of Computer Applications, 2012, 45(22):7–14.

[18]王庆锋, 卫炳坤, 刘家赫, 等. 一种数据驱动的旋转机械早期故障检测模型构建和应用研究[J]. 机械工程学报, 2020, 56(16):22-32.

[19]Ashfaq R A R, Wang X Z, Huang J Z, et al. Fuzziness based semi-supervised learning approach for intrusion detection system [J]. Information Sciences, 2017, 378:484–497.

[20]Zhao Y, Ball R, Mosesian J, et al. Graph-based semi-supervised learning for fault detection and classification in solar photovoltaic arrays. IEEE Transactions on Power Electronics, 2015, 30(5):2848- 2858.

[21]Li L L, Liu Z F, Tseng M L, et al. Prediction of IGBT power module remaining lifetime using the aging state approach[J]. Microelectronics & Reliability, 2019, 102(11):113476.

1-113476.10.

[22]Niedernostheide F J, Schulze H J, Laska T, et al. Progress in IGBT Development[J]. Iet Power Electronics, 2018, 11(4):646-653.

[23]Li C J, Lee H. Gear fatigue crack prognosis using embedded model, gear dynamic model and fracture mechanics[J]. Mechanical Systems & Signal Processing, 2005, 19(4):836

-846.

[24]Xu Q, Chen Z, Wu K, et al. KDnet-RUL: A Knowledge Distillation Framework to Compress Deep Neural Networks for Machine Remaining Useful Life Prediction[J]. IEEE Transactions on Industrial Electronics, 2021, PP(99):1-1.

[25]Chen Z, Wu M, Zhao R, et al. Machine Remaining Useful Life Prediction via an Attention Based Deep Learning Approach[J]. IEEE Transactions on Industrial Electronics, 2020, PP(99):1-1.

[26]Loutas T H, Roulias D, Georgoulas G. Remaining Useful Life Estimation in Rolling Bearings Utilizing Data-Driven Probabilistic E-Support Vectors Regression[J]. IEEE Transactions on Reliability, 2013, 62(4):821-832.

[27]Xue L, Liu Y, Xiong Y, et al. A data-driven shale gas production forecasting method based on the multi-objective random forest regression [J]. Journal of Petroleum Science and Engineering,2021,196 :107801.

[28]Mohammad S, Alireza A, Suresh P. Unified IGBT prognostic using natural computation. 2015 IEEE International Conference on Digital Signal Processing (DSP) [C]. IEEE, 2015: 698-702.

[29]刘子英, 朱琛磊. 基于Elman神经网络模型的 IGBT 寿命预测[J]. 半导体技术. 2019, 44(5): 395-400.

[30]李志刚, 梅霜, 陈红纳等. 基于饱和压降测量的功率模块老化模型研究[J].太阳能学报, 2017, 38(12): 3319-3323.

[31]Liu Z, Mei WJ, Zeng XP, et al. Remaining Useful Life Estimation of Insulated Gate Biploar Transistors (IGBTs) Based on a Novel Volterra k-Nearest Neighbor Optimally Pruned Extreme Learning Machine (VKOPP) Model Using Degradation Data[J]. Sensors, 2017, 17(11): 2524.

[32]张西应,曾文彬, 操国宏, 等. 大功率晶闸管器件加速老化试验和寿命预测方法研究[J]. 电子质量, 2022(05):64-68.

[33]李岩, 侯婷, 何智鹏, 等. 基于半直接耦合的压接型IGBT模块功率循环仿真方法[J]. 南方电网技术, 2021, 15(06):1-6.

[34]Pietruszka A, Gorecki P, Wronski S, et al. The Influence of Soldering Profile on the Thermal Parameters of Insulated Gate Bipolar Transistors (IGBTs)[J]. Applied Sciences, 2021, 11:5583.

[35]Lin D, Gao D, Fang Z, et al. Design and Realization of Intelligent Operation and Smart Grid Maintenance System for Data Center in Electric Power Supply[J]. Journal of Physics: Conference Series, 2021, 2033(1):012057.

[36]李亚萍, 周雒维, 孙鹏菊, 等. 基于模块跨导的IGBT键合线健康状态准在线监测方法[J]. 中国电机工程学报, 2018, 038(023):7035-7044.

[37]刘宾礼, 肖飞, 罗毅飞, 等. 基于集电极漏电流的IGBT健康状态监测方法研究[J]. 电工技术学报, 2017, 32(16):183-193.

[38]王贺. IGBT功率模块结温探测和寿命预测[D]. 河北工业大学, 2015.

[39]周利华. IGBT的寿命评估方法研究[D]. 安徽理工大学, 2017.

[40]Baharani M, Biglarbegian M, Parkhideh B, et al. Real-time deep learning at the edge for scalable reliability modeling of Si-MOSFET power electronics converters[J]. IEEE Internet of Things Journal, 2019, 6(5): 7375-7385.

[41]史业照, 郭斌, 郑永军. 基于LSTM网络的IGBT寿命预测研究[J]. 中国测试:1-6.

[42]陈冰, 鲁刚, 房红征, 等. 基于退化数据和DBN算法的IGBT健康参数预测方法[J]. 计算机测量与控制, 2017, 25(5): 71-75.

[43]任宏宇, 余瑶怡, 杜雄, 刘俊良, 周君洁. 基于优化长短期记忆神经网络的IGBT寿命预测模型[J]. 电工技术学报:1-13.

[44]郭子庆, 王学华. 基于神经网络的IGBT模块剩余使用寿命预测模型[J]. 电测与仪表, 2023, 60(01):132-138.

[45]葛建文, 黄亦翔, 陶智宇等. 基于Transformer模型的IGBT剩余寿命预测[J]. 半导体技术, 2021, 46(04):316-323.

[46]Ren L, Zhao L, Hong S, et al. Remaining useful life prediction for lithium-ionbattery: a deep learning approach [J]. IEEE Access, 2018, 6.

[47]Peng Y, Hou Y, Song Y, et al. Lithium-ion battery prognostics with hybrid gaussian process function regression[J]. Energies, 2018, 11.

[48]于春风，万明明，李伟林. 航空直流接触器电寿命预测研究[J]. 测控技术,2020,39(12):69-73+91.

[49]唐赓， 戴宏跃，郭志强，余远峰，张宏煕.基于UPS关键元器件的智能监控技术研究与样机测试验证[J]. 机电工程技术, 2023, 52(08):105-109.

[50]Haque M S, Choi S, Baek J. Auxiliary particle filtering-based estimation of remaining useful life of IGBT[J]. IEEE Transactions on Industrial Electronics,2017,65(3): 2693-2703.

[51]Yu J, Mo B, Tang D, et al. Remaining useful life prediction for lithium-ion batteries using a quantum particle swarm optimization-based particle filter[J]. Quality Engineering, 2017, 29(3): 536-546.

[52]秦琪, 赵帅, 陈绍炜, 等. 基于粒子群优化粒子滤波的电容剩余寿命预测[J].计算机工程与应用, 2018, 54(20):237-241+258.

[53]Duan B, Zhang Q, Geng F, et al. Remaining useful life prediction of lithium‐ion battery based on extended Kalman particle filter[J]. International Journal of Energy Research, 2020, 44(3): 1724-1734.

[54]黄琼，丁兆云. 基于粒子滤波的隧道火灾烟气速度估计方法[J]. 计算机应用, 2023, 43(03):986-990.

[55]刘亚姣, 刘振泽, 宋晨辉. 基于改进粒子滤波的锂离子电池RUL预测[J]. 吉林大学学报(信息科学版), 2018, 36(02):173-177.

[56]韩雨薇， 郑安迪， 朱俞竹， 王璐瑶， 董明泽. 蒙特卡洛粒子滤波算法应用研究[J]. 科技资讯, 2020, 18(35):12-13+16.

[57]马静雯, 李树青, 夏梦瑶. 机器学习在GDP预测中的应用研究述评[J]. 科技情报研究, 2022, 4(03):73-94.

[58]邹东池, 白富生. 基于差分进化算法的神经网络优化设计[J]. 重庆师范大学学报(自然科学版), 2022, 39(01):79-89.

[59]魏花.基于卷积神经网络的细粒度图像识别关键技术分析与研究[D]. 中国科学院大学(中国科学院长春光学精密机械与物理研究所), 2021.

[60]李炳臻, 姜文志, 顾佼佼 ,等. 基于卷积神经网络的目标检测算法综述[J]. 计算机与数字工程, 2022, 50(05):1010-1017.

[61]岑岗， 蔡永平， 岑跃峰. 基于深度学习的永磁同步电机温度预测模型[J]. 浙江科技学院学报, 2022, 34(03):216-224.

[62]周飞燕, 金林鹏, 董军. 卷积神经网络研究综述[J]. 计算机学报, 2017, 40(06): 1229-1251.

[63]杜文辽，高军杰， 杨凌凯，巩晓赟，王宏超，纪莲清.多尺度加权CEEMD-1DCNN旋转机械故障诊断[J].机床与液压, 2023, 51(17):202-208.

[64]Das Sayan,Saha Sanjoy Kumar. Diabetic retinopathy detection and classification using CNN tuned by genetic algorithm[J]. Multimedia Tools and Applications, 2022, 81(6).

[65]殷婷婷, 杨忠, 孙瑞胜, 等. 改进的Sobel算法和形态学小波变换相融合的图像边缘检测方法[J]. 金陵科技学院学报, 2022, 38(02):15-21.

[66]Torres Cesar, Gonzalez Claudia I, Martinez Gabriela E. Fuzzy Edge-Detection as a Preprocessing Layer in Deep Neural Networks for Guitar Classification[J]. Sensors, 2022, 22(15).

[67]赵文博， 马紫彤， 杨哲. 基于超图神经网络的恶意流量分类模型[J]. 网络与信息安全学报, 2023, 9(05):166-177.

[68]尹小兰，何芯芯，杜林，李远盛，张俊辉. RNN、JPR及ARIMA 3种模型预测中国农村孕产妇死亡率的比较[J]. 中华疾病控制杂志, 2023, 27(11):1308-1313.

[69]Li Wenjing, Wang Xiaoxiao, Han Honggui, et al. A PLS-based pruning algorithm for simplified long–short term memory neural network in time series prediction[J]. Knowledge-Based Systems, 2022, 254.

[70]胡凯茜， 李欣， 裴炳森. 面向目标用户的深度学习模型可视化综述[J]. 计算机系统应用, 2023, 32(11):36-47.

[71]刘建伟,王园方,罗雄麟.深度记忆网络研究进展[J].计算机学报,2021, 44(08): 1549-1589.

[72]樊顺厚.广义Hadamard积[J].天津纺织工学院学报, 2000(04):6-7.

[73]文井辉, 伍荣森, 李帅永, 等. 基于DRSN和优化BiLSTM的轴承剩余寿命预测方法[J]. 计算机集成制造系统. 2022, 1-18.

[74]Xiao H, Liu C, and Zhang B. Reconstruction of central arterial pressure waveform based on CNN-BILSTM[J]. Biomedical Signal Processing and Control, 2022, 74:103513.

[75]廖涛，马文祥，张顺香. 基于多特征融合的双通道医疗实体识别[J].计算机工程与设计, 2023, 44(10):3146-3152.

[76]冯茜, 李擎 ,全威, 等. 多目标粒子群优化算法研究综述[J]. 工程科学学报, 2021, 43(06):745-753.

[77]李发陵,彭娟.基于大数据的复杂超参数优化组合方法仿真[J].计算机仿真,2022, 39(07):332-33.