Life Prediction for IGBT Based on Improved Long Short-Term Memory Network

In order to prevent the failure of power electronic devices caused by the aging failure of Insulated Gate Bipolar Transistor (IGBT), this paper proposes an improved Long Short-Term Memory (LSTM) network for IGBT life prediction. Firstly, the collector-emitter spike voltage in the IGBT aging dataset is analyzed and selected as the aging feature, and the aging feature is extracted and reconstructed based on the Successive Variational Mode Decomposition (SVMD) method. Secondly, the grid search method is used to find the optimization of hyperparameters such as optimizer, epochs, and batchsize of the LSTM network to improve the prediction accuracy. Finally, the prediction results of the LSTM model are input to a Gaussian naive Bayes classifier to convert the IGBT remaining life. By comparing the prediction result with the traditional LSTM model, the mean absolute error and root mean square error of this method are reduced by 54.8% and 52.0%, which can predict the remaining life of IGBT more accurately.