RUL Predict of IGBT Based on DeepAR Using Transient Switch Features

Abstract

The insulated gate bipolar transformer (IGBT) has been widely used in power electrical system such as wind power converter and drive converter. However, the high load and capricious working environment make IGBT become the most vulnerable component in the electrical system. In this paper, a method based on probabilistic forecasting with auto-regressive recurrent networks (DeepAR) is proposed to predict the remaining useful life (RUL) of IGBT. Firstly, the transient data of the collector-emitter voltage signal are acquired when IGBT is turned off. Then, the different characteristics are extracted from transient data and the features that can represent the health state of IGBT best are chosen as the input of the DeepAR model to predict the remaining life of IGBT. Experiment results show that the log-log ratio of transient data can be an accurate precursor to predict RUL, and compared with other similar series predict models such as Auto-Regressive Integrated Moving Average (ARIMA) and Simple Exponential Smoothing (SES), DeepAR can get higher accuracy.