Research on IGBT life prediction based on LSTM network
To solve the problem of fatigue failure caused by cyclic impact of thermal stress and electrical stress during IGBT operation,a life prediction method based on long and short term memory(LSTM)network is proposed.Using NASA's forecast center provides the accelerated aging of data sets,analysis and selection of collector to emitter transient peak voltage as failure characteristic parameters of LSTM network built by Matlab,Adam optimization algorithm is used to train network,order to predict failure characteristic parameter data,and selected three performance evaluation indicators and ARIMA model and ELMAN neural network model of prediction were analyzed.The results show that the RMS error of LSTM network model is 0.0476,the average absolute error is 0.0322,and the average absolute percentage error is 0.4917%.The prediction accuracy of LSTM network model is higher,which can better realize the life prediction of IGBT,and has certain reference value for the life prediction of other power electronic devices.
NETL
NSTL
万方数据
